Vision and Obstacle Avoidance In Cartesian Space
Why is Obstacle Avoidance Important
• Workspace can change unexpectedly• No prior knowledge of workspace• Multiple robots in workspace• Humans in workspace!
Addressing the Issue
• Vision – Object Identification– Coordinate Transformation
• Control– Trajectory generator avoidance – Impedance controllers
Vision Introduction
• Cameras– Light sensitive chips
• Light – Visible spectrum
• Color– Red– Blue– Green
Image acquisition
• Grayscale
• Bayer
• Binary
Common feature extraction techniques
• Edge Detection– Edge(image,method)
• Sobel• Prewitt• canny
• Corner detection– Corner(image)– SIFT– SURF
• Color schemed detection– Achieved through logic
Color schemed detection demo
• Now that we have the pixel value from our image lets find the Cartesian coordinate of this object.
Camera frame
is our target in the camera perspective in the Cartesian.
Finding is not easy.
Problems with depth
Scaling the x and y coordinate by z we can make an image point Let
M
Where
=principal length
𝑃❑𝐶
Pixel point
Principal point
Pixel Origin
𝑣 𝑢
Transforming from 3d to 2d where the mapping is not one-to-one, i.e. unique inverse does not exist because of the depth
Camera Calibration
are intrinsic camera properties that can only be determined through camera calibration, since each camera has different properties.
is an extrinsic property that will change depending on the orientation of the camera.
Best resource for camera calibration is http://www.vision.caltech.edu/bouguetj/calib_doc/Once the calibration is done the intrinsic properties don’t need to be calculated again.
Obstacle avoidanceVision
Vision based controller
Haptic Geometries
Using basic geometries find the optimal path around the object and back to the normal trajectory.
Vision Impedance Controller
Vision controls the and in
Effect changes entire system response and can easily make system unstable. Takes extensive
knowledge of controller and system response.
An alternative approach
A
Vision
Model Loop
Control Loop
Questions?