Human interface guidelines_v1.8.0

Human
Interface
Guidelines
v1.8

Kinect for Windows | Human Interface Guidelines v1.8 2
©2013 Microsoft Corporation. All rights reserved.
Acknowledgement and waiver
You acknowledge that this document is provided “as-is,” and that
you bear the risk of using it. (Information and views expressed in this
document, including URL and other Internet Web site references, may
change without notice.) This document does not provide you with
any legal rights to any intellectual property in any Microsoft product.
The information in this document is also not intended to constitute
legal or other professional advice or to be used as a substitute for
specific advice from a licensed professional. You should not act
(or refrain from acting) based on information in this document
without obtaining professional advice about your particular facts
and circumstances. You may copy and use this document for your
internal, reference purposes. In using the Kinect software and Kinect
sensors described in this document, you assume all risk that your use
of the Kinect sensors and/or the software causes any harm or loss,
including to the end users of your Kinect for Windows applications,
and you agree to waive all claims against Microsoft and its affiliates
related to such use (including but not limited to any claim that
a Kinect sensor or the Kinect software is defective) and to hold
Microsoft and its affiliates harmless from such claims.
Microsoft, Kinect, Windows, and Xbox are registered trademarks, and
Xbox360 is a trademark, of the Microsoft group of companies. All
other trademarks are property of their respective owners.

Contents Introduction
Learn about the Kinect for Windows sensor and SDK, and how
placement and environment affect its use
Interaction Design Tenets for Kinect for Windows
Learn about our design principles and how to design the best
input for the task
Gesture
Learn interaction design considerations for various types of gesture
Voice
Learn interaction design considerations for voice, including language
and audio considerations
Feedback
Learn about interaction design considerations for gestural and audio
feedback, text prompts, and more
Basic Interactions
Learn about the areas of the Kinect screen and user space, plus
details on engaging, targeting, selecting, panning and scrolling,
zooming, and entering text
Additional Interactions
Learn about distance-dependent interactions, and designing
for multiple inputs and users
Conclusion
4
14
21
46
60
77
118
135

Introduction
Welcome to the world of Microsoft Kinect for
Windows–enabled applications. This document
is your roadmap to building exciting human-
computer interaction solutions you once
thought were impossible.
We want to help make your experience with
Microsoft Kinect for Windows, and your users’
experiences, the best. So, we’re going to set
you off on a path toward success by sharing
our most effective design tips, based on our
long-term design, development, and usability
work. You’ll be able to focus on all those unique
challenges you want to tackle.
Keep this guide at hand – because, as we
regularly update it to reflect both our ongoing
findings and the evolving capabilities of Kinect
for Windows, you’ll stay on the cutting edge.
Before we get into design tips, it will help
to start with a couple of basics: a quick
introduction to the Kinect for Windows sensor,
software development kit (SDK), and Developer
Toolkit, and some environment and sensor
setup considerations.
Introduction

Meet the Kinect
for Windows
Sensor and SDK
The Kinect for Windows sensor and SDK
provide the ears and eyes of your application.
You’ll want to keep their capabilities in mind as
you design your application.
Introduction > Meet the Kinect for Windows Sensor and SDK

How Kinect for
Windows sensor,
SDK, and Toolkit work
together
The Kinect for Windows sensor, SDK, and
Toolkit work as a team to provide new and
exciting capabilities to your application.
Kinect for Windows sensor
Provides raw color image frames from the RGB camera, depth image frames from the depth
camera, and audio data from the microphone array to the SDK.
Kinect for Windows SDK
Processes the raw data from the sensor to provide you with information such as skeleton
tracking for two people, player tracking for six people, and word recognition from audio data
for a given language. You can download the SDK free from www.KinectforWindows.com.
Kinect for Windows Developer Toolkit
Using the Toolkit is optional, but you can find code samples in it that show how to use features
of the SDK, as well as components such as Kinect Button and Kinect Cursor, which help you build
interfaces faster. Using the components in the Toolkit lets you focus on your unique problem, and
makes the user experience of your application consistent with other Kinect for Windows–enabled
applications. Throughout this document, whenever we reference a sample or control, it is in the
Developer Toolkit, also downloadable at www.KinectforWindows.com.
The following sections cover what this team of products does to bring natural experiences to your
application. When we say “Kinect for Windows,” we mean the “sensor” and SDK working together,
unless otherwise specified.

• Physical limits: 0.4m to 3m
• Sweet spot: 0.8m to 2.5m
What Kinect for
Windows sees
Kinect for Windows is versatile, and can see
people holistically, not just smaller hand
gestures. Six people can be tracked, including
two whole skeletons. The sensor has an RGB
(red-green-blue) camera for color video, and
an infrared emitter and camera that measure
depth. The measurements for depth are
returned in millimeters.
The Kinect for Windows sensor enables a wide
variety of interactions, but any sensor has
“sweet spots” and limitations. With this in mind,
we defined its focus and limits as follows:
Physical limits – The actual capabilities of the
sensor and what it can see.
Sweet spots – Areas where people experience
optimal interactions, given that they’ll often
have a large range of movement and need to
be tracked with their arms or legs extended.
27°
43.5°
27°
57.5°
0.4m/1.3ft
sweet spot
physical limits
3m/9.8ft
0.8m/2.6ft
2.5m/8.2ft
sweet spot
physical limits0.8m/2.6ft
4m/13.1ft
1.2m/4ft
3.5m/11.5ft
Near mode depth
ranges
• Physical limits: 0.8m to 4m
(default)
Extended depth (beyond
4m) can also be retrieved but
skeleton and player tracking
get noisier the further you
get away, and therefore may
be unreliable.
• Sweet spot: 1.2m to 3.5m
Default mode depth
ranges
• Horizontal: 57.5 degrees
• Vertical: 43.5 degrees, with
-27 to +27 degree tilt
range up and down
Angle of vision (depth
and RGB)
Note that Near mode is an actual
setting for Kinect for Windows, and is
different from the various ranges we
detail in Interaction Ranges, later in
this document.

Skeleton tracking
Full skeleton mode
Seated mode
Kinect for Windows can track
up to six people within its
view, including two whole
skeletons.
Kinect for Windows can
track skeletons in default full
skeleton mode with 20 joints.
Kinect for Windows can also
track seated skeletons with
only the upper 10 joints.

Audio input
Microphone array
Sound threshold
The Kinect for Windows
sensor detects audio input
from + and – 50 degrees in
front of the sensor.
The microphone array can be
pointed at 10-degree increments
within the 100-degree range.
This can be used to be
specific about the direction
of important sounds, such as
a person speaking, but it will
not completely remove other
ambient noise.
The microphone array can
cancel 20dB (decibels) of
ambient noise, which improves
audio fidelity. That’s about
the sound level of a whisper.
(Kinect for Windows supports
monophonic sound cancellation,
but not stereophonic.)
Sound coming from behind the
sensor gets an additional 6dB
suppression based on the design
of the microphone housing.
What Kinect for
Windows hears
Kinect for Windows is unique because its single
sensor captures both voice and gesture, from
face tracking and small movements to whole-
body. The sensor has four microphones that
enable your application to respond to verbal
input, in addition to responding to movement.
For more information about audio and
voice issues, see Voice, later in this
document.
20 decibels
cancelled

Directional microphone
Loudest source targeting
You can also programmatically
direct the microphone array
– for example, toward a
set location, or following a
skeleton as it’s tracked.
By default, Kinect for
Windows tracks the loudest
audio input.
track me!

Introduction > Consider Sensor Placement and Environment
Consider Sensor
Placement and
Environment
The situation in which your Kinect for
Windows–enabled application is used can
affect how users perceive its reliability and
usability. Don’t forget to test your application
early (and often) with the kind of environment
and setup you expect it to ultimately be used
in. Consider the following factors when you
design your application.
For more information, see Choose the
Right Environment for Voice, and Best
Setup for Controls and Interactions,
later in this document.

Will many people be moving around the sensor?
You’ll want an optimal setup and space where no one comes between the engaged user and the
sensor. For example, you might want an interaction area defined by using a sticker on the floor
to indicate where the user should stand, or by roping off an area so that people walk around.
Will you rely on voice as an input method?
Voice input requires a quiet environment for reliable results. If you can’t control ambient noise level,
try designing for user interaction closer to the sensor. If a noisy environment is unavoidable,
voice might work better as augmenting other inputs (but not as the sole input method).
How far back are your users?
Consider what they can comfortably see. Design your user interface to account for the distance
at which people will interact, within the range of the sensor. Users who are older or visually
impaired might need graphics or fonts to be larger.
What will the lighting situation be?
Lighting affects image quality in different ways. So, if you envision a particular design, you’ll
want to specify that users check their lighting; and if you expect a certain kind of lighting, you’ll
need to design around those limitations. For example, ideally the light will come from behind
the sensor. The infrared depth camera works in all lighting situations (even darkness), but better
in moderate light than in direct sunlight or full-spectrum lighting. Dim lighting is fine for depth-
sensing applications (such as using avatars); with large amounts of natural light, skeleton tracking
is less reliable. On the other hand, color image quality does require good lighting (for example, if
you’ll use green-screening).

How will your users dress?
Items that drastically change the shape of a person wearing or holding them might confuse
skeleton tracking. For example, black clothing, as well as reflective items, can interfere with the
infrared camera and make skeleton tracking less reliable.
Where will the sensor be placed?
Because your users will be interacting with the screen, place the sensor above or below the
screen, to directly face the subjects it’s expected to track. Avoid extreme tilt angles. If you’re
using the Kinect for Windows controls and interactions we provide, the distance between the
user and the sensor that will provide the most reliable behavior is between 1.5 and 2 meters.
If your scenario involves simple motion or blob detection and doesn’t rely on skeleton tracking,
you can mount the sensor to the ceiling or at other angles.

Interaction Design
Tenets for
Kinect for Windows
Kinect for Windows opens up a new world of
gesture design and voice design.
14
Interaction Design Tenets for Kinect for Windows
Kinect for Windows | Human Interface Guidelines v1.8

Overall Design
Principles
Before we go into to our design guidelines,
we recommend you first read this brief
section for some very important interaction
tenets. They reflect the best ways we’ve
found to employ gesture and voice, and to
avoid some thorny issues.
Keep the following in mind as you design
your interactions.
Interaction Design Tenets for Kinect for Windows > Overall Design Principles

The best user experiences are
context-aware.
• Your UI should adapt as the distance
between the user and the sensor changes.
• Your UI should respond to the number and
engagement of users.
• Place your controls based on expected
user movements or actions.
• Make sure your interactions are appropriate
for the environment in which your
application will be used.
• The further the user, the wider the range
of movement.
• The closer the user, the more and finer the
content, tasks, and gestures.
• Environment affects user input.
Each input method is best at
something and worst at something.
• Users choose the input that requires the
least overall effort for a given scenario.
• People tend to stick to a single input unless
they have a reason to change.
• Input methods should be reliable,
consistent, and convenient – otherwise
people will look for alternative options.
• Switching input methods should happen
intuitively, or at natural transition points in
the scenario.
The strongest designs come after
user testing.
• Kinect for Windows enables a lot of new
interactions, but also brings new challenges.
• It’s especially hard to guess ahead of time
what will work and what won’t.
• Sometimes minor adjustments can make a
huge difference.
• Conduct user tests often and early, and allow
time in your schedule for adjustments to
your design.
Confident users are happy users.
• It’s important to keep interactions simple,
and easy to learn and master.
• Avoid misinterpreting user intent.
• Give constant feedback so people
always know what’s happening and
what to expect.
Interaction Design Tenets for Kinect for Windows > Overall Design Principles

Strong Inputs
In order to provide a good experience and
not frustrate users, a strong voice and gesture
interaction design should fulfill a number of
requirements.
To start with, it should be natural, with an
appropriate and smooth learning curve for
users. A slightly higher learning curve, with
richer functionality, may be appropriate for
expert users who will use the application
frequently (for example, in an office setting
for daily tasks).
Interaction Design Tenets for Kinect for Windows > Strong Inputs
For more information about designing
for different distance ranges, see
Distance-Dependent Interactions,

A strong voice and gesture interaction design should be:
• Considerate of user expectations from their use of other common input
mechanisms (touch, keyboard, mouse).
• Ergonomically comfortable.
• Low in interactional cost for infrequent or large numbers of users
(for example, a kiosk in a public place).
• Integrated, easily understandable, user education for any new interaction.
• Precise, reliable, and fast.
• Considerate of sociological factors. People should feel comfortable
using the input in their environment.

Intuitive, with easy “mental mapping.”
Efficient at a variety of distance ranges.
Easy to back out of if mistakenly started,
rather than users having to complete the
action before undoing or canceling.
Appropriate for the amount and type of
content displayed.

The Right Input
Mode for Your Task
As you design your interactions, keep in mind
all the input methods available to you, and the
pros and cons each have to offer.
Enable users to accomplish tasks
quickly and intuitively.
Use each input mode for what it’s
naturally best at.
Take user orientation and location
into account so that input modes are
switched only when it’s convenient and
benefits productivity.
Force users to use inappropriate input
methods for certain tasks just because
they’re good for other tasks in the scenario.
Require an input method that feels
forced, unnatural, awkward, or tedious.
Do Don’t
Switch input modes for the sake of
variety or effect.
Natural actions
If one interaction is better at
a given task, consider using
that one instead of forcing
all interactions into one
input mode.
For example, for entering text, let people use their physical keyboard
or a touch interface instead of gesturing.

Gesture
This section begins by providing some
important gesture definitions and goes on to
recommend a number of important design
considerations for gesture interaction.
Gesture
21Kinect for Windows | Human Interface Guidelines v1.8

Basics
In this document we use the term gesture
broadly to mean any form of movement that
can be used as an input or interaction to
control or influence an application. Gestures
can take many forms, from simply using your
hand to target something on the screen, to
specific, learned patterns of movement, to
long stretches of continuous movement using
the whole body.
Gesture is an exciting input method to
explore, but it also presents some intriguing
challenges. Following are a few examples of
commonly used gesture types.
Gesture > Basics

Innate gestures
Learned gestures
Gestures that the user
intuitively knows or that
make sense, based on the
person’s understanding of
the world, including any skills
or training they might have.
Examples:
• Pointing to aim
• Grabbing to pick up
• Pushing to select
Gestures you must teach the
user before they can interact
with Kinect for Windows.
Examples:
• Waving to engage
• Making a specific pose to
cancel an action
Innate and learned
gestures
You can design for innate gestures that people
might be familiar with, as well as ones they’ll
need to learn and memorize.
Gesture > Basics
Wave to engage

Static gesture
Continuous gesture
A pose or posture that the
user must match and that
the application recognizes as
meaningful.
Prolonged tracking of
movement where no specific
pose is recognized but the
movement is used to interact
with the application.
Static, dynamic, and
continuous gestures
Whether users know a given gesture by
heart or not, the gestures you design
for your Kinect for Windows application
can range from a single pose to a more
prolonged motion.
Gesture > Basics
Dynamic gesture
A defined movement that
allows the user to directly
manipulate an object
or control and receive
continuous feedback.
Slide to confirm Confirmed!
Pressing to select and gripping to move are examples of
dynamic gestures.
Examples include enabling a user to pick up a virtual box or perform
a whole-body movement.
Be wary of designing for symbolic static gestures such as “okay”
that might carry different meanings across cultures.

Gesture
Interaction
Design
With Kinect for Windows, you can explore
the new and innovative field of gesture
interaction design. Here are some of our
key findings and considerations in making
gesture designs feel “magical.”
Gesture > Gesture Interaction Design

Users should agree with these statements as they use gesture
in your application:
• I quickly learned all the basic gestures.
• Now that I learned a gesture, I can quickly and accurately perform it.
• When I gesture, I’m ergonomically comfortable.
• When I gesture, the application is responsive and provides both immediate
and ongoing feedback.
Accomplish gesture
goals
The users’ goal is to accomplish their tasks
efficiently, easily, and naturally. Your goal is to
enable them to fulfill theirs.

Design for reliability
Reliability should be a top priority.
Without reliability, your application will
feel unresponsive and difficult to use, and
frustrate your users. Try to strike a good
reliability balance.
Teach users how to effectively perform
a gesture.
Instill confidence so users can show
others how to perform a gesture.
Teach users how to perform a gesture
early in the experience so they can
use it in similar contexts.
Consider the frequency and cost of
false activations.
Drive users to other modes of input/
interaction because of poor reliability.
Require such rigid gestures that users
develop superstitious behaviors, like
making up incorrect justifications for
reactions they don’t understand.
Do Don’t
Use different gestures for similar actions
unless there is too much overlap
between the gestures and natural
variation in users’ movement.
Design multiple gestures that are
too similar.
• If the gesture is too circumscribed, unique, or complex, there will be fewer “false
positives,” but it might be hard to perform.
• If the gesture is too unspecific or simple, it will be easier to perform, but might
have lots of false positives and/or conflicts with other gestures.
For more information about false
positives, see Engagement, later in this
document.

Design for
appropriate user
mindset
If you’re designing a non-gaming Kinect for
Windows–enabled application, or a specific UI
item such as a menu, keep in mind that game
mindset is NOT the same as UI mindset.
As you design, keep conscious of the purpose
of your application and the nature of your
users, and design accordingly.
Game mindset
Challenge is fun! If a user is
in game mindset and can’t
perform a gesture, then it’s
a challenge to master it and
do better next time.
UI mindset
Challenge is frustrating. If
a user is in UI mindset and
can’t perform a gesture, he
or she will be frustrated and
have low tolerance for any
learning curve.
In game mindset, a silly gesture can be fun or entertaining.
In UI mindset, a silly gesture is awkward or unprofessional.

Design for natural
interactions
Gesture might provide a cool new method of
interacting with your application, but keep in
mind that its use should be purposeful.
Allow people to interact from a distance.
Allow gesture to enable an
interaction or expression that other
input devices can’t.
Center part of the interaction on
user tracking and poses – things that
Kinect for Windows does best.
Try to force-fit gesture or voice on
existing UI that was designed for a
different input method.
For example, don’t simply
map gestures to a touch
interface.
Require gestures for tasks that could be
done faster and more easily by using
another input method.
Require that gestures be used for
productivity, speed, and precision tasks.
Do Don’t
Provide multiple ways to perform a task
if environment or context might change.

Determine user intent
and engagement
Determining user intent is a key issue, and
hard to do right. Unlike other input devices,
which require explicit contact from a user, or
only track a tiny area of a person’s body, Kinect
for Windows sees a person holistically. Kinect
for Windows users are constantly moving and
interacting with the world, whether or not they
intend to be interacting with your application.
Your challenge is to detect intended gestures
correctly and avoid detecting “false positive”
gestures. Keep in mind that users in a social
context may alternate between interacting with
the application and with other people beside
them or in the vicinity. The application must
be aware and considerate of the social context
of the users.
Hello!
Provide feedback to increase confidence.
Provide a clear and simple way for
people to engage and disengage.
Recognize unique and intentional
movement as gesture or interaction.
Ignore natural body movements
(scratching, sneezing, etc.).
Forget to provide feedback when a
gesture fails or is canceled.
Require users to modify natural
behavior to avoid interaction.
Miss, ignore, or don’t recognize the
user’s gesture.
Misinterpret natural body movements
as the proposed gesture.
Do Don’t
For more information, see the
Engagement and Feedback sections,
Part of determining user intent is recognizing when people want to first engage with the
application. It’s tricky because there’s no physical medium such as a mouse for detecting intent
with Kinect for Windows.
If the user must actively trigger engagement, be sure you demonstrate it at some point early in
the experience, or make the trigger clearly visible when it’s needed. Avoid long or cumbersome
engagement experiences, because they will become frustrating for returning users or users who
have already observed someone using the application.

Waving is one method for determining user intent. (For example, Xbox users must
wave back and forth at least three times, rotating at the elbow, for the movement to be
recognized.) Waving feels natural, yet it’s a unique movement not often misinterpreted.
It also carries meaning to people as a way to begin an interaction.
Other possible ways to recognize engagement involve how close the user is to the sensor,
or a particular pose, such as what direction he or she is facing.
You can also choose to recognize when someone disengages as a cue to explicitly turn
interactability off, rather than using a timeout or other passive method. For example, when
a user turns away from the screen, take that as a cue that they’re no longer engaged. (But
be careful not to disengage people accidentally in the middle of a task.)
Notes

Design for variability
of input
Users’ previous experience and expectations
affect how they interact with your application.
Keep in mind that one person might not perform
a gesture the same way as someone else.
Gesture interpretation
Simply “asking users to wave”
doesn’t guarantee the same
motion.
They might wave:
• From their wrist
• From their elbow
• With their whole arm
• With an open hand
moving from left to right
• By moving their fingers
up and down together
It’s hard to scan for a broad variety of interpretations; instead, it’s
best to give users clear tips about the exact gesture you require.

Make the gesture fit
the user’s task
Logical gestures have meaning and they relate
to associated UI tasks or actions. The feedback
should relate to the user’s physical movement.
Make the action and feedback parallel –
for example, users swipe left to scroll left
or move content to the left.
Require users to move a hand up to
scroll content to the left.
Do Don’t
Use logical movements that are easy
to learn and remember.
Require abstract body movements that
have no relationship to the task and are
hard to learn and remember.
Make the size or scope of the motion
match the significance of the feedback
or action.
Require a big movement for a small
result, like a whole arm swipe to move
one item in a list.
Use big, easily distinguishable
movements for important and less
frequent actions.
Use big movements for actions that
must be repeated many times through
an experience.

Design for complete
gesture sets
The more gestures your application requires,
the harder it is to design a strong gesture
set. So, we recommend that you keep the
number of gestures small, both so that they’re
easy to learn and remember, and that they’re
distinct enough from one another to avoid
gesture collisions. In addition, if you strive to
be consistent with other applications, people
will feel at home with the gestures and you’ll
reduce the number of gestures they have to
learn. You’ll also reduce the training you have
to provide.
Here are a few things to remember when defining a gesture set:
• Make sure each gesture in an application’s gesture set feels related and cohesive.
• Keep your users’ cognitive load low; don’t overload them with gestures to remember.
Research shows that people can remember a maximum of six gestures.
• Take cues from existing gestures that have been established in other Kinect applications.
• Test thoroughly for “false positive” triggers between gestures.
• Use obvious differentiators, such as direction and hand state, to make gestures
significantly different, reduce false positives, and avoid overlap.
• Make sure similar tasks use the same gesture language, so that users can guess or
discover gestures through logical inference or pairing. For example, pair a swipe right
(to move content right) to a swipe left (to move content left).
IF: THEN:

Differentiate direction changes.
Differentiate progression or path.
Risk gesture overlap by making the
direction of two similar gestures
the same.
Have two gestures that follow the same
path, especially if they’re in the same
direction.
Have vague and similar start and end
points that result in different actions.
Do Don’t
Have clear and different start and
end points.

Think about the whole scenario. What does the user do after completing a gesture?
Might that action look like the start of an unintended gesture? Will it put them in a
natural position to begin the next logical gesture for your common scenarios?
Notes
Do Don’t
Define the conceptual opposite action
of a defined gesture as something other
than a pair or mirror of that gesture.
Design logically paired gestures to
be consistent and follow the same
gesture language.

Handle repeating
gestures gracefully
If users will need to perform a gesture
repeatedly (for example, moving through
pages of content), you might encounter a
few common challenges.
Design repeating gestures to be fluid,
without jarring starts and stops.
Enable users to quickly get into a
rhythm of movement.
Let feedback get in the way and hinder
or pace the user’s rhythm. For example,
people shouldn’t feel like they must
wait for an animation to finish before
they can repeat the gesture.
Design a gesture such that repetition is
inefficient.
Do Don’t
Consider the movement of the whole
repetition. Ignore the “return” portion
of a repeated gesture if it will disrupt
the ability of the user to repeat the
gesture smoothly.
Design the opposite gesture to resemble
the “return” portion of the first gesture.

Avoid “handed”
gestures
Handed gestures are ones that can be
done only with a specific hand. They
do not provide users with a truly natural
experience, are undiscoverable, and should
be avoided.
Allow users to switch hands to reduce
fatigue.
Accommodate both left and right-
handed people.
Design as if you, the developer or
designer, are the only user.
Require specifically handed gestures;
they’re not discoverable or accessible.
Do Don’t

Vary one-handed and
two-handed gestures
For variety and to accommodate natural
interaction, you can design for both single-
handed and two-handed gestures.
Use one-handed gestures for all
critical-path tasks. They’re efficient
and accessible, and easier than two-
handed gestures to discover, learn,
and remember.
Use two-handed gestures for noncritical
tasks (for example, zooming) or for
advanced users. Two-handed gestures
should be symmetrical because they’re
then easier to perform and remember.
Require the user to switch between
one- and two-handed gestures
indiscriminately.
Use two-handed gestures for critical,
frequent tasks.
Do Don’t

Remember that
fatigue undermines
gesture
Your user shouldn’t get tired because of
gesturing. Fatigue increases messiness, which
leads to poor performance and frustration,
and ultimately a bad user experience.
Offer gesture alternatives: for example,
in a list of items, use different gestures
to navigate in small or big steps, jump
within the list, zoom in or out, or filter
to shorten the list.
Require excessive repetition of a single
gesture, such as page swiping to get
through a long list. (So, avoid long lists
or use a more appropriate modality, such
as voice search.)
Require users to assume uncomfortable
positions, such as holding their hand
above their head for a long time.
Do Don’t
natural movements awkward movements
good gestures in between
(unique and purposeful)
For more information, see Multimodal
Interactions, later in this document.

Consider user
posture and
movement ranges
User posture might affect your gesture design,
so consider where and how your users will use
your application. For example, sitting limits a
person’s possible movements.
Design a single gesture that works well
across all viable postures.
Be aware of the loss of reliability if
you view a user from the side. Joints
that are not in full view are placed in
predicted locations by the skeleton
tracking and aren’t always in the
correct position.
Require users to stand at distances or
orientations where skeleton tracking
will be less reliable. Examples include
users oriented sideways to the sensor,
blocked by objects (such as a table),
out of visible range of the sensor, etc.
Keep in mind the normal and
comfortable ranges of interaction for
the human body.
Design alternate gestures for the same
command, such as different gestures
for seated and full skeleton mode.
Do Don’t
Design one gesture that works well in
one posture but poorly in another.

Teach gestures
and enable
discoverability
Whenever a gesture is available to interact
with your application, you should find a way
to communicate it to your user. Here are
some options.
Teaching methods
• A quick tutorial for
new users
• An indication of gesture
state throughout the
interaction
• A visual cue or hint when
a user first engages
• A static image
• An animation
• A message or notification
How to
gesture
Hello!
Wave to engage

Be aware of
technical barriers
If you’re using skeleton data to define your
gestures, you’ll have greater flexibility, but
some limitations as well.
Tracking movement
Keeping arms and hands to
the side of the body when
performing gestures makes
them easier to track, whereas
hand movements in front of
the body can be unreliable.
Field of view
Make sure the sensor tilt and
location, and your gesture
design, avoid situations where
the sensor can’t see parts of a
gesture, such as users extending
a hand above their head.
Tracking reliability
Skeleton tracking is most
stable when the user faces
the sensor.
Tracking speed
For very fast gestures,
consider skeleton tracking
speed and frames-per-second
limitations. The fastest that
Kinect for Windows can track
is at 30 frames per second.

Remember your
audience
Regardless of how you define your gestures,
keep your target audience in mind so that
the gestures work for the height ranges
and physical and cognitive abilities of your
users. Think about the whole distance
range that your users can be in, angles that
people might pose at, and various height
ranges that you want to support. Conduct
frequent usability tests and be sure to test
across the full range of intended user types.
Physical differences
For example, you should
account for users of various
heights and limb lengths.
Young people also make,
for example, very different
movements than adults when
performing the same action,
due to differences in their
dexterity and control.

Iterate
Finally, getting a gesture to feel just right
might take many tweaks and iterations.
Create parameters for anything you can,
and (we can’t say it enough) conduct
frequent usability tests.
Design a gesture that works reliably
for your whole range of users.
Design a natural and discoverable
gesture.
Design a gesture that works for you
but no one else.
Build your application to be inflexible,
so it is hard to make adjustments.
Do Don’t

Voice
Besides gesture, voice is another input
method that enables new and natural-feeling
experiences.
Voice

Basics
Using voice in your Kinect for Windows–
enabled application allows you to choose
specific words or phrases to listen for and
use as triggers. Words or phrases spoken as
commands aren’t conversational and might not
seem like a natural way to interact, but when
voice input is designed and integrated well, it
can make experiences feel fast and increase
your confidence in the user’s intent.
Voice > Basics

About confidence
levels
When you use Kinect for Windows voice-
recognition APIs to listen for specific words,
confidence values are returned for each
word while your application is listening. You
can tune the confidence level at which you
will accept that the sound matches one of
your defined commands.
Voice > Basics
• Try to strike a balance between reducing false positive recognitions
and making it difficult for users to say the command clearly enough to
be recognized.
• Match the confidence level to the severity of the command. For
example, “Purchase now” should probably require higher confidence
than “Previous” or “Next.”
• It is really important to try this out in the environment where your
application will be running, to make sure it works as expected.
Seemingly small changes in ambient noise can make a big difference
in reliability.

Keyword/trigger
The sensor only listens for
a single keyword. When it
hears that word, it listens
for additional specified
words or phrases. This is
the best way to reduce false
activations. The keyword
you choose should be very
distinct so that it isn’t easily
misinterpreted. For example,
on Xbox360, “Xbox” is the
keyword. Not many words
sound like “Xbox,” so it’s a
well-chosen keyword.
Listening models
There are two main listening models for
using voice with Kinect for Windows: using
a keyword or trigger, and “active listening.”
Always on,
active listening
The sensor is always listening
for all of your defined words
or phrases. This works fine
if you have a very small
number of distinct words or
phrases – but the more you
have, the more likely it is that
you’ll have false activations.
This also depends on how
much you expect the user
to be speaking while the
application is running, which
will most likely depend on
the specific environment
and scenario.
Voice > Basics
Kinect!
Home
Home

Choose words and
phrases carefully
When choosing what words or phrases to
use, keep the following in mind. (If each
screen has its own set of phrases, these
guidelines apply within a set; if the whole
application uses one set of phrases, the
guidelines apply to the whole application.)
Distinct sounds
Avoid alliteration, words
that rhyme, common
syllable lengths, common
vowel sounds, and using
the same words in different
phrases.
Brevity
Keep phrases short (1-5
words).
Word length
Be wary of one-syllable
keywords, because they’re
more likely to overlap with
others.
Simple vocabulary
Use common words where
possible for a more natural
feeling experience and for
easier memorization.
Voice > Basics
SHOW ME THE ONE
ON THE LEFT
PUT THIS ONE IN THE
SHOPPING CART
SHOW THE LEFT ONE
PUT IN CART
Don’t
Alliteration
Same words in
different phrasesRhymes
Vowels in
Common
Syllables in
Common
CAT
KIT
KEY
MAKE
TAKE
Pair
Pear
SOMETHING
SOMEONE
INCREASE
DECREASE
MAIN
TAKE
DATE
FACE
ONE MORE TIME
SHOW ME MORE
PLAY MORE LIKE THIS
MORE RED
Do
PLAY
STOP
MORE
SCROLL
BACK
PLAY THIS ONE
STOP VIDEO
SHOW MORE SONGS
SCROLL RIGHT
GO BACK
Don’tDo
MAX OUT
CRIMSON
INITIAL
TURN UP
RED
FIRST
Don’tDo
Don’t

Minimal voice prompts
Keep the number of phrases
or words per screen small
(3-6).
Word alternatives
User prompts
If you have even more items
that need to be voice-
accessible, or for non-text
based content, consider
using numbers to map to
choices on a screen, as in
this example.
For commands recognized
with low confidence, help
course correct by providing
prompts – for example, “Did
you mean ‘camera’?”
Voice > Basics
Go home, Go back,
Next page, Previous page
1 2 3
camera 2
Did you mean “camera”?
camrwoj

Reduced false activation
Test and be flexible. If a
specific word always fails
or is falsely recognized, try
to think of a new way to
describe it.
Acoustics
Test your words and
phrases in an acoustic
environment similar to
where you intend your
application to be used.
Voice > Basics
Confidence level threshold
VS
Test Test
Triggering audio
Try using a trigger or event
to make Kinect for Windows
start listening. For example,
only listen when a skeleton is
detected in a certain area.
You can require higher confidence levels (80% to 95%) – that is, having
Kinect for Windows respond only when it’s certain that the user has given
the correct trigger. This might make it harder to users to interact, but
reduce unintentional actions.

Voice Interaction
Design
Generally, it’s best to avoid forcing people to
memorize phrases or discover them on their
own. Here are a few best practices for helping
users understand that they can use voice, and
learn what words or phrases are available.
Voice > Voice Interaction Design

Visual voice prompts
Display exactly how users
must say voice commands.
Listening mode
Indicate visually that the
application is “listening”
for commands. For
example, Xbox360 uses a
microphone icon.
User assistance
Display keywords onscreen,
or take users through a
beginning tutorial.
“Show me the next one”
“Show me the next 20”
First say “Kinect”,
then say “Home” or “Next”

Visual notifications
If there is an issue with the
microphone connection,
display an error icon and/
or message so the user can
address the problem.
Audio prompting
If for some reason the user
might not be facing the
screen, have an option for
Kinect for Windows to read
the available phrases out loud.
Alternative input
Voice shouldn’t be the only
method by which a user can
interact with the application.
Build in allowances for
the person to use another
input method in case voice
isn’t working or becomes
unreliable.
Microphone is not detecting
audio, please check connection!
if: then:
The quick brown
fox jumped over
the lazy dog

VUI (voice user interface) bars and labels
As seen in Xbox360 interfaces, VUI bars and labels are a great way to indicate what commands are
available, especially if you’re using a keyword. In Xbox360, the user says the keyword (“Xbox”), and
then the VUI bar appears, containing phrases that he or she can choose from. This is a good way of
clearly indicating what phrases are available, without having them always present on the screen.
See it, say it model
The “see it, say it” model is one where the available phrases are defined by the text on the screen.
This means that a user could potentially read any UI text and have it trigger a reaction. A variation
of this is to have a specified text differentiator, such as size, underline, or a symbol, that indicates
that the word can be used as a spoken command. If you do that, you should use iconography
or a tutorial in the beginning of the experience to inform the user that the option is available,
and teach them what it means. Either way, there should be a clear, visual separation between
actionable text on a screen and static text.
You can say:
“Home”, “Next”, “Back”
Kinect
When mic icon is blue,
you can say words in blue.
Lorem ipsum dolor sit amet,
consectetur adipisiciing elit,
sed do eiusmod tempor
incididunt ut labore et dolore
magna al qua.

Choose the right
environment
for voice
There are a few environmental
considerations that will have a significant
effect on whether or not you can
successfully use voice in your application.
Ambient noise
The sensor focuses on the loudest sound source and attempts to cancel out other ambient
noise (up to around 20dB). This means that if there’s other conversation in the room (usually
around 60-65dB), the accuracy of your speech recognition is reduced.
Amplify that to the sound level of a mall or cafeteria and you can imagine how much harder it
is to recognize even simple commands in such an environment. At some level, ambient noise is
unavoidable, but if your application will run in a loud environment, voice might not be the best
interaction choice. Ideally, you should only use voice if:
• The environment is quiet and relatively
closed off.
• There won’t be multiple people speaking
at once.
60-65dB =

System noises and cancellation
Although the sensor is capable of more complex noise cancellation if you want to build that
support, the built-in functionality only cancels out monophonic sounds, such as a system beep, but
not stereophonic. This means that even if you know that your application will be playing a specific
song, or that the song will be playing in the room, Kinect for Windows cannot cancel it out, but if
you’re using monophonic beeps to communicate something to your user, those can be cancelled.
Distance of users to the sensor
When users are extremely close to the sensor, the sound level of their voice is high. However,
as they move away, the level quickly drops off and becomes hard for the sensor to hear, which
could result in unreliable recognition or require users to speak significantly louder.
Ambient noise also plays a role in making it harder for the sensor to hear someone as they
get farther away. You might have to make adjustments to find a “sweet spot” for your given
environment and setup, where a voice of normal volume can be picked up reliably.
In an environment with low ambient noise and soft PC sounds, a user should be able to
comfortably speak at normal to low voice levels (49-55dB) at both near and far distances.
Distance of user to Kinect
Audioleveldetected

Social considerations
Keep in mind the social implications of your users needing to say commands loudly while using
your application. Make sure that the commands are appropriate for the environment, so you
don’t force your users to say things that will make them uncomfortable. Also make sure that
the volume at which they have to speak is appropriate for the environment. For example,
speaking loud commands in a cubicle-based office setup might be distracting and inappropriate.

Feedback
Whether you employ gesture, voice, or
both, providing good feedback is critical
to making users feel in control and helping
them understand what’s happening in the
application. This section covers some ways you
can make feedback as strong as possible.
Feedback

Basics
It’s important, especially if your users are
standing at a distance and have little direct
contact with the interface, to take extra care
in showing them how their actions map to
your application. Also, because a gesture is
only effective and reliable when users are
in the correct visibility range, it’s important
to give feedback to them in case they don’t
know when they’re out of range.
Feedback > Basics

Consider what users
want to know
As you design, imagine you’re a first-time user,
and ask yourself these questions.
Feedback > Basics
• What does the sensor see?
• Where’s the field of view?
• How much of me can the sensor see?
• Is my head in view?
For more information, see Engagement,
later in this document. • Is the sensor on and ready? • Am I in control?
• Can I engage now?

Feedback > Basics
Many of these questions can be answered by displaying a small User Viewer (visualizing
depth) on the screen to show what Kinect for Windows sees at any given time.
Highlighting players, hands, or other joints in the depth viewer might also be helpful.
You can also prompt people to move into the appropriate field of view whenever they’re
cropped, too close, or too far back.
Notes
• How many people can the sensor see?
• How do I know it’s seeing me and not
someone else?
• When and where can I gesture?

Feedback
Interaction
Design
This section gives tips for designing various
kinds of feedback, including selection states,
progress indicators, and other visuals, as well
as audio feedback.
Feedback > Feedback Interaction Design

Input suggestions
Use iconography or
tutorials to show users
what input methods are
available to them.
Gesture suggestions
Show what gestures are
available.
Best practices
There are several best practices that apply
whether you’re designing for gesture or voice.
Make it clear what content the user can take action on, and how
Differentiated controls
Use iconography, colors, or
tutorials to show users how
to differentiate between
controls they can activate,
text prompts for voice input,
and other text and content.
Lorem ipsum dolorLorem
Lorem
ipsum dolor
sit amet,
consectetur
adipiscing
button link normal text
Visual feedback
Show cursor visuals if you’re
tracking a user’s hand.
For more information, see Targeting
and Selecting, later in this document.
Press to select

Audio notifications
If there’s an error with a
Kinect for Windows state (for
example, a hardware or a
connection issue), show the
error message or state and a
suggestion for how to fix it.
Display some change in the
visuals. Consider whether
you should recommend that
the user switch to alternative
input methods.
Command suggestions
Show what commands users
can speak.
For an example of how to do this, see the Sensor Chooser component
in the Kinect for Windows Developer Toolkit.
You can say:
“Home”, “Next”, “Back”

Targeting and selection
Provide feedback about what
items users can take action
on, what state those items
are currently in, and how to
take action on them.
Progress feedback
When a user controls
something by direct
manipulation, show the
progress in a way that
translates to the person’s
motion.
Map feedback to gesture
Clarify selection states
For example, if the user turns a page by a gesture, show the page pulling
up and turning over as his or her hand moves horizontally.
For example, for targeting and selecting, with a gesture that requires
Z-axis movement, using size changes to show depth helps users
understand the action required (see Kinect Button, later in this
document).

Selection confirmation
Whether the user triggers
an item by voice or gesture,
have a visual indication of
recognition before changing
the state of the application.
UI controls
For controls that are not for
navigation, such as check
boxes and toggle buttons,
have clear visual changes
that show state changes.
resting hover selected
Lorem ipsum

Ergonomics
Be careful when designing
the timeout length for
progress indicators in case
users might be required to
hold a position. Also consider
how frequently users will be
repeating the interaction.
Avoid forcing users to wait
unnecessarily.
Clear visuals
If you’re using a progress
timer, or a countdown, use
clear visuals to show the
entire progression.
If you are visualizing the user’s
progress in completing a task,
make sure the visual is clear,
prominent, and placed where
the user is focused.
Use progress indicators
Hold pose for
4 seconds

User orientation
If the UI changes based on
something (such as distance)
that the user might have
triggered inadvertently, try
animation to show where
the content is located in the
new layout.
Layout continuity
If the user is navigating, use
animation to help him or her
understand the layout.
Keep context by animating
For example, horizontal animation can show that the user has moved
horizontally in space in your application’s layout.

Use skeleton tracking
feedback
Full-body skeleton tracking provides a wide
range of new application possibilities. You can
use feedback both to lead and confirm the
user’s movements.
The User Viewer
Another way to do this is to
show a small scene viewer
or visualization to show the
user exactly what Kinect
for Windows can see at any
given time. This can help
users understand how to stay
within the right range and,
when they’re outside of it,
why things aren’t working.
(For more information, see
The User Viewer, later in this
document.)
Make sure users know whether the sensor sees them
User resets
If you need to track a user
but the sensor cannot see
them, let the user know
where to stand so that it can.
If you lose track of a user in
the middle of a process, pause
the process and guide the
user back to a place where
you can track him or her. For example, in Kinect Sports, the game tells users if they’re
getting too close and shows them at the beginning where to stand
for optimal tracking.
For more user-tracking feedback
information, see Engagement, later in
this document.
Please move left

Training and feedback
If you want users to copy a
specific movement or action,
you can show an avatar or
animation, either before you
expect to track the users’
movement, or even during
the movement.
If you show an animation
during the user’s movement,
visually indicate whether
or not the user is doing it
correctly.
Lead by example
An example of this is in Xbox Dance Central, where the onscreen
dancer moves correctly and independently of the user, but limbs get
highlighted in red if the user makes a mistake.

Realistic reactions
Make other objects in the
scene react correctly to
collisions with the skeleton/
avatar.
Smooth movements
Apply smoothing techniques to
skeleton-tracking data to avoid
motion and poses that are jittery
or not human-like.
Real-time movements
Make sure the movement
is in real time so the person
feels in control.
Mirror users’ actions

Use audio feedback
Audio feedback can be very effective, and has
its own set of considerations.
Teaching cues
You can use sound patterns
or cues to communicate a
simple message or teach
users when to pay attention.
Warning sounds
Audio can be a good way
to get users’ attention if
they need to be notified of
something – for example,
audio warnings or alerts.

Activity signals
Sounds can be a good way
to signal to a user that
something has changed or
has been selected.
Click
Announce selection states
Consider using sounds that match the action the user is taking, to
enforce his or her feeling of control and familiarity – for example,
a click noise when a button is pressed.
Instructional audio
If the user isn’t facing
the screen, or is far away,
consider using audio as a
way to communicate to him
or her – for example, giving
available options, directions
or instructions, or alerts.
Base audio feedback on user orientation
You can play the
next song by
saying “next”

Combine feedback
In the physical world, we use all of our senses
to gauge whether our actions are effective
and in order to communicate naturally with
others. Similarly, combining different types of
feedback often makes for a better experience
in a virtual world.
Reinforced feedback
Combining visual and audio
inputs can make feedback
stronger.
A great example of this is the menu in Dance Central. As the user
scrolls up and down the menu, the following changes happen to the
item he or she points to:
• It grows larger.
• It glows a brighter shade of pink.
• It gets a slider visual.
• The items surrounding it become slightly
larger and a slightly different color.
• The sound changes dynamically, in real time,
based on the position of the user’s hand.
The result seems like a very “hands-on” experience, where the user
can almost feel the effects of his or her movements.

Basic Interactions
Although we leave it up to you to create
exciting and unique experiences with Kinect for
Windows, we’ve taken care of some of the basic
interactions and controls for you and included
them in the Developer Toolkit. Using our
interactions and controls saves you time and
also establishes some consistency that your users
will learn to expect as they encounter Kinect for
Windows experiences in various aspects of their
daily lives. We’ve spent a lot of time tuning and
testing this first set of interactions to give you
the best foundation possible, and we’ll add more
in our future releases.
The following sections help you set the stage for
your application, and show you how to enable
users to easily engage with it, target and select
items, and scroll or pan. We also call out some
examples that you’ll find in the Interaction
Gallery, and share some tips for interactions that
we don’t provide for you.
Basic Interactions

Basic Interactions > Best Setup for Controls and Interactions
Best Setup for
Controls and
Interactions
We’ve worked hard to provide great controls
and interactions, but to make them as reliable as
possible, we recommend this setup:
• Set your application to use Near mode. For
more information, see the SDK, at:
http://msdn.microsoft.com/en-us/
library/microsoft.kinect.skeletonstream.
enabletrackinginnearrange.aspx
• Optimally your users should be standing
between 1.5m and 2m away from the sensor.
• The sensor should be placed above or below
the screen, wherever it has the least amount
of tilt and can see the most of your users’ bodies.
It helps a lot if it can see your users’ heads.
• The sensor should be centered on the screen
so that users can stand directly in front of it.
• There should not be large amounts of natural
light, or people walking in front of your users.
For information about button controls,
see Kinect Button, later in this
document.

Basic Interactions > Best Setup for Controls and Interactions
Screen resolution
The Kinect for Windows controls we’ve built were designed for 1920x1080 resolution screens.
Whether you’re using these controls or creating your own, keep in mind that different screen
resolutions affect the intended size of the control. Also, because the Physical Interaction Zone
(or PHIZ, described in the following section) has a fixed size relative to the person, and doesn’t
adjust for the screen size or orientation, it might be helpful as you resize your controls to focus
on fine-tuning the ratio of control size to PHIZ size, to ensure that the experience is still reliable
and the control is still easily selectable.
The smallest button we’ve designed is 220 by 220px in 1920x1080 resolution. We’ve tested to
ensure that this button size and resolution combination makes it possible to have close to 100
percent accuracy for most users. If you have a different resolution, however, you need to resize
the button to make sure the ratio is maintained, to keep the reliability. The following chart
shows how this small button size translates for different resolutions.
UHDTV
(W)QHD
1080p
WSXGA+
HD+
WXGA+
WXGA
720p
XGA
SD
HVGA
Kinect Region (px) Button (px)
Width Height Width/Height
7680
2560
1920
1680
1600
1440
1366
1280
1024
720
480
4320
1440
1080
1050
900
900
768
720
768
480
320
880
294
220
214
184
184
157
147
157
98
66

Basic Interactions > Setting the Stage: the Kinect Region, the PHIZ, and the Cursor
Setting the
Stage: the Kinect
Region, the
PHIZ, and the
Cursor
The Kinect Region, the Physical Interaction
Zone (PHIZ), and the cursor are all things
that you need to get started with your Kinect
for Windows–enabled interface. You can
easily set them up so they’ll work the best
for your scenario.

The Kinect Region
The Kinect Region is the area on the screen
where Kinect for Windows–enabled interactions
are possible in your application. That means
it’s also the only area where your Kinect for
Windows cursor will appear. As you plan and
design your application, consider where and
how big you want this area to be, how itwill
affect the user experience, and how it will fit
within the orientation of your screen (landscape
or portrait).
To try out how it would feel with another setup, use a mouse to click the Restore button
at the upper-right of the Interaction Gallery sample window, change the size of the
window, and drag it to different parts of the screen. Also, notice that all controls in the
sample are mouse-enabled, but are built to be used primarily with Kinect for Windows
interactions, with mouse as more of an administrative or testing input. If part of your
application is mouse- and keyboard-enabled and part is Kinect for Windows–enabled,
consider putting the non–Kinect-enabled controls inside the Kinect Region and then
hiding them, along with the mouse cursor, when the user interacts with Kinect for
Windows. For more information, see Multiple Inputs, later in this document.
Notes
Full window
In the Interaction Gallery
sample, shown at right,
the Kinect Region is the
entire window; when the
sample is maximized, which
it is by default, that’s the
entire screen. This method
is the least jarring and
confusing for users because
it most directly translates
to the expected behavior
of cursor areas that they’ve
experienced before.

The Physical
Interaction Zone
The PHIZ aims to:
• Provide a consistent experience that
users can learn quickly, across all Kinect for
Windows–enabled applications.
• Enable users to comfortably reach everything
they need to interact with.
• Provide responsive tracking (low latency).
Area
The area of the PHIZ is relative
to the size and location of
the user, and which hand he
or she is using. It spans from
approximately the user’s head to
the navel and is centered slightly
to the side corresponding to the
hand the user is actively using to
interact. Each hand has its own
separate PHIZ.
Shape
Instead of mapping directly
from a flat, rectangular area
that has the same shape as
the screen, we take the range
of movement of human
arms into account and use
a curved surface, which
makes moving and extending
the arm within the area
comfortable and natural.
Axis
We measure X and Y
dimensions as if the curved
surface were a rectangle. We
measure Z by arm extension,
because we’ve found that
this is not always the same
as Z space between the user
and the sensor.

Ergonomics
The PHIZ stays the same
shape and size relative to
the user, regardless of the
aspect ratio of the Kinect
Region (for example, portrait
vs. landscape), to keep users
in a space where movement
is comfortable and not
fatiguing. This means that
users will always be able to
comfortably interact with
all parts of your application;
however, keep in mind that
the thinner and longer your
Kinect Region is, the harder
it will be for your users to
target precisely.
For example, if your screen is extremely tall and narrow, users will have
to move their hands much farther to move left and right, and moving
up and down will seem extremely fast.

Portrait Screen
Interactions
A large screen in portrait orientation is well
suited to individual interactions. The entire
screen can present a single user-centric
interaction surface. The user can interact with
the screen in the same way he or she might
with a dressing-room mirror.
Portrait screen design
Even though a single user can easily fit into the frame, it can be a challenge to design an
experience that involves movement spanning the entire vertical range of the portrait screen.
Mapping the PHIZ to the portrait screen will affect the cursor interaction. Moving the cursor
left and right seems slower. Moving the cursor up and down seems more rapid, and makes fine-
grained control more difficult.
UI placement
Place the controls within
easy reach of the user, for
example, near the center of
the PHIZ.
Place interaction points away
from the margins of the
screen and toward the center
so that a user can easily use
either hand.

User considerations
Avoid long interactions that
require the user to keep his-
or-her hands raised for more
than a few seconds, such as a
vertical scroll.
Avoid fine-grained
vertical motions in your UI
interactions. Fined-grained
targeting along the vertical
axis will cause fatigue in
users.

Variations in user height
Portrait screens should account for users of various heights. The UI elements will either need
to be designed to accommodate an average user height or should use Adaptive UI design to
target individual users. As we mention in Remember your audience earlier in this document,
keep your audience in mind.
Consider sensor placement and limitations
If the sensor is placed at the bottom or the top of the portrait screen, it may have difficulty
tracking the user at a steep angle. The sensor will see the user’s skeleton and gestures from an
angle and if the user is very close to the display, the skeleton may be clipped. A sensor placed
at the top of the screen may have difficulty recognizing a short user who is standing close to
the screen. As we mentioned in Consider Sensor Placement and Environment earlier in this
document, the best practice is to anticipate these limitations and to test the interactions.

The Kinect for
Windows Cursor
Much like a mouse cursor, the Kinect for
Windows cursor provides visual cues to your
users as they target, select, and scroll or pan by
moving their hand in the PHIZ.
Although we’ve built a lot of feedback into our
controls and we suggest you do the same for
any new ones you build, we’ve made the cursor
a strong mechanism for providing feedback.
Pressing
As a user extends his or her arm in a pressing motion, a color fill goes up and down within the
cursor to indicate how far along in the press the user is. This feature was heavily user-tested
and has shown to improve pressing accuracy and learnability. It also makes it easier for users to
cancel before a press is made, or retarget if a press is about to happen over the wrong control.
The cursor also has a visual state that indicates when the user has completed a press.
Gripping
When Kinect for Windows detects that the user’s hand is in a gripped state, the cursor changes
to a gripped visual and shows a color consistent with the fully pressed state. This confirms
to the user that the grip is detected, as well as being a strong visual cue for how to make a
recognizable gripped hand.
Default
targeting
state
Targeting
over
something
that is
actionable
(grip or press)
Gripped hand
detected
Fully pressed
state
Right hand vs. left
hand cursors
Standard cursor and feedback graphics
Progress
indication
(color fills
hand as the
user presses
further)

The cursor moves freely within the Kinect Region only when the user’s hand is in the PHIZ; if
the user’s hand is anywhere above and to the sides of the PHIZ, the cursor sticks to the top, left,
or right of the window, and becomes semitransparent. This provides feedback for how to get
back into the area where the user can interact. When the user’s hand falls below the PHIZ, the
cursor drops off the bottom of the screen and is no longer visible, allowing the user to rest or
disengage.
Because we’ve built only one-handed interactions (see Vary one-handed and two-handed
gestures, earlier in this document), by default we show only one cursor for one user at any
given time. This cursor is determined by the hand of the primary user who first enters the PHIZ
(see the following section, Engagement). The primary user can switch hands at any time by
dropping one hand and raising the other.

Basic Interactions > Engagement
Engagement
With most human-computer interactions, it’s
easy to know when users mean to interact with
the computer, because they deliberately move
the mouse, touch the keyboard, or touch the
screen. With Kinect for Windows, it’s harder
to distinguish between deliberate intent to
engage and mere natural movement in front
of the sensor. We leave the details of how to
handle this up to you, because your specific
scenario might have special requirements or
sensitivity.
This section describes the default behavior and
the User Viewer control, covers some things to
think about as you design for engagement, and
explains the solution we’ve implemented in the
Interaction Gallery sample.

Default engagement
behavior
The Kinect for Windows interaction model
considers the engaged user to be whoever is in
control of the application.
By default, the engaged user is the same as the
primary user, and the primary user is selected
based solely on user hand states observed.
These defaults can be overridden.
Hand states
Other than “Not tracked,” they are not mutually exclusive:
• Not tracked – No hands are being tracked.
• Tracked – Kinect is tracking at least one skeleton.
• Active – Hand is tracked, and raised anywhere above the bottom of the PHIZ,
at any reachable arm extension – for example, when the arm is reached all the way
up, out to the right, or out to the left, and anywhere in between.
• Interactive – Hand is tracked, active, and inside the PHIZ (a smaller rectangle that
is within the active area described above).
Primary user
Only one user can control the single cursor at any time; this person is the primary user. Default
primary user selection can be overridden by your application to fit your scenario:
• The primary user is designated when the first tracked person raises his or her hand.
• When the primary user drops his or her hand below the PHIZ (out of the active area), or
leaves the view of the sensor, he or she stops being the primary user, and another skeleton-
tracked user can raise a hand to become the new primary user.
• While the primary user has an active hand, another user cannot take over.
• Even though the sensor can see up to six people, only two of them will have tracked
skeletons. The primary user can only be one of these two fully tracked users.
It might make sense for your application to cycle through tracked skeletons, to enable
all (up to six) users to have the opportunity to be active.
Notes

Colors
• Color for the primary (engaged) user
• Color for any other users who are detected but not tracked
We’ve chosen some generic colors as default for this control, but you can set them to
whatever’s appropriate for your design and brand.
The User Viewer helps to answer the following engagement
questions for your users:
• Can the sensor see me?
• Is any part of my body clipped? Is my hand still in view?
• Am I the primary user in control of the experience right now?
• Is it possible for me to take over control of the application?
• Someone just walked in between me and the sensor: am I still tracked?
• Is the person behind me or next to me confusing the sensor and making my cursor jumpy?
The User Viewer
We’ve built the User Viewer control to
help make engagement easier for users to
understand. It’s a simple view that shows
silhouettes for any user tracked by the
sensor and allows developers to set colors
for any visible players. By default, it uses two
colors to distinguish between the primary
user and others.

Including the User Viewer in your application helps your users understand how to place
themselves at the right distance and orientation for the sensor to see them. It also reduces any
feeling that the application or Kinect for Windows is buggy, unresponsive, or unexplainably
jumpy, and helps users figure out what is happening and how to make their experience better.
As mentioned above, this feedback is especially important in Kinect for Windows interactions,
because users moving freely in space will quickly feel a frustrating lack of control if the system
does not behave as they expected. For more information, see Feedback, earlier in this document.
You can size and place the User Viewer anywhere in your user interface.
In the Interaction Gallery sample, we use the user viewer in two different ways:
A full-screen overlay on the
Attract view
The purpose of the Attract view is to show
users that we can see them, and entice them
to interact with the application. The large
User Viewer helps to get this message across
and serves as an interesting visualization.
Small and centered at the top
on all other screens
This visually cues users to what the sensor
sees, who the engaged user is, and how to take
over engagement. We also use the small User
Viewer as a launching point for application
messaging and user education, and replace it
with the Sensor Chooser UI if there is a Kinect
for Windows connectivity error.

Considerations
for designing
engagement
It’s important that when a user is ready to
interact with your application, they can do
so easily.
The top engagement interaction challenges are to ensure that:
• Users who want to engage, can.
• Users who aren’t trying to engage, don’t.
• Engaging feels natural and intuitive.
• Engaging doesn’t feel like significant additional effort.

Ease of engagement
When considering how to design your
application’s engagement strategy, consider how
and where you expect your users to interact
with it. In some scenarios, it should be harder for
users to engage.
Quick, easy engagement
If you’re not concerned about false positives,
a “low barrier” solution might simply look
for a tracked user’s hand entering the PHIZ
(being interactive), such as our default
behavior does. For example, this might be a
good method for a simple interactive sign.
Deliberate, safe engagement
A “high barrier” solution could build on
top of a low barrier, looking for additional
factors in order to remove false positives. For
example, you might want this when data or
consequences are more critical.
With the high-barrier solution, after the low-barrier criteria are met, hinting could show
what else is required to complete engagement.
You might trigger engagement when the user:
• Faces the screen
• Raises a hand into the engagement area
• Is in front of the sensor and remains stationary for a short time
(less than 500 milliseconds)
Notes

Common false
positives for
engagement
Your challenge is to design actions that
feel easy to the user, but also don’t risk
false positives.
Here are some common user behaviors that can
potentially be misinterpreted:
• Holding something (such as a drink or mobile phone)
• Moving a hand toward the body or resting it on the body
• Touching the face or hair
• Resting an arm on the back of a chair
• Yawning and extending arms
• Talking and gesturing with other people

Initial engagement
In the Interaction Gallery sample, we
demonstrate an example of a more
complicated engagement model that has a
“speed bump” for initial engagement and
then allows deliberate engagement handoff
between users.
Here is how the engagement flow works:
• The application starts in Attract view, which cycles through images,
enticing users to come and interact.
• As users who walk by are detected, they appear as gray silhouettes in
a full-screen overlay on the Attract view.
• When a user pauses and faces the sensor, her player mask turns
purple and she is directed to raise her hand. (For more information
about player masks, see Multiple Users, later in this document.)
• Once the user raises her hand, her silhouette is replaced with a
cursor, and she is presented with a large button and instructed to
“Push Here.” Note that this also serves as user education, teaching
the press-to-select interaction. In user research, we found that this
was enough to teach people to successfully press to interact with the
rest of the application.
• After she presses the large button in the center of the screen, she is
successfully engaged and enters the Home view of the application.
This is a fairly high barrier to entry, requiring a user to perform a deliberate action in
order to engage. Another option would be to skip this step and go directly to the home
page after a user raises his or her hand into the PHIZ.
Notes

User handoff
After initial engagement, the Interaction
Gallery sample demonstrates how users can
hand off to another user.
The following occurs in the event that one user decides to
relinquish control to another:
• If, at any time, the primary (interacting) user drops her hand below
the PHIZ, a second skeleton-tracked user has a chance to take over
control and is assigned a blue color in the User Viewer.
• Messaging comes out from the small User Viewer to inform the second
tracked user that he has an opportunity to raise a hand to take over.
• If the second user raises his hand, he has temporary control of the
cursor and is prompted to press a button to take over.
• If, at any time before the second user presses, the first user raises
her hand, she is still the primary user and will remain in control of
the application, removing the ability for the second user to take over.
• If the second user completes the press, he is now the primary user,
and has control over the cursor.
• If the primary user leaves, the application will give any other tracked
user the chance to confirm engagement and remain in the current view.

This solution overrides the default engagement behavior
in the following ways:
• The primary user drops his or her hand but does not completely
disengage and other users cannot immediately take over control.
• The primary user does not immediately switch when one user drops
his or her hand and the second user raises his or her hand.
• A third User Viewer color is used for the candidate user who has the
option of taking over control.
• The candidate user who is trying to take over control is temporarily
the primary user, but will not remain so if he or she does not
complete the required speed-bump action.

Basic Interactions > Targeting
Targeting
After users have engaged with a Kinect for
Windows application, one of the first tasks is
to target an object (get to an area or item they
want to open or take action on).
Where people traditionally move a mouse or
move their hand over a touch screen, they can
now move their hand in the PHIZ to control a
cursor on the screen. We’ve enabled targeting
from a distance to enable natural integration
with other distance-based interactions.
We’ve found that people naturally move
their hands in the X, Y, and Z axes even when
they think they’re moving only in a single
axis. Our ergonomic PHIZ helps translate to
the screen what users think and feel they are
doing. As with any new input method, there’s
a slight learning curve, but users quickly learn,
especially as they continue to encounter the
same experience in all Kinect for Windows–
enabled applications.
As you design interfaces where users target
objects, consider the following.

Make controls easy to target, with
appropriate spacing, placement, and sizes.
Make sure visual feedback matches
user intent.
Provide audio feedback when the user
targets an actionable item.
Provide ambiguous feedback.
Make it clear which objects the user
can take action on.
Crowd multiple controls together or
make them too small to reliably hit.
(Our recommended smallest button size is
220px; see Screen Resolution.)
Make users guess which items they can
interact with.
Do Don’t
Basic Interactions > Targeting
Lorem
Lorem
ipsum dolor
sit amet,
consectetur
Lorem ipsum dolor Lorem ipsum dolor
actionable non-actionable

Basic Interactions > Selecting
Selecting
Typically, users target an item in order to take
action on it. For example, they may want to
select a button that triggers a reaction, such as
navigation, in an application. In this document,
when we refer to selecting, we mean selecting
an item and triggering an action.

With the Kinect Button and Kinect Cursor components in the Developer Toolkit, we are
introducing a new interaction for selecting with Kinect for Windows: pressing. We’ve
found that when presented with a Kinect for Windows control, users naturally think of
pressing with their hand in space; it maps nicely to their understanding of the physical
world where they press physical buttons or extend their arm to point at objects.
Currently, the components we’ve built support only the ability to select an item to
trigger an action, but you could extend and build on the interaction to make it work
for other selection scenarios, such as toggling multiple items in a list.
Along with the press interaction, we’ve provided you with two styles of Kinect Button
as components in the Developer Toolkit, designed specifically to provide a good
pressing experience as well as feedback. We’ll cover our new button styles and tips
later in this section.
Although we’ve built only buttons so far, we think it’s a good idea to use the press
interaction for other selectable controls where it feels natural. If you’re building your own
Kinect for Windows controls, consider using the press gesture where it translates well, to
keep experiences consistent and avoid forcing users to learn a large gesture set.
Notes

Why is pressing
better than hovering?
The Basic Interactions sample in the 1.6 version
of the Developer Toolkit showed hovering as
the solution for selection. In that model, a user
must target a control, and then hold his or her
hand over that control for a specified amount
of time to make the selection. Although it’s
easy to learn, and dependable, hovering
over controls is not something users can get
better or faster at, and as a result, selection
interactions can seem tedious and slow. People
also tend to feel anxious about the hover
progress timer starting, so they keep moving
their hand, or feel they have to hold their hand
in a blank area so as not to inadvertently select
something. Pressing allows users to be precise,
but to go at their own pace and use a more
natural movement.
As we designed the pressing model, our goals were:
• Users can select with no training.
• Users can select anywhere on the screen.
• Performance improves over the hover model.
We want users to be confident when they press to select and we want to make
inadvertent selections rare. Here are some of the problems we’ve worked to solve while
translating users’ hand movements and arm extensions in space to a press on the screen:
• A user means to simply target (moving in X and Y axes) when
they are physically making some movements in Z.
• A user means to push or press “straight out” in Z when, in fact,
he or she is moving quite a lot in X and Y.
• Different users will start pressing at different arm-extension levels
(for example, some with their arm close to their body, some with
their arm almost fully extended).
• Users need to cancel an inadvertent press gesture.
• Users need visual progress feedback: how much more do they need
to press? Are they pressing when they don’t mean to?
• Users need to learn to press in order to select.
1.6 hover model 1.7 press model

Kinect Button
The Developer Toolkit provides you with two
styles of Kinect Button, a component we’ve
built to make the most common selectable
control easy to add to your application and
to customize for your experience. We’ve
built in feedback to help users understand
that the control is targetable, which control
they’re targeting, and how to interact with it
in 3D space. A combination of the button’s
changing size and the cursor visuals helps
users understand that they need to press and
how far they need to press in order to select
it. It also shows them if they’re in the middle
of a press so they can cancel if they need to.
In designing pressing with our button controls, we decided to trigger on release, similarly
to standard touch and mouse behavior. This enables users to cancel a press at any time
by moving their hand up, left, or right. Moving their hand down will trigger a release,
however, because we found that as people get confident with their presses, they often
drop their hand as they release.
Although the buttons were designed for Kinect for Windows interactions, they also
respond to mouse, keyboard, and touch input.

Button styles
We are providing two button styles you can
use: tiles and circle buttons.
You can change button sizes, shapes, and
colors, and also create your own, but we think
these styles are a good baseline and will cover
most of the common scenarios. The buttons are
the recommended sizes for 1920x1080 screens.
If you’re developing for a different screen
resolution, be sure to adjust the size, to make
sure your users can both hit them accurately
and read button text from the distance you
expect them to interact from. For a chart
of how button size translates for different
resolutions, see the chart in Best Setup for
Controls and Interactions, earlier in this
document.
In the Interaction Gallery sample, we show
an example of how to use the tile button style
tailored to specific experiences and styled for a
specific brand. Look at the Video Player module
to see a slightly different use of a circle button
(which is outlined on the following page).
Tile buttons
We provide three sizes for tile buttons: small
(220px), medium (330px) and large (440px).
• Tile buttons can fit in a grid, with equal padding between tiles. Padding can vary, if
hit areas don’t overlap. For usability, button size is more important than padding.
• Tile buttons are composed of a background (which can be a solid color or an
image/symbol) and a text area that overlays the background on the bottom edge.
• If text is omitted, the overlay does not appear. If an image is omitted, a solid color appears.
• You can easily resize them to fit your design.
• Tile buttons are good for listing items, launching pages, or navigating.
• Tile buttons are built to be easy for users to target and select – keep them
somewhat large so that they‘ll remain easy to target.
• The border property allows for invisible padding that is hit-targetable. This allows
buttons without dead space between them, making it easier for users to select
within a group of items. By default, tile buttons have invisible 5px borders.
In the Interaction Gallery sample, we use tile buttons for:
• Tiles on the home screen that navigate you to each of the modules
• Images that launch shadow boxes in both the horizontal scrolling view and the Article view
• The buttons for the engagement speed bumps

Circle buttons
These are used mostly for simple
navigation buttons (such as Back or
Home) or for settings.
• The design is a circle surrounding a glyph of your choice; you can also add text
below or to the right of the circle.
• You can scale the size of the circle and text to fit your needs and resolution.
• You can replace the circle and glyph with an image of your choice.
• Make sure that they are easy to target and select on your resolution and screen size.
• The button and text are inside a rectangular area that is all hit-targetable – this
enables users to be less accurate and still hit the button. The area of the button is
larger than the visuals, which helps reduce clutter without making users struggle
to select the buttons.
• By default you can use them in black or white, or you can re-template them to
use a brand color of your choice.
In the Interaction Gallery sample, we use circle buttons for:
• Back buttons in various views
• The Close (X) button on shadow boxes
• The Play/Pause/Replay button on the video view
Normal Hover Pressed Disabled
PLAY PLAY PLAYPLAY

Basic Interactions > Panning and Scrolling
Panning and
Scrolling
Scrolling enables users to navigate up and
down, or left and right; panning can enable
users to navigate freely in X and Y within a
canvas, like dragging a finger over a map
on a touchscreen. Experiences often allow
users to scroll and pan continuously, pixel by
pixel through content, at intervals through
a surface or page, or between consecutive
screens or views.

With the Kinect Scroll Viewer component, we’re introducing a new direct-manipulation,
continuous panning and scrolling interaction: grip and move. Kinect for Windows can detect
the user’s hand closing into a fist, called gripping. This interaction is a solution for scrolling
through lists or panning on large canvases, but might not be the strongest interaction for
navigation between discrete pages, screens, or views. The gesture allows for a high level of
control, but can be fatiguing to do repeatedly or for large jumps. You can place content or
controls inside a Kinect Scroll Viewer to add this experience to an application.
Although we’ve built grip recognition to work specifically for scrolling or panning,
consider using it for similar interactions, such as zooming, drag and drop, or rotating.
Gripping works best if the user is no more than 2m away from the sensor.
Gripping works best if users’ wrists are easily seen. Encourage users to remove large coats
or items on their wrists before interacting by gripping.
For paging or view-changing scrolling scenarios, consider using Kinect Buttons in the
scroll viewer, or above it, to help jump users to the place they’re looking for. When there
are discrete sections, it may be faster and less frustrating to navigate straight to them,
rather than scroll to them with direct manipulation.
Notes

Why is gripping to
scroll better than
hovering?
The Basic Interactions sample from the 1.6
version of the Developer Toolkit showed an
example of scrolling through a list of items by
targeting a large button and hovering over it,
making the canvas move at a constant pace.
Like using hovering to select, it was very easy
and reliable to use, but also frustrating and slow.
Although there are ways to make hovering to
scroll work better, such as allowing acceleration,
we’ve found that direct manipulation with grip is
a fun interaction and allows users to control their
speed and distance more deliberately.
As we worked on this new interaction with panning and scrolling in
mind, we had the following goals:
• Provide feedback when the user grips and releases.
• Enable users to successfully scroll short distances with precision.
• Enable users to scroll longer distances without frustration or fatigue.
1.6 hover model 1.7 grip model

Kinect Scroll Viewer
In order to provide you with the best arena for
panning and scrolling through lists by gripping,
we’ve built the Kinect Scroll Viewer for you
to use in your applications as a component
from the Developer Toolkit. Although it was
designed for Kinect for Windows input, it also
responds to mouse, keyboard, and touch input.
Developer Options
• You can enable and disable scrolling in X or Y axes.
• The control allows free panning when both X and Y are enabled (imagine dragging
around a canvas).
• Use the Kinect Items Control to create a databound list of items to scroll through.
detail
User Experience
• Users can grip anywhere within the Scroll Viewer and drag to directly manipulate
the canvas.
• Users can grip and fling to scroll longer distances, and the canvas will continue
moving while being slowed by a set amount of friction.
• The Kinect Scroll Viewer tries to correct for accidental scrolls in the wrong direction
as users repeatedly fling.
• Users can stop a moving canvas at any time by gripping or pressing on the scrolling area.
• When the end of a scrollable area is reached, it has a slight elastic bounce to provide
feedback to the user.

Users should be able to scroll or pan by gripping any portion of the screen that actually
moves when scrolled (any part of the Kinect Scroll Viewer). When the Kinect Cursor is over the
Kinect Scroll Viewer, by default the background color of the Kinect Scroll Viewer changes to
indicate the grippable area. We found that this color change makes the interaction easier to
understand and complete. The Kinect Scroll Viewer enables users to move their gripped fist
slowly for finer control, or “fling” the content if they want to traverse a longer distance. The
fling gesture is particularly helpful when users want to reach the beginning or end of a list
quickly. Kinect for Windows ignores pressing when the user’s hand is gripped, except to stop
scrolling, or unless the user has already completed a press when a grip is detected.
The visual padding at either end of the Kinect Scroll Viewer, along with the elastic effect
during scrolling, and the bounce when the end is hit from a fling, help to indicate to the user
that they’ve reached the beginning or end of a list.
We suggest that you avoid using long scrollable lists of content in applications, because
repeatedly doing any gesture can be fatiguing and frustrating. Try to ensure that most users
can reach either end of a list with no more than two or three repetitions. Grip-and-move to
scroll or pan can be a fun and novel experience, but grip recognition while users are quickly
moving their hands is not extremely reliable, so we suggest that you reserve the Kinect Scroll
Viewer for non-critical tasks. Combining grip-and-move with other gesture interactions might
also make them both slightly less reliable.
Ergonomically, horizontal scrolling is usually easier and more comfortable for people than
vertical scrolling. Where possible, structure your user interface to allow for horizontally
scrolling content.
Also, remember that, as with any new gesture, user education is important. Many users
have never experienced a grip-and-move interaction before. Grip recognition works best
when users are deliberate about their hand positions. Sometimes half-closed hands are
misrecognized as grips. Many users figure this out quickly, but having clear messaging can
help avoid initial confusion or frustration.

User Interface
• The color of the overlay that appears when the Kinect Cursor enters the Kinect Scroll Viewer
is configurable. By default, it is a semi-transparent gray.
• The only visible part of the Scroll Viewer is the scrollbar.
• A small panning indicator shows the users’ current location in the canvas and the amount of
canvas available. Users cannot grab the panning indicator; it is only provided as visual feedback.
The panning indicator only appears when the user is actively scrolling by gripping (not with
a mouse).
• If mouse movement is detected, the scrollbar appears, with a traditional thumb control and
two arrows. After mouse movement stops, the scrollbar times out and disappears, returning
to the panning indicator.

Horizontal scrolling is easier ergonomically than vertical scrolling. If you have vertical
scrolling, do not design it to span the entire height of the screen.
Kinect Scroll Viewer areas that take up larger screen space are easier to scroll through.
It is less fatiguing for users if they don’t have to reach across their body to scroll.
Be sure to provide clear user education when you include grip scrolling in an interface.
Notes
In the Interaction Gallery sample, the Scroll Viewer is used:
• In the Pannable Map view, for
2D panning
• In the Article view, for scrolling
the text vertically

Basic Interactions > Zooming (Z-Axis Panning)
Zooming (Z-Axis
Panning)
Zooming makes objects on the screen larger
or smaller, or displays more or less detail. Many
people are familiar with using a zoom control
with a mouse and keyboard, or pinching to
zoom on a touch screen. Zooming with Kinect
for Windows can be especially challenging
because it’s much harder to be precise about
distance, or start and end points, when users
aren’t directly touching the surface.
Currently, we don’t provide controls or
interactions that help with zooming, but we
recommend using our grip recognition if you
do your own implementation of zooming, to
keep it consistent with the scroll interaction.
The following suggestions are based on existing
zooming UI that people might be familiar with.

Triggered zoom mode
UI that represents the hands,
or a symbolic visual, such
as arrows, to indicate that
zooming has been initiated.
This is useful if, for example,
the user triggers zooming by
holding both hands up.
Zoom control UI
UI that can be similar to a
slider from 0 to 100 percent.
It might be useful to enable
users to grab and drag the
slider, or press and hold
Forward or Back buttons on
either side.
VUI
Voice commands like
“Zoom 100%,” “Zoom in,” or
“Zoom out,” which enable
users to jump to different
zoom increments. Avoid
forcing people to do too
much repetition of voice
commands.
+

Proportion of change in
two-handed zoom
Look for the user to hold up
both hands, and then base
the percent of zoom on the
percent of change in the
distance of the hands toward
or away from each other.
Z-axis zoom
Use a literal translation
between a user’s hand and
object, by zooming with the
hand moving in Z-space.
That is, increase zoom as the
user’s hand pulls in towards
the body, and decrease as
the hand pushes out.
This might feel most intuitive and familiar to users, because it’s similar
to the touch interaction. However, it can be a challenge to implement it
well. For example, you need to define the start and end of the gesture
accurately to make the zoom level stick at the appropriate location.
Z-space is typically harder to work with, so this can be a more
challenging, although intriguing, approach.
The goal of zooming is to manipulate an object on the screen and see the results. Here
are some ways to map user actions to zoom responses. As with anything that requires
direct manipulation, try to avoid lag, and keep the UI as responsive as possible.

Trigger + zoom gesture
Enable users to trigger
zooming mode by voice or
a distinct gesture, and then
to zoom in or out by using
another gesture.
A set increment triggered
by a directional gesture
Specify the same increment
for every zoomed gesture.
This can be tedious and frustrating for big zooms, but can work well
for semantic zooming.
Semantic zooming is when the zoom makes a more meaningful
content change – such as zooming out of one page in a document to
see all the pages in the document, or zooming in on a collection of
tiled items to display titles or dates.
It’s important to avoid false activations with this method, because it
could be very disorienting to trigger zooming by mistake.

Basic Interactions > Text Entry
Text Entry
Voice and gesture aren’t strong input methods
for text entry, and you should avoid designing
for it. Keep in mind that text entry with gesture
is usually a series of targeting and selecting
actions, which can get frustrating and tiring if
the user must do it multiple times in sequence.
If you expect users to need to enter text, and
they have access to alternative inputs such as
a touchscreen, it’s best to direct them to that
input method.

Avoiding ineffective input
Writing content by gesturing
or voice is very difficult, slow,
and unreliable. Searching or
filtering can be successful if
the user only needs to select
a few letters.
Virtual keyboard
A virtual keyboard is a
text-entry UI that people
might be familiar with. It an
allows for brief text entry by
targeting and selecting from
a list of letters.
Most targeting and selecting enhancements we’ve described for other inputs can be combined
to make text entry easier. For example, it can be useful to increase collision volume (a specified
range beyond the visual boundary within which an object responds to input) based on predicted
words, or filter letters available based on completion. As always, be sensitive to being too forceful
or presumptuous about what your user’s intent is, and leave them in control.

Enable text entry for searching, or filter
through a small set where only a few
letters are required.
Enable voice text entry with a small
number of recognized words.
Use voice for short phrases and for a
limited and appropriate set of tasks.
Require long text entry with a gesture
that imitates a keyboard experience.
Do Don’t
Require voice text entry with individual
letters (sounds are too similar: “B,” “P,” “V”).
Require long phrase dictation or
conversational voice input.

Additional
Interactions
In addition to the basic interactions of
targeting and selecting, scrolling, zooming,
and entering text, Kinect for Windows
enables more complex distance-dependent
interactions, as well as multiple input modes
and multiple-user scenarios.
Additional Interactions

Distance-
Dependent
Interactions
With Kinect for Windows, users no longer
need to directly touch a computer in order to
interact with it. Of course, this introduces an
interesting set of considerations for designing
interactions and interfaces for larger distances.
This section describes how you can make your
interface more usable at any distance.
Additional Interactions > Distance-Dependent Interactions

Interaction ranges
Users can interact with Kinect for Windows
from a variety of distances. These distances
are divided into the following categories:
Out of Range, Far Range, Near Range, and
Tactile Range. (Note that Near Range does
not refer to the Near mode setting detailed
in Meet the Kinect for Windows Sensor,
earlier in this document.)
Out of range (over 4 meters)
Most Kinect for Windows
interactions aren’t feasible
at this range. Your UI should
focus on broadly informing
users that an interesting
interaction is available, and
enticing them to move closer
with an Attract view.
Gestures that are fairly coarse-grained, and short commands, work
best in this range. The UI must be large enough to be visible from
far away.
Far (2.0-4.0 meters)
In Far Range, the user’s full
skeleton typically is visible.
Seated mode is also available.
Visuals must be very large and simple, and in some cases you can use
audio to get users’ attention. You could use this range to see general
shapes – for example, to see where movement is in the room.

Near (0.4-2.0 meters)
In Near Range, a full
skeleton might be partially
obscured or cut off, but
this is a good distance for
seated mode.
Because the user is near to the sensor, you can have fairly detailed
visuals, longer phrases for speech, and finer gestures and depth data.
This is also a good range to work in if you plan to require object or
symbol recognition.

Tactile (0.0-0.4 meters)
This is the range at which
people could use a touch
screen; thus by definition,
they must be no further
than an arm’s length away.
Because the user is close to the screen, this range can have the
highest level of detail for visuals and controls. (Many items, small
items, and fine detail.)
If your user is in Tactile Range, mouse, keyboard, and touch might be
more appropriate input methods than voice or gesture. Neither depth
nor skeleton data is available at this distance; depth data is limited to
at least 40cm in Near Range, and at least 80cm in Far Range.

Content, text, and
target sizes
As the user gets further away from the screen,
the text and target sizes should adjust to
account for what he or she can see, and what
Kinect for Windows can adequately hear.
Also consider the increasing imprecision and
decreasing detail in hand-tracking and gestures
over distance.
Graphics-to-text ratio
As distance increases, consider increasing
the graphics and size of text, and reducing
the amount of text to maintain user
engagement and reduce visual strain.
Transitions
As the UI adjusts to users’ distance, use
transitions to help them keep their context
and understand how their movements affect
what they see.
User orientation
Consider using anchor points or unchanging
sections to orient users and provide consistency.
Visual consistency
Be consistent in the general location of items
that users can take action on.

Item placement and
quantity
Two things factor into determining the
number of items on the screen that users
can take action on, and where the items
should be placed:
• The distance between the user(s) and
the sensor
• Screen size and resolution
When designing screens, keep the following
in mind.
Object grouping
In general, grouping items in an easy-
to-reach area can help users have faster
interactions. Corners and edges tend to be
harder to target than the center.
Distance adjustments
With longer distances, text has to be larger
and gestures less detailed, which will most
likely lead to fewer gestures and voice
commands in Far Range.

Adaptive UI
With an adaptive UI, your application can
dynamically adjust to optimize the experience
for a user based on his or her position and
height. An adaptive UI is most useful on large
displays where the user may not be able to see
or reach all of the content on the screen. The
application can adapt to show an appropriate
UI for the user based on his or her distance
from the screen.
Points of orientation
Remember to establish and maintain visual consistency as the application adapts to the user
in each interaction range. Refer to Content, text, and target sizes earlier in the document.
Consider the user-experience flow, and design the application so a user in the Tactile Range
can comfortably see and reach the interface elements related to his or her task without having
to move away from the screen.
Adapt to variable positions and user height
Since users can interact with Kinect for Windows sensor from a variety of distances, you can
design your application for the user at each interaction range.
• Users may begin interacting with the application from any range. Design your application
to adapt to the current position of the user. Refer to Interaction ranges earlier in this
document.
• Guide the user through the different ranges with appropriate cues and transitions so
he or she is never confused about what can be done at any given location. Refer to “User
orientation” and “Layout continuity” in Feedback Interaction Design earlier in this
document.

Transitions
Use transitions to help the user keep context as they move between interaction ranges.
Design a transition that reveals or hides
additional information or detail without
altering or obscuring the anchor points
in the overall UI.
Guide the user with clear feedback so
they can control transitions from one
range to another. If a specific distance
is used as a transition threshold, allow
minor changes in the user’s stance so the
UI doesn’t change without specific intent.
Design UI where users can accomplish all
tasks for each goal within a single range.
Reflow text or content and discard visual
context from one range to another.
Do Don’t
Rapidly toggle between states when a user
is positioned at the boundary between
interaction ranges.
Don’t separate out tasks within one goal
and disrupt the flow of the experience by
having the user repeatedly cross ranges.

User’s field of view
The user’s field of view and reach determines the most comfortable area for his or her
interactions based on his or her distance from the screen. At a Far Range the field of view may
be the entire screen. At a Tactile Range the field of view may be a small portion of the screen.
In adapting the UI to the user:
• Adjust the interface to account for the user’s field of view.
• Position the UI for Tactile Range interactions based on the user’s field of view and reach. This
is important on a larger screen where the user cannot see or reach the entire screen.
Adaptive UI at Tactile Range
Make sure the user has easy-to-reach access to the controls and information relevant to the
Tactile Range action. Avoid displaying information out of the immediate field of view that will
require the user to step back to see or access.

Adaptive UI with multiple users
For experiences with multiple users, it may be necessary to provide individual UI controls for
each user. The individual controls can also be adaptive and may be based on the position
of each user. In addition, the application may continue to provide information, visuals, and
sounds targeted to other users observing the application. For more information see Multiple
Users later in this document.

Additional Interactions > Multiple Inputs
Multiple Inputs
Much as we often use both voice and
gesture to communicate with each other,
sometimes the most natural way to interact
with an application can involve multiple input
methods. Of course, with some options and
combinations, you risk creating a confusing
experience. This section outlines the ways to
make multimodal actions clear and usable.
We recommend designing your Kinect for
Windows controls to work with multiple
inputs to allow for greater flexibility and
easier troubleshooting. Kinect Button and
Kinect Scroll Viewer are examples of controls
that work with Kinect for Windows as well as
mouse, touch, and keyboard.

Based on your use scenario, think about how you want your controls to
handle multiple inputs:
• Will users interact with the application by using more than one input
at any given time?
• Do you want users to be able to use any input method or input at any time?
Only with certain controls? Only at specified times in your task flow?
• Does one input method take priority over the others if multiple methods
are used at once?
If the Kinect Button and Kinect Scroll Viewer detect multiple inputs, they respond to
the first one detected.
In the Interaction Gallery sample, either the mouse cursor or Kinect Cursor can be
visible and used at any given time. The Kinect Cursor is on by default, but if mouse
movement is detected:
• The Kinect Cursor is disabled and invisible.
• The mouse cursor becomes visible.
• The Mouse Scrollbar appears on the Scroll Viewer.
• The application frame with windowing controls becomes visible if the application
is in full-screen view (which it is by default).
• After the mouse has been inactive for two seconds, Kinect for Windows mode
comes back on and reverses the changes.
• All controls in the application respond to mouse, keyboard, and touch input
at any time.

Single mode
interactions
With single-mode interactions, multiple
input modes are available but only one
is used per action. For example, your
application might allow people to use voice
to trigger some commands, and gesture to
navigate; or it might use keyboard for text
input in Tactile Range and skeleton tracking
and gestures at a distance. Here are a few
things to remember.
User cues
Provide visual and/or audio cues to
the user about what input methods are
available.
Backup methods
Whenever possible, have a back-up input
method that users can switch to in order to
complete any given task (for example, either
mouse or voice for selecting a button).

Multimodal
interactions
With multimodal interactions, the user employs
multiple input methods in sequence to
complete a single action.
Speech + gesture
The user points to a product,
and then says “Add to cart.”
Speech + touch
The user presses and holds a
photo, and then says “Send
photo.”
Speech + skeleton
The user says “Record,” and
then makes a movement.
Add to
cart
Send
photo
Record

Record
Reduce the number of steps for a
more complex action
For example, sending a photo typically takes
two steps using the same input method –
clicking the photo and then clicking Send.
With multimodality, this can become one
action with two input methods.
Enable triggering controls without
moving the user’s body
For example, if a task relies on a user’s body
being in a certain position, voice is a great
way to trigger an action without the person
having to move.
Here are some reasons to use multimodal interactions:
Increase confidence
For example, it’s more likely that a command
will be not be executed accidentally if the
user has to do two things to perform it.
Make the best use of each input
For example, you might require voice input
to put the application into volume-setting
mode, but require gesture to give the input
for the actual volume level.
Send
Volume

Additional Interactions > Multiple Users
Multiple Users
Kinect for Windows can track up to two
skeletons and, therefore, detailed movement
information from two users simultaneously.
This opens the way for some interesting
collaborative interactions, as well as new
challenges regarding control and inputs.
Currently our Kinect for Windows controls
and interactions don’t support multiple
simultaneous users. Only one person can be
actively interacting with them at any time.
This section covers what we know so far about
designing for multiple users.

Tracking
Kinect for Windows can be aware of up to
six people, tracking two of them in skeleton
detail, returning full joint information. For the
other four users, Kinect for Windows returns
a small amount of information: a player mask
specifying the mask or volume of the person,
and a center position.
Your application needs to select the two people who will be primary. We recommend that
the first two skeletons detected become the primary users by default.

Multiple users and
distance
As you design experiences and build your
expectations, keep in mind the number
of people who can physically be in view,
depending on their distance from the sensor.
You can expect Kinect for Windows to track
the following number of people in each
distance range.
Number of People 0 1-2 1-6 0
Tactile Near Far Out of range

Multiple users,
multimodality, and
collaboration
Enabling multimodality with multiple users
means that you can have many combinations
of people at different distances, using
different input methods. Depending on your
model for collaboration, there are some
important considerations for the different
ways of handling this.
Single-driver model
This model assigns one of the users as the
“driver” at any given time and registers
actions taken by that person only. The driver
role can be selected or transferred in a
number of ways, including designating the
first user to engage as driver, or the user
that is closer to the sensor. This is one way
to avoid conflicting inputs. Typically, this
model has visuals that show which person is
being tracked, and has only one cursor on
the screen at any time.
Equal participation model
This model takes input from both users, often
simply in the order that it’s given. This works
very well for scenarios where each user has
a unique experience to drive, as in games or
avateering (making one’s skeletal movement
affect the avatar’s movement on screen).
However, it can be a very complex model for
navigating or for basic interactions.
Collaborative interactions
Collaborative interaction is when two users interact with the same screen space and
inputs. This means that any action taken by one user affects the state of the application
for both. There are two options for balancing control in this case:
For the Interaction Gallery example of
handing off between single drivers, see
User Handoff, earlier in this document.

You can handle this with screen partitioning or sectioning. In general, this experience
should be very similar to the single-driver experience described above; however, it’s an
extra challenge to correctly map each person’s speech and actions to the corresponding
interface. Voice commands can be a challenge. Gestures and movement should have
separate visual feedback per user.
Non-collaborative interactions
In a non-collaboratve interaction, each user has his or her own sub-experience within
the interface.

Conclusion
Kinect for Windows is at the forefront of
the revolution known as NUI—Natural User
Interface. The next generation of human-
computer interaction, NUI lets people
interact with any device, anywhere, using the
movements and language they use every day
in their lives. Microsoft Kinect for Windows–
enabled applications open a broad range of
new possibilities for people to interact with
computers in ways that feel natural. From
business to the arts, from education to gaming,
and beyond, NUI expands the horizons of
application development.
In this document we’ve provided you with
the tenets and best practices for creating
NUI applications, and guidance on both the
basic and more advanced Kinect for Windows
interactions for gesture and voice. We hope
our suggestions help you create “magical”
experiences for your users. Your development
of touch-free, natural UI will shape the way
people experience and interact with software
applications for years to come.
Conclusion

Human interface guidelines_v1.8.0

More Related Content

What's hot (20)

Viewers also liked (8)

Similar to Human interface guidelines_v1.8.0 (20)

Recently uploaded (20)

Human interface guidelines_v1.8.0