Matching And3D Space Points Reconstruction Based On Binocular Stereo Vision

Abstract:With the rapid development of China’s railway transportation and the substantial improvement of train speed, the railway security problem is becoming more and more prominent. The intrusion detection is one of the main means to prevent railway accidents, and3D space points reconstruction system based on binocular stereo vision has an important application prospect in the field of target recognition and warning in the railway intrusion detection technology.This paper systematically studies the methods of3D space points reconstruction based on binocular stereo vision, further discusses camera calibration technology, corner detection and image matching algorithm and3D space points reconstruction.Firstly, the Levenberg-Marquardt nonlinear optimization algorithm is brought in to put forward an improved Tsai two-step calibration method based on coplanar. In actual experiments, the error evaluation function shows that the improved algorithm is verified to be simpler at the aspect of calculation, stronger in the real-time performance and is0.183pixel less than tranditional algorithm in absolute error of image coordinate. The error evaluation function of the dual camera system is put forward according to the dual camera calibration method based on binocular stereo vision system, which further optimizes the camera calibration parameters.Then, the advantages of SIFT and Harris corner detection are combined in addition to a deeply analysis of the similarity measure of NPC normalized product correlation. As a result, the image matching algorithm based on the improved SIFT corner detection and NPC similarity measure is proposed. Through the actual test images, it is proved that the matching accuracy and speed of the modified algorithm have been improved significantly.Finally, combined with the proposed camera calibration and image matching algorithm, the3D coordinate of the object space points in railway intrusion scenario is worked out according to the influence of the two camera installation structure to the precision of the binocular stereo vision system and the calculation model of the3D space points reconstruction. At last the three-dimensional information is restored. After3D reconstruction, the railway invasion target has been identified and classified. Experiment results show that the algorithms proposed in this paper have a good effect in the3D space points reconstruction system and then have high engineering application value.