Three-dimensional Visual Odometer Critical Technology And Applied Research

Vision based navigation technology has been applied more and more in the navigation and localization of autonomous mobile platform, with the maturity of computer vision algorithm. This thesis aims to study the stereo visual odometry in vision based navigation system, visual odometry from stereo sequences has been described as an important assistant of conventional odometer to obtain more accurate result of navigation and localization.Firstly, this thesis describes the basis principle and methods of computer vision, and then discusses the concept and the main framework of the visual odometry, and introduces the research at home and abroad in recent years.In Chapter II, the basic concepts and methods of stereo vision and its application in the navigation and localization are introduced. This chapter mainly focuses on the basic principle of stereo vision system, stereo matching and 3D reconstruction.Chapter III studies visual odometry system, describes the main framework of the stereo visual odometry, and then feature detection and matching has been talked about. This chapter focuses on the Harris corner detection algorithm used in visual odometry. Compared with SIFT(Scale Invariant Feature Transform) feature detection and matching algorithm, Harris corner detection algorithm has bad performance at invariance properties, while it is fast and easy to realize, on the epipolar geometry constraint, the sufficient features can be matched precisely when the change between two images is small, Harris corner detection algorithm is suitable for the application when good real-time performance is required. This chapter also studies on the motion estimation which mainly affects the accuracy of the result. Combination of the linear and nonlinear least square method (LSM) are proposed to ensure the real-time performance and accuracy of the algorithm.Chapter IV introduces the application of visual odometry in wide baseline stereo matching. Wide baseline stereo matching mainly solves the problem that normal stereo vision system is difficult to reconstruct the objects far from the camera. The main difficulty is how to rectify the stereo images precisely when motion is uncertain. Visual odometry can get accurate motion information, so it can be used to rectify the stereo images in wide baseline stereo matching system.Finally the summarization of the thesis and the future work are presented.