Research And Design Of Key Area Security System Based On Target Detection

With the rapid development of computer vision related theories,intelligent security systems are gradually replacing traditional security systems.In the security of important places,the sudden invasion of UAVs and the abnormal behavior of pedestrians will pose a security threat to important places.The traditional security system requires relevant security personnel to monitor and observe.After a long time of work,the security personnel will not be able to detect abnormal events in time due to their inattention,which makes the detection of abnormal events not timely.The intelligent security system can not only reduce the consumption of manpower,but also detect abnormal events in a timely manner.This thesis mainly focuses on the pedestrian abnormal behavior detection and UAV intrusion detection in the security of key areas.The main research contents are as follows:(1)In the abnormal behavior detection based on pedestrian detection,in order to further improve the accuracy of pedestrian detection,a YOLOv4-tiny pedestrian detection algorithm SK-SN-YOLOv4-tiny is proposed,which combines adaptive selection convolution kernel and flexible non maximum suppression.First,an adaptive convolutional attention mechanism(SKNet)is added between the backbone network and the feature fusion layer(FPN)to improve the detection accuracy of the network in the multi-scale case.Secondly,the improved flexible non maximum suppression(DIOU Soft NMS)is used to improve the detection accuracy of pedestrians in dense scenes.Finally,after the detection of pedestrians,according to the position of pedestrians in the image and the movement track of pedestrians,it detects whether pedestrians have abnormal behaviors such as wall climbing,regional intrusion and wandering.(2)To solve the problem of low detection accuracy of small target UAV,this thesis proposes a CS-MF-CA-YOLOX UAV detection algorithm.First,the pyramid pooling module(SPP)is integrated with the structure of cross stage feature fusion to retain the details and semantic information of small target UAV as much as possible.Secondly,a small-scale convolution operation is added to the backbone network to fully extract the semantic information of small target UAV.Finally,coordinate attention mechanism is added to the residual block of the backbone network to improve the antibackground interference ability of the network and further improve the detection accuracy of the UAV.(3)This thesis studies and designs an intelligent security system based on target detection.Combined with the business requirements of intelligent security,the system architecture and functional modules are analyzed and designed,and the abnormal behavior detection based on pedestrian detection and low altitude UAV intrusion detection algorithm are applied to the system.It well meets the business needs of security in key areas.