Pedestrian Detection And Tracking In Subway Station Based On Deep Learning

Pedestrian detection and tracking technology has been a popular topic of research in the field of computer vision,which can be applied to public places such as banks,shopping malls and streets.With the progress continuously-achieved in the deep learning technology and with the improvement of target detection algorithm,pedestrian detection and tracking technology based on deep learning has gradually been applied to the emerging fields such as intelligent driving and intelligent transportation.In recent years,subways have been constructed all over the country and opened to the public one after another.Travel by subway not only avoids the traffic jams,but also provides a means of transportation at very affordable prices.Therefore,subway has become the first choice of short-distance travel for citizens,and for this reason there is a very high demand of pedestrian detection and tracking for the management of the passenger flow in such subway stations.First of all,Based on the existing actual needs of the scenes,the shortcomings regarding the performance and efficiency of the traditional target detection algorithm has been fully expounded in this paper,with focus attached on var Io Us target detection algorithms based on deep learning,optimizing the YOLOv3 algorithm.With this optimization the algorithm demonstrate not only a higher accuracy,but also a greatly improved speed.Following that,the YOLOv3 model is then trained by using the data set of pedestrian detection made by time-framed video recording at a certain exit of a station of Hangzhou Metro;and finally,the YOLOv3 model is used to realize the pedestrian detection in the subway station,which isbased on the real-time detection scene of the subway station.Based on the real-time scene of detection,the possibility of applying Kalman filtering and other theories to the pedestrian tracking has also been studied in this paper.The Deep SORT algorithm based on Kalman filtering theory is selected in this paper as the algorithm for multi-target pedestrian tracking,and is taken as the focus of the study.The algorithm employs the Kalman filtering theory to accurately predict the position of the target pedestrian's next movement;and with the Hungarian Assignment Algorithm being combined to detect and predict the optimal matching of pedestrian results,the purpose of pedestrian tracking is then achieved.In this paper,the YOLOv3 model and the Deep SORT algorithm are combined with consideration given to the vector inner product method to realize the real-time statistical calculation of both inbound and outbound passenger flows.