Markerless motion capture for 3D human model animation using depth camera

TELKOMNIKA, Vol.17, No.3, June 2019, pp.1300~1309
ISSN: 1693-6930, accredited First Grade by Kemenristekdikti, Decree No: 21/E/KPT/2018
DOI: 10.12928/TELKOMNIKA.v17i3.8939  1300
Received February 19, 2018; Revised January 18, 2019; Accepted February 17, 2019
Markerless motion capture for 3D human model
animation using depth camera
Maulahikmah Galinium*, Jason Yapri, James Purnama
Bachelor of Information Technology, Swiss German University, The Prominence Tower, Indonesia
*Corresponding author, e-mail: maulahikmah.galinium@sgu.ac.id
Abstract
3D animation is created using keyframe based system in 3D animation software such as Blender
and Maya. Due to the long time interval and the need of high expertise in 3D animation, motion capture
devices were used as an alternative and Microsoft Kinect v2 sensor is one of them. This research analyses
the capabilities of the Kinect sensor in producing 3D human model animations using motion capture and
keyframe based animation system in reference to a live motion performance. The quality, time interval and
cost of both animation results were compared. The experimental result shows that motion capture system
with Kinect sensor consumed less time (only 2.6%) and cost (30%) in the long run (10 minutes of
animation) compare to keyframe-based system, but it produced lower quality animation. This was due to
the lack of body detection accuracy when there is obstruction. Moreover, the sensor’s constant assumption
that the performer’s body faces forward made it unreliable to be used for a wide variety of movements.
Furthermore, standard test defined in this research covers most body parts’ movements to evaluate other
motion capture system.
Keywords: 3D animation evaluation, human computer interaction (HCI), keyframe-based system,
microsoft kinect depth camera, motion capture
Copyright © 2019 Universitas Ahmad Dahlan. All rights reserved.
1. Introduction
The most common practice nowadays to animate 3D model with human skeleton is
using keyframe based system in 3D animation software such as Blender and Autodesk Maya.
The system can define each translation and rotation of each human skeleton joints per
keyframe or set of keyframes. These sequences of keyframes define the animation. Another
technique for creating 3D animation is using motion capture. It eliminates the 3D human models
animating process using keyframe based system which is time consuming and requires high
level of expertise in 3D animation. There are two types of motion capture, marker based and
markerless based motion capture [1]. In marker-based motion capture, every performer must
use a special suit with sensors and there are multiple cameras placed in a room to detect them.
Although they produce accurate result with fine details, it is very costly and difficult to
implement. One product example of this is the Vicon Blade which is used by Bethesda
Game Studio [2].
Thus, the research problem is that creating animation of 3D human model by complex
keyframe based system requires high end animation software with the high level of expertise,
which needs more time and learning curve to create a simple 3D human model animation.
Therefore the research objective is to solve the problems by analysing the capabilities of a
single Microsoft Kinect v2 sensor as a motion capture device to animate 3D human model by
replacing keyframe based animation technique. In our previous research, Kinect is used for
several applications such as gaze tracking based on head-pose estimation in advertising
area [3] and geometric learning tool for elementary school [4]. By using the Kinect sensor
technology, a markerless-based motion capture, which does not require any suit to be worn by
the performer, can be implemented [1]. Furthermore, the skill and time requirements of
animating 3D human models using keyframe based animation can be eliminated and the
complexity for marker based motion capture can be reduced. The sensor is used to replace
keyframe based animation technique in order to help people who create 3D human model
animation with the decision choosing between the two. A third-party application with MIT license
called Kinect Animation Studio was used during the experiment which only focused on capturing

TELKOMNIKA ISSN: 1693-6930 
Markerless motion capture for 3D human model animation using... (Maulahikmah Galinium)
1301
the motion of one performer at a time. The research focused on evaluating the quality, time and
cost to produce 3D human model animations using motion capture and keyframe based
animation technique. These focuses are based on Martin Barnes’ Triangle of Objective / Iron
Triangle [5].
Tong et.al. [6] conducted a research on skeleton animation motion data based on
Kinect. It converted the original Kinect coordinate data to rotation angles of each human
skeleton joint. As an output, a standard Biovision Hierarchy (BVH) motion capture file is
produced in real time for animating a 3D human model. It used openNI and NITE program
library to access the original human skeleton data from the Kinect. Reference [6] is essential for
this research since this research also used similar basic principles in preparing the information
needed, such as the length of each bone segment to calculate the initial hierarchical offsets, the
basic principles of rotation matrices of Euler angles and the base formula of the relation
between joints’ coordinate, the local coordinate and the rotation matrix. However, Tong et.al [6]
is not calculating each rotation angle.
Therefore, our research also used a method to calculate each rotation angle based on
the three-dimensional coordinates obtained from the computations of the tracked marker points
using multiple cameras to rotation angles as also mentioned in reference [7]. However
reference [7] is used to marker point while our research is used more to markerless point.
Another approach is used by Han [8] by converting 2D coordinate to 3D coordinate that
reconstruct 3D visual motion from the 2D visual human motion in a sequence of images. Thus,
the contribution ot this paper is using the basic principles of rotation matrices of Euler angles
based on reference [6] and combining with each rotation angle calculation based on
reference [7] for markerless motion capture.
2. Proposed Method
As illustrated in Figure 1, to conduct the motion capture (mocap) animation experiment,
we used three methods including develop application by processing positional data, develop
application by processing quaternion data, and implement Kinect Animation Studio. These three
methods will be explained in details in section 2.1. Initial research was done to conduct a study
on the fundamental elements of motion capture (motion capture file types, libraries for Microsoft
Kinect v2 [9], process of applying motion capture file to 3D human model in Motion Builder, and
attempting to create Motion Capture Application that produces BVH file [10]). Motion capture
animation experiment was aimed to animate a 3D human model using motion capture technique
and produce animation video. Narrated movements (T-pose, walking front and back, jump on
the spot (with both legs), jump forward (one leg at a time), saying "no" head rotation, stretching
1 (hand relief), stretching 2 (waist relief), punch 3x, front kick, side kick, soccer-kick ball,
T-pose) were defined to cover almost all body parts movement including hand, head, waist, and
leg. This may be used as a standard to evaluate the animation quality result of other motion
capture system as well. The performer performed the aforementioned movement narration to
produce fbx file using Kinect Studio and Kinect Animation Studio [11]. This was then used to
animate an inbuilt 3D human model “Aragor” in Motion Builder and an animation video was
produced. The whole process was documented using RGB cameras placed on two angles
along with screen recording on the computer as illustrated in Figure 2.
Keyframe based animation experiment was aimed to animate a 3D human model using
keyframe based animation technique and produce animation video. This was done by a 3D
animator expert using Autodesk Maya [12] software based on the reference of the RGB video
recordings of the performer’s performance from previous experiment. Lastly, an analysis was
made based on the two aforementioned experiments to compare the two animation techniques.
The animation quality was defined in a scale of 1–10 based on the animation principles using a
questionnaire to a 3D animation expert. The time taken was calculated based on the screen
recordings from each experiment. The cost of the motion capture animation technique was
calculated by adding up the value of the essential hardware with minimum specification while
the keyframe based animation technique was calculated based on the estimate fee of 3D
human model animation production inquired to the 3D animation expert in the questionnaire.

 ISSN: 1693-6930
TELKOMNIKA Vol. 17, No. 3, June 2019: 1300-1309
1302
Figure 1. Research process Figure 2. Simple architecture diagram
OpenNI library along with Nite middleware allows developers to access audio, RGB,
infrared, depth, and also skeletal information from the Kinect sensor. This was the most popular
library to be used by Kinect v1 developers. The main reason for this is because other 3
rd
party
libraries such as OpenKinect [13], LibFreenect [14] often do not support skeletal tracking.
Currently, the official Microsoft SDK for Kinect v2 has greatly improved since its previous
version. So, it is recommended to just use the Kinect for Windows SDK 2.0 to start development
for Microsoft Kinect v2 sensor. When applying a motion file to a 3D human model, several
processes (as illustrated in Figure 3 and Figure 4) must be done. A skeleton control rig must be
created from the loaded motion file in Autodesk Motion Builder. A skeleton definition template
must then be assigned to the control rig as in Figure 4. This is to let Autodesk Motion
Builder [15] aware of each specific joint available in the 3D model. Characterizing the skeleton is
to lock the skeleton definition which is required before assigning it to a 3D model.
Figure 3. Applying motion file to 3D model in
autodesk motion builder process
Figure 4. Defining skeleton
definition template
2.1. Motion Capture Application Development
A motion capture application was developed for the experiment to produce bvh motion
capture file. Figure 5 shows the motion capture application that has been developed. Notice on
tracking process has just been completed, 22 frames have been captured in 5 seconds, and
both hierarchy and sample motion data has been written to the desired output file formatted (in
bvh) can be seen on Figure 6. However, the rotation angle of the joints in motion data is not
completely processed to fully represent performer’s motion.

1303
Figure 5. Motion capture application Figure 6. Produced BVH file
BVH file includes hierarchy and motion data as illustrated in Figure 7. The hierarchy
data represents the illustration of 3D human model bone posture in T-Pose. Figure 8 shows the
first frame of the BVH files with 0-degree rotation in each joint, thus, illustrating the hierarchy
data. The bvh file was opened with bvhacker (a free source software by davedub to view bvh
file). The motion data produces a 3D human bones animation. It is represented by the rotation
angle in x, y, and z axis of each body joint with the parent joint as the pivot. From the Kinect
SDK itself, each body joints coordinate is provided in body data stream and can be used to
produce hierarchy data by calculating the length of each bone in the body using (1). There are
two main methods to solve the rotational angles required for BVH motion data; processing the
positional data and processing the Quaternion data of each joint acquired from the Kinect SDK.
Figure 7. BVH File data structure Figure 8. BVH file initial T-Pose stance
𝐿𝑖 = √(𝑥𝑗 𝑖+1
− 𝑥𝑗 𝑖
)
2
+ (𝑦𝑗 𝑖+1
− 𝑦𝑗 𝑖
)
2
+ (𝑧𝑗 𝑖+1
− 𝑧𝑗 𝑖
)
2
(1)
2.1.1. Solving BVH Motion Data by Processing the Positional Data
Figure 9 shows the illustration of a bone illustrated using cartesian coordinate system
where the parent joint is the origin, the initial child joint coordinate / offset (as defined in the
hierarchy section) is v and the current child joint coordinate is v’. The figure illustrates front view
in which the rotation of z axis can be clearly pictured. When pictured from a different view point,
the rotation of another axis can be seen. The main formula of retrieving the rotational angle from
two positional vectors is illustrated in (2) and (3). The variable R is rotational matrix consisting of
rotational matrix X, Y and Z, illustrated in (4), (5) and (6). Solving it with basic algebra can be
very complicated, if not possible as illustrated in (7).
𝑣′
= 𝑅𝑣 (2)
(
𝑥2
𝑦2
𝑧2
) = 𝑅 (
𝑥1
𝑦1
𝑧1
) (3)

 ISSN: 1693-6930
1304
𝑅(γ) = (
cos γ − sin γ 0
sin γ cos γ 0
0 0 1
) (4)
𝑅(α) = (
1 0 0
0 cos α − sin α
0 sin α cos α
) (5)
𝑅(β) = (
cos β 0 − sin β
0 1 0
sin β 0 cos β
) (6)
𝑅 = 𝑅(β) 𝑅(α) 𝑅(γ) =
(
cos α cos γ + sin α sin β sin γ sin α cos γ − cos α sin β sin γ cos β sin γ
− sin α cos β cos α cos β sin β
− cos α sin γ + sin α sin β cos γ − cos α sin γ − cos α sin β cos γ cos β cos γ
) (7)
One way to solve the rotational angle is using the traditional method by eliminating one
angle. However, this may result in some deviations. A novelty method was proposed in [7]. The
method utilizes imaginary variables which conclude the formula for the value β and γ in (8). The
problem with the proposed formula is that it always results in 0-degree rotation angle. For
instance, the spine’s initial coordinate in BVH Hierarchy setup is (0, 20, 0) and it was moved to
(10, 15, 8.66) as illustrated in Figure 10 (9) shows the solution to find R(γ) using the proposed
formula in (8). The formula only utilizes X1, Y1 coordinates which consists of value 0 and the
length of the bone between the hips and spine joint. In this case, the proposed formula cannot
be used to calculate the rotation angle.
cos β =
𝑧2
√𝑥2
2+𝑧2
2
; sin 𝛽 =
𝑥2
√𝑥2
2+𝑧2
2
; cos 𝛾 =
𝑦1
√𝑥1
2+𝑦1
2
; sin 𝛾 =
𝑥1
√𝑥1
2+𝑦1
2
(8)
𝑅(γ) = cos−1
(
20
√02+202
) = cos−1(1) = 0 (9)
Figure 9. Illustration of joint rotation Figure 10. Child joints coordinate
2.1.2. Solving BVH Motion Data by Processing the Quaternion Data
Another solution to produce the motion data in BVH file that may be found in the
quaternion data [16] is by assessing it through Kinect SDK. It represents rotation information in
any axis in the form of a 1x4 vector consisting of W, X1, Yj, and Zk values. Using quaternion data
directly from the SDK can actually improve the quality of the movements captured especially
when there is a specific body part linear rotation on a certain axis, i.e. saying “no” head rotation.
Producing the motion data using only positional data will fail to acknowledge the rotation of the
head along the y axis for the aforementioned movement. These quaternion values are used in
JointOrientationBasics application [17] in Unity to represent the rotation of each bone mesh from
xz
y
Rz
Ending point (v’)
based on current coordinate
Starting point (v)
based on hierarchy data
X
Y

1305
an upright position as illustrated in Figure 11 to the current pose of the performer as illustrated
in Figure 12. The quaternion data provided by the Kinect SDK differs from what is required for
the BVH motion data. In BVH motion data, the rotation angle is based on the parent joint and
the initial pose specified in the hierarchy part. When the quaternion data was used directly, it
leads to an unwanted posture as illustrated in Figure 13.
Figure 11. Bone renders in
unity without quaternion data
Figure 12. Joint orientation
basics unity application [5]
Figure 13. BVH file
representation from
quaternion data
2.1.3. Kinect Animation Studio
Eventually, Kinect Animation Studio was used in this research for the motion capture
animation experiment. It is developed by Marco Furtado with MIT license [18]. It is used to
retrieve body joints data and process it to produce fbx file that contains motion data. This fbx file
is essential for the motion capture animation experiment to animate the 3D human model in
Autodesk Motion Builder. The main drawback of Kinect Animation Studio software is it still uses
positional data to produce the motion data stored in the bvh file. Thus, some animations are not
perfectly depicted. Moreover, there is a jittering in the produced animation that significantly
reduces the animation quality.
2.2. Summary of Motion Capture Applications Research
The development of motion capture application was halted during the phase of
producing rotation angles for the motion data in bvh file as previously explained. During the
attempts, several methods were examined and explored. These methods are summarized in
Table 1.
Table 1. Methods to Produce BVH Motion Data
No. Method Problem Drawbacks
1
Processing the
Positional Data
The proposed formula from the related work
cannot be applied
Some movements cannot be
depicted perfectly
2
Processing the
Quaternion Data
The algorithm to process further the
Quaternion data must be further explored
(-)
3. Result and Analysis
3.1. Experiment Result
Figure 14 and Figure 15 show the hardware setup of motion capture animation
experiment and the video recording of the live performance based on the narrated movements
done by the performer. Figure 16 and Figure 17 show the screen recording of the motion
capture animation experiment while the live performance is being recorded in the computer to
produce the FBX file then the application of FBX file will drive the 3D human model in Motion
Builder respectively. After applying the FBX file, the animation was rendered in Motion Builder
and produced an animation video. Figure 18 shows the process of animating 3D human model
using keyframe based technique done by the 3D animator, while Figure 19 shows the animation
video result for keyframe based technique.

 ISSN: 1693-6930
1306
Figure 14. Motion capture animation experiment
hardware setup
Figure 15. Video recording of performer
Figure 16. Screen recording for motion capture
animation experiment
Figure 17. Screen recording for motion
capture animation experiment
Figure 18. Animating 3D human model using
keyframe based technique
Figure 19. Keyframe based animation video
3.2. Result Analysis
An interview is done to evaluate the quality, provide the feedback and retrieve
estimation cost of video production. The interviewee is a 3D animation expert who is also a
lecturer in 3D animation. Animation principles are used to measure the quality. Out of 12
animation principles (Squash and Stretch, Timing and Motion, Anticipation, Staging, Follow
Through and Overlapping Action, Straight Ahead Action and Pose-to-Pose Action, Slow in and
out, Arcs, Exaggeration, Secondary Action, Appeal, and Personality), the principle Staging
cannot be evaluated. Based on the expert evaluation, the quality score of motion capture is
26 pts while keyframe based is 50 pts. Keyframe based animation technique indeed produces
higher quality of animation. The total score that can be achieved is 110 pts (from 11 animation
principles). The quality of keyframe based animation result may be further increased by
investing in a more skilled 3D animator or investing more time in animation process.
On the other hand, the motion capture animation result may also be further increased
by using a more developed motion capture software up to certain extend, but due to the Kinect

1307
v2 sensor limitation itself, the sensor cannot detect body joints that is obstructed. Moreover, the
sensor always assumed that the performer’s body is facing towards the sensor. These two
factors reduce the quality of the animation significantly and they also limit the movements that
can be captured. This can be seen in the animation video result from motion capture animation
experiment where the movement “Front Kick”, “Side Kick”, and “Soccer - Kick ball” suffer deeply
from the obstruction limitation. Figure 20 and Figure 21 display the performer’s head towards
the camera, even when it should be facing away the camera when performing the movement
“Walking backwards”.
Figure 20. Motion capture animation video snapshot
(from left to right-front kick, side kick, kick soccer ball)
Figure 21. Motion capture animation video snapshot (walking backwards)
An alternative solution to the problem faced by Kinect sensor as suggested by the 3D
animation expert is using multiple Kinect sensors to get better motion data or using more
advanced motion tracking system such as marker based motion capture system with cameras
“Optitrack” (name of a brand) or markered based motion capture system with gyroscope
sensors [19] without any camera for a more affordable alternative. This motion capture system
with gyroscope is also equipped by accelerometer and magnetometer. Together they are called
IMU (Inertial Measurement Unit) [20]. The screen recordings from the previous two experiments
show the process to animate 3D model using two different techniques based on a reference of
59 seconds life movements performed by real performer. As shown in Table 2, the motion
capture animation technique resulted in 7 minutes and 48 seconds, while the keyframe based
animation [21] technique resulted in 3 hours 19 minutes and 33 seconds. It shows that within 10
minutes animation duration, the time to create animation using motion capture only needs
2.6-3.9% out of time for keyframe based animation.
Table 2. Time Comparison
Animation
Duration
Motion Capture Keyframe based
Percentage Motion Capture vs
Keyframe-based
59 seconds 7 minutes and 48
seconds
3 hours 19 minutes and 33
seconds
468/11973 = 3.9%
2 minutes +
/- 13 minutes +
/- 6 hours 13/360 = 3.6%
10 minutes +
/- 52 minutes +
/- 33 hours 52/1980 = 2.6 %

 ISSN: 1693-6930
1308
Cost breakdown for motion capture animation process is shown in Table 3. This
includes a laptop to run the motion capture application, the Kinect v2 sensor with tripod to be
attached together and the adaptor to connect the sensor with the computer. The cost needed for
keyframe based animation is IDR 60,000/second of animation produced. This data was given by
the 3D animation expert who filled in the questionnaire. To produce a 59 seconds animation, the
cost would be IDR 3,540,000 (IDR 60,000x59 seconds) for keyframe based animation.
However, a longer duration of animation would cost more, and motion capture has fixed cost.
This is illustrated in Table 4.
Table 3. Cost Breakdown for Motion Capture Animation Technique
# Item Name Price
1 Asus Laptop A456UR – i5 [22]
Intel Core i5 7200 U Processor, Windows 10, 14” HD Screen (1366 x 768), GTX
930MX – 2GB Graphics Card, 1TB HDD – 4GB DDR4 Memory, USB 3.1 Type C – BT
4.0 – Wi-Fi
IDR 8,299,000
2 Microsoft Kinect v2 sensor [23] IDR 1,995,000
3 Microsoft Kinect Adaptor [24] IDR 1,200,000
4 Tripod [25] IDR 225,000
Total IDR 11,719,000
Table 4. Cost Comparison of Motion Capture and Keyframe Based Animation Technique
Animation Duration Motion Capture Keyframe based
Percentage Motion Capture vs
Keyframe-based
59 seconds IDR 11,719,000 IDR 3,540,000 331 %
2 minutes IDR 11,719,000 IDR 7,200,000 163 %
10 minutes IDR 11,719,000 IDR 36,000,000 33 %
*motion capture application license not included (when available)
4. Conclusion
Although a single Microsoft Kinect v2 sensor can be used as a quick and cheap solution
for motion capture device, it has some drawbacks that made it unreliable to be used for wide
variety of movements. These drawbacks include the lack of body detection accuracy when there
is any obstruction and the sensor’s consistency to assume that the performer’s body is facing
towards the sensor. Movements such as “Front Kick”, “Side Kick”, “Kick soccer ball”, and
“Walking front and back” resulted in unfathomable animation. Overall, keyframe based
animation technique is more preferable in this case even if the time needed is far more than
motion capture. In terms of quality, the keyframe based animation is superior to the motion
capture animation, scoring 50 pointswhile the motion capture animation only scores 26 points.
However, both animations still have many rooms for improvement as the total score that can be
achieved is only 110 points, judging by the 11 animation principles with a scale of 10 points
each. In terms of time, the motion capture technique only requires 7 minutes and 48 seconds to
complete the animation. On the other hand, Keyframe based animation technique requires
3 hours 19 minutes and 33 seconds and it is directly proportional to 59 seconds of animation
video produced. So, when producing a lengthier animation video, it requires far more time for
keyframe based animation technique rather than motion capture technique. In terms of money
invested, the motion capture animation technique costs IDR 11,719,000 while the keyframe
based animation technique costs IDR 3,540,000. Even if the keyframe based animation
technique is cheaper, the cost surpassed that of motion capture when producing an animation
video of 4 minutes.
References
[1] Shingade A, Ghotkar A. Animation of 3D Human Model Using Markerless. International Journal of
Computer Graphics & Animation (IJCGA). 2014; 4(1): 27-39.
[2] Vicon. Bethesda choose Vicon for new motion capture stage | VICON. [Online] Available at:
https://www.vicon.com/press/bethesda-choose-vicon-for-new-motion-capture-stage, 2013. [Accessed
10 June 2017]

1309
[3] Galinium M, Yulius Purnama, J Gaze Tracking based on Head-Pose Estimation using Depth Camera.
International Journal of Applied Engineering Research (IJAER). 2016; 11(15): 8656-8661.
[4] Purnama J, Andrew D, Galinium M. Geometry Learning Tool for Elementary School using Augmented
Reality. Proceedings of International Conference on Industrial Automation, Information and
Communications Technology, IAICT. Bali, Indonesia. 2014: 145-148.
[5] Lock D. Project Management. 9th ed. Aldershot: Gower Publishing Limited. 2007.
[6] Tong X, Xu P, Yan X. Research on skeleton animation motion data based on Kinect. Proceedings of
Fifth International Symposium on Computational Intelligence and Design. Hangzhou, China. 2012:
347-350.
[7] Pan H, Meng Q, Gao C, Lei Y. Novel Data Conversion Method for Motion Capture. Journal of
Computer Engineering. 2012; 38(2): 17-20.
[8] Han Y. 2D-to-3D Visual Human Motion Converting System for Home Optical Motion Capture Tool and
3-D Smart TV. Journal of IEEE Systems, 2015; 9(1): 131-140.
[9] Zhang G, Li J, Peng J, Pang H, Jiao X. 3D Human Body Modeling based on Single Kinect.
Proceedings of International Conference on Biomedical Engineering and Informatics. Dalian, China.
2014: 100-104.
[10] Zhu X, Li KF. Real-time motion Capture: An Overview. Proceedings of 10
th
International Conference
on Complex, Intelligent, and Software Intensive Systems (CISIS). Fukuoka, Japan. 2016: 522-525.
[11] Arai K, Asmara RA. 3D Skeleton Model Derived from Kinect Depth Sensor Camera and its Application
to Walking Style Quality Evaluations. International Journal of Advanced Research in Artificial
Intelligence. 2013: 2(7): 24-28.
[12] Li C, Yin C, Lu J, Ma L. Automatic 3D Scene Generation based on Maya. Proceedings of 10th
International Conference on Computer-Aided Industrial Design & Conceptual Design. Wenzhou,
China. 2009: 981-985.
[13] Colvin CE, et.al. Multiple user motion capture and systems engineering. Proceedings of IEEE
Systems and Information Engineering Design Symposium. Charlottesville, USA. 2011; 137-140.
[14] El-laithy RA, Huang J, Yeh M. Study on the use of Microsoft Kinect for robotics applications.
Proceedings of the 2012 IEEE/ION Position, Location and Navigation Symposium. Myrtle Beach,
USA. 2012; 1280-1288.
[15] Obdržálek S, et.al. Accuracy and robustness of Kinect pose estimation in the context of coaching of
elderly population. Proceedings of Annual International Conference of the IEEE Engineering in
Medicine and Biology Society. San Diego, USA. 2012: 1188-1193.
[16] Xia Y, Jahanchahi C, Nitta T, Mandic D. Performance Bounds of Quaternion Estimators. IEEE
Transactions on Neural Networks and Learning Systems, 2015; 26(12): 3287-3292.
[17] Sirignano C, Lower B. GitHub - Kinect/samples: Sample applications which demonstrate platform
features and APIs. [Online] Available at: https://github.com/Kinect/samples. 2015 [Accessed 15 June
2017].
[18] Furtado M. Kinect Animation Studio. [Online] Available at:
http://marcojrfurtado.github.io/KinectAnimationStudio/.n.d. [Accessed 31 July 2017].
[19] Dong H, Gao Y. Comparison of Compensation Mechanism Between an NMR Gyroscope and an
SERF Gyroscope. IEEE Sensors Journal. 2017; 17(13): 4052-4055.
[20] Korkishko YN, et.al. High-precision Inertial Measurement Unit IMU-5000. Proceedings of IEEE
International Symposium on Inertial Sensors and Systems (INERTIAL). Moltrasio, Italy. 2018: 1-4.
[21] Pranatio G, Kosala R. A Comparative Study of Skeletal and Keyframe Animations in a Multiplayer
Online Game. Proceedings of Second International Conference on Advances in Computing, Control,
and Telecommunication Technologies. Jakarta, Indonesia. 2010: 143-145.
[22] Asus. PG 2017 - Mei Juni.indd. [Online] Available at:
https://www.asus.com/event/2017/productguide/id/ASUS_Product_Guide.pdf. 2017. [Accessed 14
June 2017].
[23] HorlickGame. Jual Kinect Xbox One New [Original] - HorlickGame | Tokopedia. [Online] Available at:
https://www.tokopedia.com/horlickgame/kinect-xbox-one-new-original. 2016. [Accessed 14
June 2017].
[24] Waroengame. Jual Xbox Kinect Adapter for Xbox One S & Windows 10 PC (ORIGINAL Product)
-Waroengame | Tokopedia. [Online] Available at: https://www.tokopedia.com/waroengame/xbox-
kinect-adapter-for-xbox-one-s-windows-10-pc-original-product. 2017. [Accessed 14 June 2017].
[25] Tripod. Jual Tripod Kamera DSLR /Tripod Kamera Digital Excell Platinum Mini + TAS - tripod |
Tokopedia. [Online] Available at: https://www.tokopedia.com/tripod/tripod-kamera-dslr-tripod-kamera
-digital-excell-platinum-mini-
tas?trkid=f=Ca592L000P0W0S0Sh00Co0Po0Fr0Cb0_src=directory_page=1_ob=24_q=_ref=csearch
_catid=592_po=11. 2017 [Accessed 14 June 2017].

Markerless motion capture for 3D human model animation using depth camera

More Related Content

What's hot (18)

Similar to Markerless motion capture for 3D human model animation using depth camera (20)

More from TELKOMNIKA JOURNAL (20)

Recently uploaded (20)

Markerless motion capture for 3D human model animation using depth camera