Research On Application Of Binocular Vision In Obstacle Detection

Image proceeding is very important in intelligence traffic system, because images include mach useful information. This paper introduced a method of obstacle recognition and location based on binocular stereo vision, and designed an obstacle detection system. The system includes two CCD cameras, a computer and a video capture card. Firstly, the CCD cameras on different location of vehicle capture the images of obstacle. Then, the computer process and analyze these images, and compute the obstacle location. This process includes image acquisition, camera calibration, image division, stereo matching and 3D position computation.The stereo matching algorithm is one of mostly content in the research. There are two stereo matching methods, area-based matching method and feature-based matching method. The former needs mass of calculation, while the latter depend on exist the same matching features in different images. This paper put forward an improved method of stereo matching. In the method, the horizontal information is more important than the vertical information, so the vertical data in images were neglected. For insuring the real-time and precision, the horizontal-projection of gray area was matched.This stereo vision system showed good performance on obstacle detection experiments. The camera calibration and stereo matching algorithm can be applied to other stereo vision system. This thesis can be the foundation for future related work.