Image Understanding And Polarization Property Research Of Off-road Robots

With the rapid development of robot technology, autonomous off-road navigation robots receive more and more attention from national researchers and experts as its particularity application areas and military value. Scene understanding and obstacle recognition are a major focus in the research of off-road robotics. This paper aim to the detection of two types of obstacles in the off-road environment, one is obstacles like stones, and the other is water hazards like puddles and ponds.We use stereo vision to detect the first class of obstacles. Discussed the theory of stereo vision, method of camera calibration and image rectification. Focused on analysis of stereo matching principle and some challenging points. Compared with the matching results and real-time performance of three kinds of stereo matching algorithms (BM, SGBM, GC) included in the OpenCV library. Then choose the BM algorithm for its good capability both on real-time performance and matching result. Finally to obtain obstacle position by stereo information.We use two different methods to detection the water hazards. First method used water texture and color variation to get water position. By some experiments, we proved that it is workable and has some advantages compare with traditional methods based on luminance information. The second method used polarization characteristics of the environment. Detailed the formation principle of polarized light, analysis of the polarization properties of the wild water. Then shows that polarization information can be used to segment water regions. Finally, by comparing a series of data, analyze the advantages and disadvantages of two different detection methods. And ultimately determine to use polarization information to detect water regions.Depending on the hardware and software platforms of AS-RF robot, we proposed a method which can get polarization and depth information synchronously. Described its principles and the specific hardware structure. Finally use collected data to obtain the location information of obstacles and water hazards. The experiments show that the success rate of this method for obstacles and water hazards detection can reach90%or more, meanwhile, the accuracy meets the requirement of off-road robot navigation. So the joint detection method is feasible and effective.