Forward obstacle detection system by stereo vision

Abstract

We have developed forward obstacle detection system which has good detectability and the accuracy of distance only by using stereo vision. The dense disparity images of a region-based matching can be obtained at real time by using FPGA-based stereo processing system. Road surfaces are detected and the space to drive can be limited. A smoothing filter is also used. Owing to these, the accuracy of distance is improved. This system runs in real time. In the experiments, this system could detect forward obstacles 100 m away, and also it could immediately detect cutting-in objects.