Disparity analysis for real time obstacle detection by linear stereovision

A new approach is presented for real-time obstacle detection in front of moving cars by computer vision. The experimental set-up makes use of two line-scan cameras in order to analyse the scene in front of the vehicle. The camera calibration, based on a procedure specifically developed for line-scan stereo vision, allows the system to know precisely where a point of the real world appears in each of the two linear images. Prominent features in the linear images are extracted by a differential operator and matched by means of a correlation measure of similarity.<
>