Deep Learning-based Train Station Video Scene Detection Of Abnormal Human Behavior

For nowadays,the continuous development of network,integration and high level of intelligent security makes it more and more urgent to monitor the emergency detection system in railway stations.The research of emergency detection system in train station video scene based on deep learning becomes a key part of smart city solutions.This paper designs an emergency detection system based on deep learning for the video monitoring of the current train station scene.It uses the method of combining detection and regression to realize the number counting,and then the improved target tracking algorithm is used to track the specific person,and the human body attitude estimation algorithm is used to extract the skeleton data of the person with a specific identity,and finally the fall behavior recognition is completed.Firstly,according to the characteristics of waiting rooms in railway stations,this paper improves the existing CSRNet regression method network.In addition,YO LOv3 algorithm is used to realize the detection method count of sparse parts.By referring to Decide Net network,the counting result of the regression model is weighted to solve the problem of inconsistent training data set and application scenario and inaccurate crowd counting.Thus,the combination of detection method and regression method makes the algorithm more accurate in crowd counting in various scenes with dense and sparse crowd scenes.Then,it studied the target tracking algorithm of railway station waiting room.Based on the existing two-stage detection algorithm,this paper tries to use the one-stage detection YOLOv3 algorithm to realize the rapid detection of individual population.And after realizing the detection box of the individual crowd under the camera field of vision,the tracking algorithm is used to realize the continuous tracking of the single target.Among them,the original fixed size tracker was changed to a tracker that changed size with the detection box,which improved the existing tracking algorithm and increased the accuracy of target tracking.Finally,the system imp lements the particular persons’ identity matching so as to extract the specific personnel data skeleton with using the tracking model.And then to realize the unattended detection of emergency in railway station,the fall behavior is identified from the speed,acceleration and moving frame rate of human body’s key points.The emergency detection algorithm proposed in this paper can make the alarm of decision-making command in time,guide the station staff to deal with the emergency quickly and effectively,and greatly improve the work efficiency as well as the safety of passengers.