Research On The Algorithms Of Moving Vehicle 3D Reconstruction Under Surveillance Scene

Three-dimensional(3D) reconstruction is an important branch of computer vision. It uses multi-discipline knowledge and directly imitates the human visual system by way of perception of the objective world.. It is a comprehensive and cross research area and has a very wide application prospects in the fields of aerial mapping, visual navigation, medical diagnosis, electronic commerce and virtual reality, etc. With the application of these aspects,it also promotes the development of 3d reconstruction.This paper focuses on three-dimensional reconstruction of the camera calibration technique, match strategy and matching algorithms, reconstruction and other key issues and difficulties in study. The process of obtaining a three-dimensional spatial information from two-dimensional images, the camera calibration is indispensable link. Based on the comparative analysis of the foundations of several methods of camera calibration, a new camera calibration method based on road markings is introduced. The calibration template of this calibration method is the marking on the road, this kind of template can be seen everywhere in the street. So using this calibration template in the surveillance video is relatively applicable. In order to obtain a dense reconstruction effect, a method based on regional growth is proposed. To improve the reconstruction precision, design and implement a dense matching method based on region growing,which solved the problem of lacking features, reconstruction localization, and even distortion and other problem. The method exploits the method of extracting features, find the robust seeds in the image first; then take the seeds as starting point, through a certain growth method to spread the matching relations to the whole image; finally, using RANSAC role in the matches, which to obtain the final matches. In searching for matching points, the most important constraint is geometric constraints. This constraint limits the matching point locating on the pole line, which improves the efficiency of search elements.Finally, this paper gives the corresponding results, By contrast with the conventional method, Experimental results show that the matching point obtained by this method are uniformity both in quantity and distribution. This method improves the accuracy of 3D reconstruction and has a higher efficiency of the algorithm.