ALV Study On Stereo Vision Technology

Autonomous Land Vehicle (ALV) is an intelligent mobile robot, which can run on road or cross-country autonomously and continuously. ALV is of great value in research and application. This dissertation aims at the research on stereo vision methods and technologies applied in ALV navigation from the aspects of theoretical and practical.The accurate calibration methods of stereo cameras are studied, the way of erasing the radial distortion of cameras has been discussed and proved in experiment. In various stereo matching algorithms, our research emphasis area-based local matching algorithms. The qualitative and quantitative analyses on the performances of various similarity measurements are made, and the influence of matching window size is also studied. Otherwise, two accelerating techniques( Pyramid Matching and Box Filtering) are introduced.In order to overcome the drawbacks of traditional stereo algorithms for ALV navigation, we propose a direct height-computing stereo algorithm. This algorithm can sharply improve system performance both in effect and speed. We also propose an area re-project algorithm to deal with the texture lack area in stereo images.A height-gradient-based obstacle detection algorithm is proposed to process the height images outputted by stereo vision. The obstacles in scenes can be detected exactly and quickly in this way.The fusion of stereo vision and laser-radar (Ladar) is also studied preliminarily in the background of ALV navigation. By introducing the Ladar data into the height prediction before matching, the fusion can occur on data layer. We have also found and practiced a feature layer fusion method based on Dempster-Shafer evidential reasoning approach.