Research On Extraction Technologies Of Environmental Characteristics Based On Parallel Binocular Vision

Extraction technology of environmental characteristics based on binocular vision is an important research element in the robot navigation field. It uses two vision sensors to capture a great deal of environmental information, which other sensors can not compare with. The environmental information can guarantee a reliable decesion for the control of robot navigation.In this paper, the traditional theory of binocular vision is studied. Combining with the requirements in the practical application of mobile robot navigation, the paper carries out a research on the imaging models and calibration of camera parameters. Then the paper designs experiments to achieve the practical requirements. In the later chapters, extraction of the feature points and stereo matching, generation of disparity map, irregular obstacle detection, as well as the calculation of three-dimensional information are all studied. In the same time, experiments have been done to verify the effectiveness of the algorithm in all parts. The principal innovations of the paper are listed below:(1) Camera calibration method raised by Zhengyou Zhang is studied in the first place. Then the paper designs the binocular vision system in line with the practical application and through the design of experiments, the paper calculates the parameters of the cameras. The paper puts forward the problems which should be paid attention to in the practical applications and gives appropriate solutions.(2) The paper improves an image matching method of binocular stereo vision based on feature points. Firstly, the Harris corner detector is improved to detect feature points and the method to describe the features of the points with SIFT(Scale Invariant Feature Transform) is enhanced. Secondly, Euclidian distance is utilized to get exact matching in the points set above. Finally, a simple efficient way to eliminate wrong matches is given. This method is proved to meet the needs of real-time binocular vision by actual test, and it can be applied to practice.(3) The paper describes the research progress in the target detection area of binocular vision, and based on the actual requirements of the project, the paper presents an irregular obstacle detection method based on disparity map. The method considers the hight information and continuous range of the matching points as the primary basis for distinguishing obstacles, excluding the difficulties by using shape and color. In the last part, the paper verifies the validity of the algorithm through experiments.The author expects the paper may have some reference meaning in research on robot navigation based on binocular vision, so that it can make certain contribution to the theory construction and practical application of binocular vision and robot navigation.