A new edge detector for obstacle detection with a linear stereo vision system

The problem addressed is obstacle detection in front of a moving vehicle. One of the main difficulties encountered when scanning stereo pairs of linear images is the extraction of significant features from the left and right images. The better the features, the more reliable is the matching procedure for 3-D reconstruction of the car environment. In this paper we present a new edge detector which has been developed for this application. Its performances, in terms of detection, localization and multiple responses, are determined according to Canny´s criteria. The system has been evaluated in real traffic conditions. It operates in real-time, due to the highly efficient recursive implementation of the filters proposed in this paper