Research On Method Of Intrusion Detection Based On Stereo Vision System

With the rapid development of China’s high-speed railway system, the scale of railway network is increasing, as a result, the security of the operational railway lines should be given more attention. However, the current railway clearance intrusion detection technology lags behind the practical application requirements, thus, railway accidents occurred frequently, causing tragic casualties and economic losses. This thesis presents a method for detecting foreign matter intrusion based on stereo vision system, it overcomes the problems of current clearance intrusion detection technology, thus improving the accuracy of the intrusion detection.Firstly, the overall scheme design of our intrusion detection system according to the specific application is introduced in this paper, and then the principles and techniques of stereo vision, next are the system hardware architecture and software design, and finally we give the experimental results and analysis to test the performance and accuracy of this system.This thesis focus on the research in camera calibration, image correction, image matching and3D reconstruction techniques. The MATLAB calibration toolbox is used to calibrate the binocular camera, OpenCV is used in the image correction with the parameters got from camera calibration, the image matching is completed with the SAD matching method, and the3D reconstruction is based on the cross binocular stereo vision model, in this way, we overcome the incorrect problem in the parallel binocular stereo vision model. In order to solve the slow speed of computing, CUDA(Compute Unified Device Architecture) is used to optimize the design of the program.Experimental results show this system can detect the intrusion bigger than the size of0.3m X0.3m X0.3m accurately and timely. The result shows, our system can judge the existence of intrusion accurately as expected. The method of stereo vision is used to extract3D information from the binocular images. Compared with the traditional method based on image difference, the intrusion clearance detection based on3D information effectively overcome the problem of ignoring the static foreign matter, mistakenly identifying the shadows, can’t detect the location of the foreign matter etc.