| In recent years, with the advance of China’s High-Speed Rail strategy, the technology of HSR in our country made a great process that attracted worldwide attention. It also puts forward higher requirements for safe operation of HSR. The pantograph-catenary system, which is the power source of trains, is the main factor that restricts the safety and acceleration of trains. As the increase of train speed would lead to high failure rate of pantograph-catenary system, intelligent monitoring the pantograph-catenary system in real time is an important research. Accurately identify the train number, can locate the faults of pantograph catenary system, and can eliminate the safety hidden trouble of the train in time. Therefore, how to recognize the train number and extract the picture of pantograph area efficiently and accurately is the basis of pantograph monitoring system.Automatic recognition of train number and pantograph monitoring, was implemented in the paper using computer image processing technology. The images of CHR trains as a data source in experiments were sampled by pantograph monitoring devices.Before train number recognition and pantograph monitoring, the images with uneven illumination and low resolution should be enhanced by using MSR image enhancement algorithm. Due to the reasons for the shooting angle, the train’s roof image is tilted, in this paper, edge detection and line detection are used to tilt correction.In the part of train number detection and recognition, edge detection algorithm and Hough transform were used to divide the whole picture into regions. MSER algorithm was adopted to extract the picture of train number area, of which perspective transformation was implemented by using vanishing point detection. Character segmentation method based on profile detection was proposed to divide picture of train number area. SVM classifier was trained to recognize characters using the divided picture.In the part of pantograph detection, this paper proposed an algorithm based on bag of features to detect the pantograph. First, correcting the roof image tilt, then using Sobel filter and linear detection method to extract the initial extraction of the pantograph region. Then extracted Bag-of-Feature of the pantograph region, in the end the feature is used to train a cascade classifier to detect the pantograph.Substantial images were used to test the methods proposed in this paper. The experiment results revealed that the proposed methods by this paper performed high accuracy rate and good adaptability in CHR trains’number recognition and pantograph detection, which means the methods have certain application value. |
