A Study On Railway Obstacle Detection Using Machine Vision

Track clearance is the foundation of the railway safety. This thesis is focused on the railway obstacle detection using machine vision, with emphasis on obstacle detection between rails and obstacle detection on the railway catenary. The results show that this technology can provide technical support for the safety of trains.This thesis first introduces the overall design of the railway obstacle detection using machine vision system, including the overall system design, the system configuration and the function of each component. After a brief introduction of the existing fence detection technology, the thesis is focused on obstacle detection between rails and obstacle detection on the railway catenary.In the system of obstacle detection between rails, according to the object characteristics, this thesis introduces two methods of detection:large-size-based detection and small-size-based detection. In large-size-based detection, the system is based on the track location detection algorithm using Kalman filter to judge whether there are obstacles in front of the train. In small-sized-based detection, in order to recognize the obstacle correctly, the system develops the feature abstracting algorithm and use Support Vector Machine (SVM) to classify obstacles and the normal objects.In the system of obstacle detection on the railway catenary, image processing method is used to detect obstacle candidate area. And then, a pulse coupled neural network (PCNN) is used to process the obstacle candidate area and candidate obstacle area can be obtained by the characteristic sequences. Then these characteristic sequences are compared with characteristic sequences in database. According to correlation coefficient, obstacles can be identified.The proposed algorithm is verified in laboratory environment and the railway field environment. The results show that the proposed algorithm has high accuracy, and, to some degree, can improve the safety of the trains.