Intrusion Detection, Intelligent Video-based Railway

Trespassers on railway's right-of way, include rolling stones, people, animals, etc, lead to railway traffic accidents frequently which pose a great threat to peoples' live, properties and countries' transportation. Trespass detection has become the focus of international transportation safety. With the rapid development of intelligent video surveillance, the video-based trespass monitoring system is studied widely and applied in railroad safety protection as an attempt.The intelligent video-based trespass detection system is discussed for its function target and application structure, which suggest the properties and installation position of test image acquisition instruments. After studying and analyzing types and character of trespassers, a railway trespass detection algorithm is key researched and designed for both accumulation and moving object.The whole trespass detection algorithm consists of three parts:image preprocess, rails recognition and template matching, moving object trespass detection. Image preprocess is designed to correct the image when distorted by lens, filter the image when polluted by noise, stabilize image when dithering, enhance the image when having low contrast. In the part of rails recognition and template matching, the ROI (region of interesting) is determined by the rails which are recognized from the monitor image automatically and the accumulation trespassers are detected. With the part of moving object trespass detection, the trespass in the monitor image is detected effectively if the detected and tracked objects move into the ROI and become trespassers. Every part of the algorithm tested with videos photographed in varies conditions yields to very good results in detecting trespassers.The innovations of the thesis are as follows:Large quantity of snow, sand or other accumulation on the railways is considered firstly as trespassers which detected effectively by technology of orientation field feature template matching. Orientation field which used to identify fingerprint, is calculated according to the feature of rails, filtered by directional consistency with so-called direction-distance, and does good job to recognize the rails automatically. The trespass detection algorithm has been designed for color video monitor by days as well as infrared images taken during the night.