Research Of Vehicles Self-localization Method Based On Custom Vision Landmarks

Localization system is important equipment of vehicles、aircraft and spacecraft, wherein the vehicle localization system is widely used. It can improve high accuracy positon information. It has become the automotive industry technology developmenthot spots, which playing an increasingly important role in people’s daily lives, but also widely used in military, transportation and geological exploration and other fields.In a number of vehicle positioning method vision-based vehicle positioning algorithm is not affected by weather conditions, not to the outside world radiation energy, good concealment and large amount of information. With the development of computer vision and visual sensor in the price and quantity of information advantages, vision based vehicle localization system is attracting more and more attention from researchers. Therefore, the content of this thesis is based on computer vision vehicle positioning technology and propose a vehicle positon method based custom visual landmarks. The main research contents and results are as follows:Firstly, this thesis use SURF algorithm for image feature extraction to get visual landmarks. Redundancy and duplication of existing information for the landmarks, this thesis proposes three improvement programs. The first is a visual landmark selection strategy based on the corner measure. The second selection strategy is based on the edge of the visual landmark information. The last one is based on geometric distribution of visual landmark selection strategy. This paper realizes the visual landmarks featured selection and matching precision.Secondly, this thesis research fundamental matrix estimation algorithms. Estimating the fundamental matrix based on visual vehicle positioning technology is the core problem, its accuracy on the vision based vehicle positioning technology has a great influence on. Therefore, this thesis presents a cunstom visual landmarks improved RANSAC algorithm for fundamental matrix estimates. The algorithm first uses the SURF image feature extraction to get matching points, and the matching point based on corner points and geometric distribution method for selection of optional, be used for estimating the fundamental matrix. Through the RANSAC to remove the false matching points, to use the normalized eight point algorithm for estimating fundamental matrix, finally using nonlinear iterative approach to optimize the results.The experimental results show that the new algorithm has high precision and small error.Thirdly, traditional camera calibration methods need to reference the requirement of the scene or require camera to do the special sports. Based on this thesis algorithm, we using genetic algorithms and camera stratified self-caibration to get internal parameters of camera. Then get outside parameters of the camera by normalization and decomposition of fundamental matrix. Then it calculates the vehicle’s pose. The experimental results show that the algorithm has high positioning accuracy and robust.Finally,based on the design of the localization algoritm, to achieve a vehicle in the campus building environment vehicles localization. Experimental results show that the algorithm can quickly get the vehicle location information and the feasibility and reliability of the algorithm.