Research On Identification And Detection Of Birds' Nest And Insulator Based On Machine Vision

With the rapid development of electrified railway,the railway contact network pl ays an important role in ensuring the normal operation of catenary.Analysis and detection of the common fault and abnormal operation state of the contact network,the development of a series of preventive measures,is an important guarantee for the safe o peration of catenary.At present,the main way to detect the contact network is to analyze the vehicle video,through careful observation,to find out the fault contact network of existing or potential threat to the normal operation of the contact net,thi s detection method takes a long time,affected by human factors.In this paper,based on the automation and intellectualization of railway catenary detection,the method of computer vision is studied.This paper focuses on the following aspects of work:Combined with the features of platform of bird nesting,a novel foreign body detection method was proposed based on relative position invariance between hard beam and birds' nest.The proposed method aimed to overcome the difficulties in recognizing and locating the birds' nest in the various global images of the high speed railway contact net.Firstly,the image edge is obtained by using the horizontal Sobel edge detection operator.Secondly,the image is corrected by the method of probability Hough transform line detection,and get the most front hard beam combined with the length of lines in the image to be analyzed.Thirdly,the image is processed by Otsu and corrected with the same angle.Through the statistics data of white area between the two beams,whether any birds' nest existed were able to be distinguished.The experiments show that the method can exactly recognize and locate the birds' nest between the beams.The algorithms has the characteristics of good accuracy,high automatic recognition rates and efficiency,can be provided for the safety inspection of railway contact network.In the process of insulator recognition,considering the unique shape and texture characteristics of the insulator in the contact network,the method of machine learning is adopted,and the problem of insulator identification is transformed into the classification problem.Firstly,the target and other background images are cut out from a large number of catenary images by hand,set up a database of positive and negative sample s.Secondly,after the image preprocessing,the Hog features are extracted,and these features are input to the classifier for learning and training,and then the trained classifier is used to identify the target on the test image.The traditional target recognition algorithm detect the image in the form of sliding window after obtaining the classifier,this searching method aimlessly,produces more windows and causes high computational complexity aiming at the defects,this paper proposes a new method for insulator target recognition based on selective search.Firstly,the target is segmented from the test image by selective search algorithm,and then recognize it with the trained classifier.This method gets rid of the brute force search of the traditional sliding window,which greatly saves the computation time.Experiments show that the new algorithm has a good separation effect and fast separation speed.