Intelligent Processing System Of Monitoring Video Based On Machine Vision

With the development of large-scale infrastructure such as residential quarters,public business places,ATM self-service outlets,bank halls and so on in recent years,it is gradually changing from construction to operation.These areas are related to people's life and property security.Sometimes,criminal acts threatening public security,such as carrying dangerous goods,theft,terrorist attacks,occur.At present,security means are added near public areas.With surveillance cameras installed,it is of great significance to develop an intelligent surveillance video processing system applied in the field of security.In order to effectively monitor the risk factors in the monitoring area,this paper studies and designs an intelligent surveillance video processing system applied in the field of PC security video.The system mainly includes two main subsystems: face occlusion detection and legacy detection.In the part of face occlusion detection,through research,the following scheme is designed: Make OwnFace data set suitable for the needs of this paper,use SSD target detection model to detect face,and then use YCrCb color space based skin detection algorithm to detect skin color,calculate the proportion of skin color in the face area to determine whether the face occlusion occurs.Because the SSD model is slow to detect and cannot meet the actual demand,some improvements are needed to improve the detection speed.The improvement methods are as follows: First,replace the original VGG-16 basic network of SSD with a lightweight MobileNet network,and combine them into MobileNet.The SSD target detection model detects the face of the person,and then adopts the BN layer and convolution layer strategy for the trained MobileNet-SSD model to further improve the running speed of the model.The skin color detection method based on YCrCb color space is used to detect the skin color of the face region,and a face occlusion evaluation method based on skin color ratio is proposed to determine whether the face region has occlusion behavior.In the remnant detection part,by analyzing the foreground characteristics of the mixed Gaussian background modeling method(GMM)and the three-frame difference method,a residual detection method based on GMM and three-frame difference method is proposed.Firstly,the shadow obtained by the GMM is eliminated by the shadow elimination method based on the YCrCb color space,and then the foreground obtained by the three-frame difference method is differentiated to obtain a suspicious remnant mass.The dynamic region in the suspicious remnant mass is screened by the centroid method to obtain a temporary stationary target.On this basis,a tracking method based on the mean value of gray scale and histogram similarity is proposed to track the temporary stationary target,and the time t at which the target continues to be stationary is counted,and compared with the set residual judgment threshold T.t>T,the target that is temporarily stationary is considered to be a remnant.In order to solve the problem of loss detection of residuals in practical applications,this paper proposes an improved residual detection method based on fault-tolerant mechanism,which preserves the initial frame features of the detected remnants: gray average and histogram features.In the block area,calculate the gray mean value and the histogram feature,and perform the difference matching and the similarity matching with the gray average value and the histogram feature of the initial frame,so that the residual occlusion factor can still remain after the exclusion factor is removed.The trace of the remnants is proved by experiments that the method can effectively deal with the situation that the residuals are missed after being occluded and improve the stability of the detection.Finally,based on the above research results,combined with the use of the subject,under the Windows,compiled the Caffe depth framework,and using the C++ programming language and Qt interface development,the face occlusion detection system and the remnant detection system were integrated,and developed a Set of monitoring video intelligent processing software for PC.