Flow Statistics And State Detection System At The Entrance And Exit

With the improvement of people's quality of life,sports venues,movie theaters,shopping malls,etc.have become places where people gather.The security problem brought by crowd gathering has become the focus of today's society.At present,the mainstream method is to use crowd density estimation method in surveillance video and count people by estimating static image density map.Because the fundamental factor that causes crowd gathering is the unrestricted flow of imports and exports,leading to internal crowd gathering,the crowd density method can not be used to count the crowd gathering in a timely manner to control the crowd in a timely manner through the flow of people to avoid the occurrence of abnormal events.In view of the above problems,this paper takes the dynamic changes of the flow of people in the surveillance video as the research object,realizes the automatic flow statistics of the detection port and the automatic detection of abnormal events,and predicts the flow of people and predicts the occurrence of abnormal events.Managers can Make corresponding decisions according to the actual situation to avoid crowds and abnormal events inside the venue.Most of the people flow statistics use different detection ports to separate the flow of people into and out of the detection statistical method.In this paper,it is more difficult to select the same detection port for bidirectional flow of people statistics.Use different methods to count the flow of people in and out,and detect and judge whether there are abnormal events in the flow of people.The research contents of this article are as follows:(1)Incoming flow statistics based on the head detection method.Considering that the general surveillance camera monitors the flow of people from top to bottom,the head is more obvious,the head is used as the detection target,different networks are used for parameter adjustment training and model comparison selection,and the multi-task convolutional neural network(MTCNN)is selected.)As aninflow detection network,combined with the target matching tracking algorithm between frames to achieve inflow statistics.(2)Outflow statistics based on human detection methods.Different from the head detection of inflow,this paper adopts the method of identifying and tracking the human body for the statistics of the number of people in outflow.The loss function that reduces the impact of occlusion is combined with the single-stage network SSD and the two-stage network Faster RCNN,using public pedestrian data sets for corresponding parameter adjustment and model training,comparing the detection accuracy and speed of the model,and selecting the SSD network to reduce occlusion The affected loss function is used to leave the pedestrian detection network,and is combined with the flow statistics and inter-frame target matching tracking method to achieve the entire flow statistics.(3)Flow statistics and state detection system display.According to abnormal events,this article uses the video reconstruction algorithm and multiple sample learning algorithm experiment respectively,will be the result of the experiment comparison and analysis from several aspects,abnormal events to choose the appropriate detector,and combined with traffic statistics method is applied to the two-way flow situation,implementation of stream of people in video statistics and abnormal event detection,and set up the system in visual way to test results show,for warning abortion beyond and abnormal events.