Research On Key Technologies Of Video Processing System On Railway Handcarts

With the rapid development of railway transportation, and the grate increase in the capacity of train for passengers, the rail abrasion becomes a serious problem. In China, the main methods are that workers detect the lines by eyeballing or detect the lines collected from railway handcarts by vision. These methods could not satisfy the requirement of the development of high speed railway, as they require rich experience of workers and much workload but with low efficiency.In this paper, video processing system on railway handcarts for automatic detection lines was studied. After studying on its key technology, automatic detection technology of cement hardening and automatic identification technology of fastener state were put forward, and it provided a new thought for automated detection of rail lines. Compared with the traditional method, the automatic detection system has the characteristic of accurate orientation, low omission and high automatization.In this paper, related works were studied, the necessity and feasibility of automatic detection of state line with video processing technology were discussed, and the main contents and key technology of video processing system on railway handcarts were put forward firstly. Then the main frame of the system was introduced, the design of the image process subsystem was put forward, after studying Image preprocessing algorithm, regional extraction algorithm based on improved Hough transform was put forward, and it has high rate of detection on edge curve and less impact of noise and intermittent curve. At last, automatic detection technology of cement hardening based on image texture analysis was mainly studied, texture feature of cement image distilled by the adaptive algorithm based on gray level co-occurrence matrix was put forward, and automatic detection of cement hardening was achieved by BP neural network as a classifier; Feature extraction of fastener image was studied, and automatic recognition algorithm of fastener state based on moments invariant was put forward. Experiments show that the omission rate of disease is 0, the average accurate rate is 91%, and the system can meet the basic requirement of the actual detection.