This paper reports on a stereo vision technique for use in a driver assistance system. In the detection of obstacles using the stereo vision technique, it is required to both reduce the computational load of stereo matching and improve reliability of detection. Furthermore in the process of detection, another challenge is to distinguish the road surface from three-dimensional objects. These problems have been solved using stereo matching with multi-resolution images and the segmentation of disparity in local regions to separate the road surface from three-dimensional objects. Experiments, performed on video sequences, have confirmed that the stereo vision can detect obstacles with low false positive detection and high true positive detection. For the covering abstract see ITRD E134653.
Abstract