Research On Comprehensive Detection And Monitoring System Of High Speed Railway Pantograph-catenary Based On Image Processing

In the hot era of artificial intelligence(AI),a large number of AI applications develop rapidly,continually impacting on traditional industries' s development.As an important basic support for the development of national economy,the railway of China plays a crucial role in development of China' economy.Meanwhile,as one of the busiest on-land transportation system,it's high running speed and complex running environment can cause serious accidents and pose a serious threat to passenger's life and property.As a result,in addition to stricter and safer control of railway infrastructure and technical specifications,it is exceedingly import to impose intelligent dectection and monitoring on railway system.However,the intelligent detection and monitoring system is under developing at present,the technical indicators are not comprehensive,and more importantly,a large amount of image data have not been fully exploited,and so the application of image processing technology needs to be improved.All countries in the world are alse actively exploring,in order to provide more intelligent detection and monitoring while continues to raise speed.As the train' speed continues to increase,the accuracy requirements of the inspections and monitorings continue to grow.While ?China speed? gradually walks away from home,the detection technology is still not mature enough and the detection accuracy is still not high enough that many detections stagger in the stage of manual detection,playing the roles of auxiliary detection as a result of later starts.The intelligent transformation of the railway shall to be the long-term trendency for the railway industry.Therefore,this paper aims at researching on the key technologies of comprehensive detections of pantograph-catenary system and improving the accuracy,which has certain practical significance on the future development.The main work and research results of this paper are as follows:Firstly,we in-depthly analysed the application range of the the parameters to be detection after a deep understanding of the overhead catenary detection technology standards and discovered the main factors of affecting the experimental accuracy,that is the uneven illumination.We decided to use the multi-scale retinal techniques(MSR)to all preprocessing procedure of contrast enhancement by designing comparative experiments to histogram equalization.Experiments show that MSR can not only increase the recognition accuracy by 15%,but also improve the recall rate of ROI location,which is improved by 23%.Secondly,we designed the overall structure of automatic positioning and detection to the carriage numbers of the train based on the scene test recongnition technology.The whole framework could be divided into two main parts.One was a coarse screening process based on Maximal Stable Extremal Region(MSER)and manual-designated rules of filtering.The other was the classification,identification and positioning part which was based on strong seeds and weak seeds.Besides,a certain amount of comparative experiments were carried out on the design of the features.The experimental results show that the accuracy rate is 98.34% by use of HOG,which basically reaches the industry application standard.Thirdly,we designed the framework of key area of position of the pantograph,which was based on the Bag of Features(BOF)and Support Vector Mechine(SVM).The whole positioning procedure could be seprated into three parts.The first was a coarse extraction procedure according to the geometric characteristics of the pantograph key region,which could extract a large member of interest of regions(ROI),and the key ideals were the Canny-edge detection and the Hough line transformation.The second was the construction of visual dictionary based on the Scale Invariance Feature Transformatin(SIFT)and k-means++ clustering algorithm.The last was to complete the automatic positioning of the pantograph skateboard key area precisely by the combination of BOF constructed according visual dictionary and SVM classifier.The 96% recognition rate is realized,which provides a certain technical indicator comparison for the industry,and improves the utilization of the image.