Abstract
This work describes the development of a stereo vision system used as a 3D sensor to detect obstacles in outdoor environments. Camera captures frames with high sampling rate which makes the stereo vision capable for 3D scanning of mobile obstacles. We implemented the stereo system with two IEEE1394 cameras using wide angle lenses and casing developed using rapid prototyping. We use a fast stereo algorithm provided as a function in OpenCV to generate highly reliable disparity maps calculated with sub pixel accuracy. From the disparity map we reconstruct a 3D environment, and then filter out the objects according to their height from ground level up to 2 meters in order to detect only the objects that could block the robot's path. The filtered objects are then clustered using a blob extraction technique. From various experiments we confirmed this system to be capable of detecting obstacles. This system uses resolution of 640x480 and is capable of reconstructing the 3D environment and detecting obstacles at a rate of 20fps+.
