Research And Implementation Of Real-Time Security Video Monitoring System Based On Human Key-Point Detection

Nowadays,the analysis of urban surveillance video is mainly done manually,which requires a lot of human resources,and it is difficult to quickly identify in the real-time and warn unsafe behaviors that may exist in the city.The intelligent video security analysis system can analyze the unsafe behaviors that may exist in the video automatically and accurately,then providing early warning of dangerous behaviors at critical moments,which can greatly reduce the occurrence of dangerous behaviors in the city and save human resources in the same way.Moreover,It can provide important support for the construction of automatically independent and modern cities.The urban video security analysis system mainly analyzes the behavior of the target person in the video in real time,so as to realize the prediction and early warning of unsafe behavior.For the behavior analysis of the target person in the video,the most important thing is to detect and recognize the target person in the video.The target person recognition method is based on the detection of the key points of the human body is the current mainstream method.The human body key point detection method can obtain the position information of each key point of the target person while detecting the target person in the video,which can provide a basis for further human behavior analysis.Aiming at the limited accuracy and low efficiency of current video character behavior analysis methods,this paper proposes a high-resolution human key point detection network based on feature fusion.Based on this network,three types of video ID tracking,fall detection and cross-border detection are designed.According to these methods,automatically independent city surveillance video real-time analysis system is constructed.The main contributions of the thesis are as follows:(1)A high-resolution network model based on feature fusion is proposed to achieve rapid and accurate key point detection of the target person in the video.First,normalize the video frame image to realize the algorithm compatibility with most surveillance videos on the market;Secondly,by fusing the high-resolution and low-resolution features of the frame image,the high-resolution is constructed based on the deep convolutional network.The overall accuracy of this model for the detection of human key points is 84.2%.Compared with other human key point detection networks,it has a higher accuracy rate.It also has better accuracy for key points such as knee joints and hip joints that are difficult to check.Detection effect;Finally,the high-resolution network model is accelerated based on Tensor RT technology to realize real-time processing of surveillance video.(2)Based on the high-resolution feature fusion network,design video ID tracking,fall detection and cross-border detection methods to achieve real-time monitoring and analysis of the behavior of the target person in the video.First,using the high-resolution feature fusion network to obtain the key points of the target person in the video;secondly,calculating the cosine distance of the key point between the adjacent frames of the video,and assign the same ID to the target less than the threshold to achieve the target person in the video.The accuracy of the algorithm for fall detection is 96.3%,and the recall rate is92.5%,which can avoid false positive pairs.Interference of the early warning system;Finally analyzing the movement direction of the person in the video,it is determined whether the person has a cross-border phenomenon to the pre-calibrated cross-border area,and Early warning of cross-border phenomena.(3)Based on video ID tracking,fall inspection and cross-border detection methods,a set of urban surveillance video automatic analysis system was constructed.The system integrates the following functions: video selection,image acquisition,image preprocessing,key point detection,ID tracking,fall detection and cross-border warning,and related functions can be turned on and off through interactive buttons.Through the system constructed in this article,users can analyze the video data recorded by multiple cameras in the city in real time,and provide early warning of behaviors that may endanger city safety.The urban video surveillance security analysis system constructed in this paper can effectively and automatically analyze the surveillance video in the city,realize the tracking of the target person in the video,fall detection and cross-border detection,which can satisfy the needs of urban supervision in a approaching way and has a wide range of application prospects.