Obstacle detection is primarily accomplished using three Point Grey Inc. stereo cameras. Each camera captures two simultaneous images from different cameras.

The images are processed by Point Grey software. By comparing the two rectified images, a 3D depth map can be constructed.

Our own software determines which points in the image are obstacles to be avoided. It is based on an algorithm decribed in a paper by Manduchi.

Our software then groups the obstacle points into distinct obstacles, which are tracked in time.