Presentation ICTS 2009

A Survey On Outdoor
Water Hazard Detection
Mohammad Iqbal
Olivier Morel
Fabrice Meriaudeau

Agenda
Introduction : Location, Motivation, Water
Appearance, A Variety outdoor Water Hazard
Scene.
Water detection method : Color Imagery, Short
Wave Infra red (SWIR) Imagery, Thermal Infra
red Imagery, Laser Range Finder, Multi feature
based and Polarization Imaging based
Classification Scheme of method water
detection

Location
Gunadarma University, Indonesia
Simple Information :
25.000 students and 1000 lecturers
9 campus locations
12 bachelor , 6 master and 2 PhD
programs
internal news, radio and TV station
Each department of each faculty at
Gunadarma University has
laboratories that have divisions:
Basic, Intermediate and Advance
Laboratories. The laboratories are
facilities for students as supporting
requirment for some subjects that the
students receive during the class.
www.gunadarma.ac.id

Location
Université de Bourgogne, France
Simple Information :
27,000 students divided into six
city, Dijon, Auxerre, Chalon-sur-
Saône, Le Creusot, Mâcon and
Nevers.
The Humanities and Sciences are
well represented on the main
campus along with Medicine and
Literature in separate buildings.
The IUT (Institute of technology) is
also on the campus, providing
specialist higher level diplomas in
Business, Biology,
Communications and Computer
Science.
Lab LE2I IUT (Institut Universitaire
de Technologie), Le Creusot
webcreusot.u-bourgogne.fr
12, Rue Fonderie
71200 Creusot (Le), France
+33 3 85 73 10 00

Motivation
Autonomous off-road navigation is a highly
complicated task for a robot owing to the
different kinds of obstacles it can encounter.
Water hazards such as puddles and ponds are
very common in the outdoor environment and
are hard to detect even with ranging devices
such as LADAR as the signal does not get
reflected back to the sensor.

Water Appearance
Ambient light (day versus night operation)
Scene elements reflected by the surface (sky,
vegetation or buildings)
Width of surface water (a pond versus a lake)
Water depth
Water turbidity and presence or absence of
ripples on the surface or shadow

A Variety outdoor
Water Hazard Scene
brighter intensities where the sky
is reflected, darker intensities
where the water is in shadow,
and reflections of ground cover
that are close and far away.
changes in illumination along the
body of water and are not clearly
distinguished from the
surrounding terrain.
Image from the DARPA 2004 Grand Challenge media galle
Image from Matthies, BelluttaImage from Matthies, Bellutta
Big Challenge on Water Detection
to formulate a single feature capable of
characterizing water under the different
scenarios

Water Detection Method
Color Imagery
Short Wave Infra red (SWIR) Imagery
Thermal Infra red Imagery
Laser Range Finder
Multi feature based
Polarization Imaging based

Color Imagery
Color classification for rippled water
Result :
Color classification for still water
Image Processing : Saturasi (left),
brighness (right)
Source :
Evaluate the brightness
[I = (R+G+B)/3] and color
saturation [1- min(R,G,B)/I]
of the terrain, water, and
sky
If [S=0] or
[S ≤ 0.27 and B ≥ 0.73] or
[sky and S ≤ 0.1 and B >
Bmin(S)] or
[sky and S ≤ 0.3 and B >
Bmin(S) and 240<H<285]
The thresholding
criteria derived for
labeling a pixel a
water cue from color
S = Saturation, B=Brightness, H=Hue

Color Imagery
During the day, colour image classification can be
used to recognize water by its reflection of the sky; in
off-road, open terrain, this is a fairly reliable, easy-to-
compute signature. On the other hand, when still water
reflects other aspects of the terrain, such as trees,
hills, or buildings, it may be particularly difficult.
This method usually combine with texture detection of
water. texture quantifies grayscale intensity differences
(contrast), a defined area over which differences occur,
and directionality, or lack of it (Haralick, 1973). For this
water cue, water regions having low texture.

Short Wave Infra Red
(SWIR) Imagery
Basic Principal : Analysing image of it deepness Infra red Sensor
As the absorption coefficient of
pure water in the near infrared
wavelengths is higher than that
in the visible range, water
bodies of any appreciable
depth will appear very dark in
the near infrared imagery.
Ability of SWIR cameras
with wavelength sensitivity
from nearly 0.9 to 1.7 μm

Short Wave Infra Red
(SWIR) Imagery
Fact from SWIR : Snow and ice have very strong absorption beyond about 1.4
μm therefore the wavelength region around 1.5 μm to 1.6 μm may be useful
for recognizing water, snow and ice.
Result : The water is very dark
at the bottom of the image
where it reflects the sky; the
angle of incidence at that point
is at least 80 degrees.
Beyond that, strong reflections
of clouds and trees are evident
on the water surface. Note that
the vegetation all around the
reservoir is highly reflective at
these wavelengths.
SWIR is capable of detecting water at
moderate angles of incidence and has the
potential to discriminate snow and ice
from other terrain materials. But the
performance of this technique degrades in
regions of water strongly reflecting
vegetation or the clouds.

Water Appearance from operating
time detection (day or night)
The water bodies tend to be cooler
than surrounding terrain during the
day and warmer at night.
The size of water body will be
affected to the temperature
distribution and smaller bodies
equalize temperature quickly while
there is significant contrast for
larger bodies of water.
Moreover, water has a higher
emissivity than other terrain
materials and hence, improves the
contrast during night.
Ability of Thermal Camera
(Mid-wave IR - MWIR) with
wavelength sensitivity from
nearly 3.5 to 20 μm
Water bodies have a thermal region
distribution region around 3 to 5 μm
and other gases in the atmosphere
restricts aerial systems 8 to 15 μm.

We can use thermal images of water
measures only on the very top layer
of the water surface because those
wavelengths are attenuated/absorbed
very rapidly.
In the water content, this is will
makes confusing results sometimes,
unless you know for certain what
covers the area you are looking at or
have very precise control of the
wavelengths sensed by the
instrument (which makes in
expensive).
Most thermal imaging systems have
strict technical parameters, for
example, detector materials must be
kept extremely cold during use
(because the emitted radiation being
sensed is very weak)
Result Water Detection with Thermal IR taken at 3 pm (left) & 4 am (right)

Laser Range Finder
A LADAR system works by firing photons
(lasers) at a given object or area, it then
measures the time of flight of photons (time
to reflect back), this enables a CCD style
setup to encode each photon as a pixel and
produce a 3D image of the object or area.
The water bodies can return signals for
robot mounted LADAR ranging systems
owing to the specular reflection at the air
water interface.
It has been suggested that some of the
LADAR energy penetrates the interface
and could even produce a range
measurement of the bottom of the water
body depending on the angle of incidence
LADAR System

Laser Range Finder
LADAR has been concluded for detecting water body within short
ranges of shallow water bodies. The visible and near infrared
LADAR may provide a return value and could be used in detection
and depth measurements.
Scene elements within certain range (between 6.5 and 11
metres), provide return values could be used to characterise
water. Beyond this limit, no return signal is obtained from any
object and hence, it is mean not possible to detect water hazards.
For detecting the presence of water body, LADAR
system need careful installation, depend on time
operation (day or night), and the precision angles of
incident laser beam to water bodies object.

Multi feature based
A lot of scene of water body in outdoor
environment, bring the idea to researchers
to combine various techniques to detect
the water.
The technique employs brightness,
texture and range reflection features
that are fused to detect water
hazards. [9]
The brightness and texture cues
Extracted by common image
segmentation methods such as region
growing and clustering.
Find a parameter the 3D Information
Use an NCC (normalized cross
correlation criteria) based stereo
matching method.
Features are fused
to detect the water regions in the scene
Brightness textures

Multi feature based
The multi-cue approach allows each detector to
target different water characteristics.
A certain amount of false detections from each
detector is tolerated by applying fusion rules that
are, in part, designed to eliminate false
detections.
Generally, this technique have a better results
than any technique based on a single feature.
But, to apply this approach we need found many
task to formulating a robust fusing scheme
capable of accommodating the different factors
influencing the appearance of a water body.
This is will increase a computational overhead,
hence it will complicate their use in real time
applications.

This technique based on the physical principle
that the light reflected from water surface is partial
linearly polarized and the polarization phases of
them are more similar than those from the scenes
around.
Water hazards can be detected by comparison of
polarization degree and similarity of the
polarization phases.
Two step :
I0, I45 and I90 is a
representation of
intensity image
measurements that
are taken at 0, 45
and 90 degrees of
polarizer lens.
1. Extraction of
polarization
imaging
information

Extraction of polarization imaging information result :
(Binxie, 2007) Degree of Polarization
Phase of Polarization
2. Segmentation polarization image
Degree of Polarization Phase of Polarization
self-adaptive threshold
segmentation algorithm
and a morphology
filtering technique to
label water region

Classification Scheme
of method water detection

Presentation ICTS 2009

More Related Content

What's hot (20)

Similar to Presentation ICTS 2009 (20)

Recently uploaded (20)

Presentation ICTS 2009