2. Road Map
Robot Vision
Imaging Sensors
Vision Systems
Visual Servoing
Configuration of Vision System
Image Processing
Gray level histogram
Image Segmentation
Region based segmentation
Image interpretation
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 2
3. Robot Vision
A robot vision system consists of one or more cameras,
special-purpose lighting, software, and a robot or robots.
Vision sensors are used in robot to provide information
about the work area and objects to the robot.
Images of the working area or object are processed using
image processing software to determine position and
orientation of objects in the work cell.
Vision is also used in mobile robots to navigate.
Depending on the application, the camera might be
mounted on the robot or could be in a fixed position within
the cell.
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 3
4. Imaging Sensors
Image sensors convert light into electric charge and process it into
electronic signals
Image Sensors
◦ Charge Coupled Device CCD
◦ All pixels are devoted to light capture
◦ Output is uniform
◦ High image quality
◦ Used in cell phone cameras
◦ Complementary Metal Oxide Semiconductor CMOS
◦ Pixels devoted to light capture are limited
◦ Output is not uniform
◦ High Image quality
◦ Used in professional and industrial cameras
5. Lighting Techniques
The three lighting techniques used in vision
applications are:
◦ Front lighting,
◦ Back lighting
◦ Structured lighting
6. Vision Systems
Vision Systems are of two types namely: (a) Stand alone and (b)
PC based. Standalone systems are
Smart Camera: These are self-contained and do not require
separate computers, there are two types of image sensors used
in smart cameras namely (a) CCD image sensors and (b) CMOS
image sensors.
Vision Sensors: These are integrated devices which do not
require any programming and are systems between smart cams
and vision systems
Digital Cameras are classified based on the type of sensors and
memory storage devices used namely (a) CCD image, (b) CMOS
image, (c) Flash memory, (d) Memory stick (e) Smart Media cards
(f) Removable drives
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 6
7. Image File Formats
Images are stored in a computer in one of the
following formats, depending on the application of
the images stored.
◦ Tagged Image Format [.tif]
◦ Portable Network Graphics [.png]
◦ Joint Photographic Experts Group [.jpeg, .jpg]
◦ Bitmap [.bmp]
◦ Graphics Interchange Format [.gif]
◦ Raster Images [.ras]
◦ Postscript [.ps]
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 7
8. Vision based Robot Control
Vision-based robot control also known as Visual Servoing,
is a technique which uses feedback information extracted
from a vision sensor (visual feedback) to control the
motion of a robot.
There are two fundamental configurations of the robot
end-effector (hand) and the camera:
Eye-in-hand, or end-point closed-loop control, where the camera is
attached to the moving hand and observing the relative position of
the target.
Eye-to-hand, or end-point open-loop control, where the camera is
fixed in the world and observing the target and the motion of the
hand.
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 8
9. Vision allows a robotic system to obtain
geometrical and qualitative information on the
surrounding environment
high level control motion planning
(look-and-move visual grasping)
low level control measures used in the control
loop
Visual servoing control is broadly classified into the
following types and they are based on feedback of visual
measurements
image-based visual servoing
position-based visual servoing
hybrid visual servoing
Visual Servoing
SENSORS FOR ROBOTICS LECTURE
9 DR. HEMA C.R.
9
10. Image based Visual
Servoing
The control law is based on the error between current and
desired features on the image plane, and does not involve
any estimate of the pose of the target.
Image processing is aimed at extracting numerical
information referred to as image feature parameters.
The features may be the coordinates of visual features,
lines or moments of regions.
IBVS has difficulties with motions very large rotations,
which has come to be called camera retreat.
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 10
11. Position based Visual
Servoing
PBVS is a model-based technique (with a single camera).
The pose of the object of interest is estimated with respect to the
camera and then a command is issued to the robot controller, which
in turn controls the robot.
In this case the image features are extracted as well, but are
additionally used to estimate 3D information (pose of the object in
Cartesian space), hence it is servoing in 3D.
Pose estimation methods are based on the measurement of a
certain number of points or correspondences
Numerical pose estimation methods are based on the integration of
the linear mapping between the camera velocity in the operational
space and the time derivative of the feature parameters in the
image plane
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 11
12. Multi-camera systems
information about its depth by evaluating its
distance with respect to the visual system
3D vision or stereo vision
Mono-camera systems
two images of the same object from two different
poses
if only a single image is available, the depth can
be estimated on the basis ofgeometrical
characteristics of the object known in advance.
This is cheaper and easier to calibrate
Visual System Configuration
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 12
13. Visual System Configuration
Eye-to-hand : fixed location
advantage is that the camera field of view does
not change during the execution of the task,
implying that the accuracy of such
measurements is constant
the manipulator occludes, in part or in whole,
the view of the objects
Eye-in-hand : mobile configuration
the camera is placed on the manipulator
high variability in the accuracy of measurements
the accuracy becomes almost constant and is
usually higher than that achievable with eye-to-
hand cameras
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 13
14. Visual System Configuration
Hybrid configuration consisting of one or more
cameras in eye-to-hand configuration, and one
or more cameras in eye-in-hand configuration
ensures a good accuracy throughout the
workspace, while avoiding the problems of
occlusions
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 14
15. Visual information is very rich and varied
complex and computational expensive
transformations before it can be used for
controlling a robotic system
extraction of numerical information from the
image image feature parameters
Two basic operations
segmentation a representation suitable for the
identification of measurable features of the
image
interpretation measurement of the feature
parameters of the image
Image Processing
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 15
16. Image Processing
The source information is contained in a two
dimensional memory array representing the spatial
sample of the image
image function I (x,y) is a vector function whose
components represent the values of one or
more physical quantities related to the pixel in a
sampled and quantized form
light intensity in the wavelengths of red, green
and blue
or in shades of gray (number of gray levels
depends on resolution 256 gray levels)
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 16
17. Gray-level Histogram
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R.
Provides the frequency of occurrence of each gray level in the image
The gray levels are quantized from 0 to 255
The value h(p) of the histogram at a particular gray level p ∈[0, 255]
is the number of image pixels with gray level p
If this value is divided by the total number of pixels, the histogram is
termed
Normalized histogram
17
18. Consists of a grouping process, by which the image is
divided into a certain number of groups, referred to as
segments (component of each group similar with
respect to one or more characteristics)
Distinct objects of the environment
Or homogeneous object parts
Finding connected regions of the image
Grouping sets of pixels sharing common features
into two-dimensional connected areas
High memory usage
Low computational load
Image segmentation
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 18
19. Image segmentation
Detection of boundaries
Identifying the pixels corresponding to object
contours and isolating them from the rest of the
image
The boundary of an object, once extracted, can
be used to define the position and shape of the
object itself
Complementary
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 19
20. Obtaining connected regions by
continuous merging of initially small
groups of adjacent pixels into larger
ones
If the pixels belonging to these regions
satisfy a common property, termed
uniformity predicate
(Verifying gray level)
Binary segmentation or image
binarization by comparing the gray
level of each pixel with a threshold l
The peaks of the histogram are termed
modes (for the dark objects the closest
minimum to the left)
Region-based segmentation
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 20
21. Region-based segmentation
In the presence of multiple objects, a further
elaboration is required to separate the
connected regions corresponding to the single
objects
The gray-scale histogram is noisy and the modes
are difficult to identify
various techniques have been developed to
increase the robustness of binary
segmentation
appropriate filtering of the image before
binarization
algorithms for automatic selection of the
threshold
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 21
22. Boundary-based segmentation techniques usually
obtain a boundary by grouping many single local
edges
Corresponding to local discontinuities of image gray
level
Local edges are sets of pixels where the light intensity
changes abruptly
The algorithms for boundary detection
Derive an intermediate image based on local edges
from the original gray-scale image
Construct short-curve segments by edge linking
Obtain the boundaries by joining these curve
segments through geometric primitives often
known in advance
Boundary-based segmentation
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 22
23. Boundary-based segmentation
Edge detection is essentially a filtering process whereas
boundary detection is a higher level task usually requiring
more sophisticated software
Edge detection can be performed by grouping the
pixels where the magnitude of the gradient is
greater than a threshold
In case of simple and well-defined shapes, boundary
detection becomes straightforward and segmentation
reduces to the sole edge detection
Several edge detection techniques exist, most of them
require the calculation of the gradient or of the laplacian of
function I(XI, YI )
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 23
24. Boundary-based
segmentation
Visual servoing is based on the mapping
between the feature parameters of an object
measured in the image plane of the camera
and the operational space variables defining
the relative pose of the object with respect to
the camera
Often it is sufficient to derive a differential
mapping in terms of velocity (easier to solve
linear, numerical integration algorithms)
SENSORS FOR ROBOTICS LECTURE 9 DR. HEMA C.R. 24