Research And Application Of Common Anomaly Detection Technology In Video Monitoring System

With the wide application of video monitoring technology in many fields and the increasing number of surveillance cameras,how to maintain and manage them efficiently becomes an urgent problem.Moreover,it is difficult to monitor some soft faults such as camera occlusion,position deviation,focus blur,color distortion,black screen and so on.In the past,these video image quality problems need manual troubleshooting,but this not only greatly increases the workload of inspectors,but also the accuracy of detection completely depends on the inspection strength of inspectors,so video monitoring image is different Automatic and intelligent detection is particularly important.In view of the above problems,this paper focuses on the four common problems of video image black screen,gray level anomaly,camera occlusion and offset,and proposes the corresponding detection algorithm.To solve the problem of black screen in surveillance video image,a detection algorithm of black screen is proposed by analyzing the gray histogram of black screen and normal image.The black pixels are identified by setting the threshold value of pixels,and the black pixel rate of the image is calculated to determine whether it is a black screen.The experimental results show that the method can effectively detect the problem of black screen in surveillance video image.Aiming at the problem of camera position offset,a method of camera position offset detection is proposed.Firstly,the edge eigenvalues of the normal image background area and the same area of the image to be tested are extracted,and then the histogram similarity is calculated to determine whether the image to be tested is offset by the similarity degree.In view of the problem of camera occlusion and position deviation,this paper analyzes the shortcomings of traditional image algorithm which adopts the method of manual feature extraction for camera occlusion and offset detection,while the advantage of convolution neural network in the field of machine vision is the automatic feature extraction of image.The convolution neural network algorithm is used to automatically extract image features and solve the disadvantage of manual feature extraction.A method based on depth learning and feature similarity is proposed to solve the problem of camera occlusion and offset anomaly detection in video monitoring.The validity and adaptability of the algorithm are verified byusing real scene and virtual scene sample data.At the end of this paper,based on the above image anomaly detection algorithm,a monitoring video image quality anomaly diagnosis system is designed and implemented.