Design Of Intruder Detection System Based On Binocular Stereovision

Rail traffic safety is becoming more and more important with the developments of high-speed railway and urban railway transport. In the process of high-speed railway's normal operation or construction over and paralleling existing railway, objects invading clearance limit would make serious impact on traffic safety, so the design of stereo vision-based intruder detection system is meaningful and critical.The theory of binocular stereo vision simulates the mechanism of human eyes which capture a pair of images at different angles. Based on the disparity of the pair of images after using stereo matching and other image-processing operations, three-dimensional information can be reconstructed by using trigonometry. Firstly, the key technologies of binocular stereo vision, i.e., the camera calibration, stereo matching and 3D reconstruction, are studied. The feature is extracted by Harris algorithm, and the camera is calibrated by using feature points based on BP neural network. The image processing algorithm is studied, and gray correction, median filtering and contrast enhancement are used on original images, the feature is extracted by image binaryzation after subtracting from the leading frame. Then the model of 3D reconstruction is built after getting disparity by stereo matching, thereby the 3D info is obtained.Secondly, the system design is described from the aspects of algorithms, hardware and software respectively. The algorithm includes clearance calibration and real-time detection. Hardware includes binocular camera and PC104, which compose the platform of image grabbing, storage and operation. Software includes host and front parts. The host software is consisted of remote calibration module, manual detection module and communication module. The front software, consisted of clearance calibration, real-time detection and communication module, detects intruders by using mixed programming, multithreading, serial port communication and other techniques.Finally, the system precision and the influences of parameters are analyzed. A modified method is proposed and verified in experiments. The performance of the system in both environments of laboratory and field are also tested. Results show that efficient and accurate intruder detection are achieved.