Obstacle Detection System Of Railway Crossing Based On Binocular Measurement And Image Processing

China is a big railway country,131,000 kilometers of railway covering 57% of China's region,is the backbone of China's integrated transport system.The crossing,as the intersection of railway and highway,needs to ensure that there are no obstacles on the crossing that can affect the safety of driving,otherwise it is very easy to have traffic problems or even serious accidents.And the monitoring mode of the crossing across the country is still the traditional way of manual monitoring of this time-consuming and laborious.Therefore,in order to solve the problem of intelligent monitoring and reconstruction of railway crossing obstacles,this dissertation studies a railway crossing obstacle detection system based on binocular measurement and image processing,and can realize the intelligent transformation of crossing obstacle detection by using mobile detection algorithm and binocular vision algorithm to detect and monitor the obstacle of railway crossing.Based on binocular machine vision and image processing algorithm,this system studies the key technologies of large field of view binocular calibration,moving frame front and rear view extraction,SIFT feature point extraction and binocular obstacle size detection with the help of OpenCV Machine Vision Library and VC++ development platform.Finally,the key technology research and software system development of railway crossing obstacle detection are realized,and the experiment is carried out under the condition of laboratory and actual crossing,and there are interference factors such as shade,water and so on in the actual crossing.In this dissertation,the Homomorphic filtering algorithm and the legacy algorithm are also studied to further improve the detection accuracy of the system to the uneven illumination and the small obstacles.Because of the huge computation of the algorithm,it will take a lot of time and system resources,and the low image quality of the camera will further aggravate the burden of the computer.In order to improve the efficiency of the system,this dissertation uses multiple exposure image acquisition,camera frame synchronization,data structure conversion and other means to ensure that the camera output high-quality images at the same time also uses the extraction of areas of interest and filter feature points to reduce the calculation of the algorithm.The experimental results show that the algorithm detects the error of 0.5cm in the 3m range in the laboratory,and the detection error in the 10m~15m range is only 3cm~7cm in the actual scene,and the common obstacles such as beverage bottles,bricks and pedestrians have been successfully detected,and the intelligent detection of the obstacles in the railway crossing has been basically realized.Has the strong practical significance and the promotion prospect.