Research On Contenary Steady Arm Recognition Method Based On Video Image

Catenary systems is an important part in traction power supply system of electrified railway. It guarantees the normal operation of the traction power supply system, and it would generate serious consequences if the system was failure. The traditional catenary detection method requires manual inspection, it costs a lot of manpower and material resources and the testing cycle time is time-consuming. In recent years, computer vision technology improved speedily, and target intelligent detection algorithm based on image processing technology was widely developed. The HD camera was installed on high-speed rail inspection system, and we will get massive catenary pictures by shooting catenary components. We need to identify pictures intelligently by using digital image processing technology, because artificial discrimination pictures is time-consuming and huge workload. At present, there is no re-search of catenary steady arm intelligent recognition based on complex background images. Therefore, recognizing the catenary steady arm efficiently is significant for the normal oper-ation of high-speed railway.In this thesis, target intelligent recognition program based on image is used to achieve the steady arm identification. There is no relevant image data. First of all, The image data is building based on massive complex backgrounds, changeable weather images, which can ensure the diversity of the image data. This thesis established steady arm image data which contains the steady arm positive sample image data, negative sample image data, the whole picture test sample image data.Secondly, this thesis studied the classification and feature extraction methods. Combin-ing with AdaBoost algorithm, HOG and LBP feature classifiers were trained based on the image data which was built in this paper and this thesis compared the performance of the classifiers.Finally, in the light of current features existing disadvantage of time-consuming and without scale-invariant, the thesis proposed Gray self-similarity feature based on the struc-tural similarity of steady arm, the feature is scale-invariant and the feature extraction rate is faster than HOG and LBP feature. There is no need to resize image when conducting steady arm multi-scale detection, so it can further improve the detection rate, but the performance is equal to LBP and HOG feature. GSS used AdaBoost algorithm to train feature classifiers and achieved accurate detection of steady arm. Finally, the steady arm was detected by hough line detection.