The Study Of Stereo Visual Odometry For Unmanned Vehicle

The unmanned vehicle is a highly intelligent automation system which consists of environment perception, dynamic decision and planning, behavior control and execution etc. function. The main application fields include unmanned combat, intelligent transportation and driving assistance.To realize the autonomous navigation of unmanned vehicle, three problems have to be solved, namely: Where am I? Where am I going? How should I get there? That is orientation problem, sport goals problem and path planning problem of unmanned vehicle. Among the three problems, the first problem which is to find the position of unmanned car is one of the most basic problems to be solved firstly in all kinds of application.This article mainly studies the positioning problem of unmanned vehicle based on binocular stereovision odometry, which contains theories and algorithm the binocular stereovision odometry uses. The binocular stereovision odometry is a main section of simultaneous localization and mapping(SLAM),it can be used as effectively complement for global positioning system and dead reckoning.Firstly, the article introduces the research background and research situation of unmanned vehicle and visual odometer, compares some kinds of positioning methods for unmanned car and points out the advantage and problems of stereo visual odometry.Secondly, the basic principle binocular stereo vision system of is analyzed, it includes the research contents of stereo vision system, the camera’s imaging model and the relationship between each coordinate system for stereo vision system, the parallax and3d measurement principle of binocular stereo vision, epipolar geometry constraint relation and basis matrix and essence matrix related with internal parameters and external parameters of cameras. The fundamental principles of stereo binocular vision system are the basis of study for binocular stereo vision.Thirdly, the system framework of binocular stereovision odometry is introduced, and a deep study of the key techniques it includes is taken. For feature point extraction, the regular methods Harris and SIFT are studied, based on this two methods, SURF is introduced, and experiment is taken to compare the three methods, and the advantage of SURF method compared with Harris method and SIFT method is pointed out, that is, SURF can extract more stable feature points than Harris method and performs better than SIFT method in real-time aspects; For feature point matching, the nearest vector matching method is taken, and regional restrictions based method is taken to reject wrong matching points.At last, the experiment of linear motion and curve motion for binocular stereovision odometry is taken, the validity of the method this paper taken has proved.