Study Of Track Obstacle Detection System Based On Machine Vision

In recent years, along with the mass speedup of train and the change of train operating mode, to meet the conditions of safe operation of trains, the situation how to freed the train driver from the heavy outlook job brought by single driver mode, becomes an urgent problem. Research of Obstacle detection system based on machine vision is not only an emerging application direction, but also one of the cutting-edge topics in recent years. This paper takes obstacle detection as research background, proposes a rail obstacle detection system combined with the actual situation in orbit. This method can help the train driver judge the existence of track obstacles in front of train, and ensure the safety of railway operation. The main contents are as follows:(1) Design of the vision system. Combined with the train feature of high speed and long braking distance, a vision system is proposed as the first design in this paper. The vision system of class straight(include straight track and small angle turning) utilizes car camera with ultra-long focal length in the locomotive. And the vision system of corners(include large angle turning, right-angle track) utilizes high-definition surveillance camera with fixed design. Wherein the obstacle detection section of class straight, automatic detection and identification of track obstacles can be realized by means of video image processing, detection and recognition,This is also the paper's focuses.(2) Image enhancement. As a result of using car camera, the train jitter can affect the quality of collected images directly. In order to solve this problem and realize image enhancement, the homomorphic filtering method is adopt in this paper, first of all, the image would be transformed the spatial domain into the frequency domain by the Fourier transform. Then establish of a low-pass Gaussian filter as image smoothing filter. Finally, the enhanced image can be obtained through the Inverse Fourier Transform, and the image blur and noise caused by train jitter would be eliminated ultimately.(3) Creation of detection window. Because of the complex and varied background during train running can affect the detection process, and only the obstacles on the track before the train can cause run threat. Against this problem, how to locate the track is the key. This paper proposes an edge extraction method based on top-hat transformation and Otsu threshold. Firstly, prominent the rail section through top-hat transformation. Secondly, obtain the optimal threshold T by using Otsu method. Finally, obtain a binary image with clear rail edge line through the connected component labeling, realizing edge detection of rail. On this basis, establish a detection window by linear fitting technology, and this method can shorten the detection time by narrowing the detection range,and improves the operating efficiency of the obstacle detection system.(4) The detection and recognition of track obstacle. This paper proposes a fusion differential method based on mathematical morphology. Firstly, using mathematical morphology improve the effect received from frame difference moving target detection. Then fusing with effect by background difference method. Finally, realizes the track obstacle detection. In the part of obstacle classification and recognition, the Harries angular point matching algorithm was used to realize the obstacle tracking velocimetry. Take comprehensive analysis of size, direction and speed, finally achieved recognition and classification of track obstacles.This paper do a depth research of the obstacle detection system by establishing the system framework, algorithms research and system simulation. The simulation results show that the system proposed run faster with higher accuracy, so this system has some practical value.