| Binocular stereo vision technique, which directly simulate man's eyes processing information manners, it can be widely used in many domains. For example, obstacle detection, virtual reality, industry detection, navigation of robots, aviation and spaceflight, etc. The underlying obstacle information will be supplied for drivers or non-driver's car by picking the information from binocular images in this paper. And the obstacle detection in the intelligent driving is the background of application, the images from binocular sensors as the studying object. This research is significative for intelligent driving and navigation of robots.Based on Marr's theory, binocular stereo vision's basic theory, method and related techniques have been researched in this paper. This article mainly aims at automatically obstacle detection in this research area. And one kind of technology which has been called the V-disparity algorithm is introduced. It can detect the obstacles which has the surface characteristic, for instance, pedestrian, vehicle, wall and so on. Moreover, this algorithm is insensitive to illumination, shadow and so on. It is suitable to detect obstacles under the complex background of the city.This article mainly includes three partial contents:(1) Stereo image rectification. It is hard for stereo image matching, because the stereo image can't satisfy the epipolar lines condition. This article introduces a rectification method without calibration.(2) Stereo matching, that is disparity image generating. We search the homologues from the binocular images fast and accurately by Stereo matching. And get high-quality stereo disparity of the images. It then will be the basis of the obstacle detection.(3) V-disparity image generating. Building the V-disparity according to the disparity image, extracting the straight line information in the V-disparity, and through these straight line information can detect the obstacles. |
PDF Full Text Request