DeepSort Subway Passenger Tracking Algorithm Based On Effective Channel Attention Mechanism

Multi-target real-time tracking is a very hot research field at present,which belongs to the middle zone of visual perception field.The main purpose of multi-target trajectory tracking based on vision is to realize real-time and continuous tracking of pedestrian targets in public places.During this period,the ID assigned to each target pedestrian does not change.After tracking,the motion state and motion position of pedestrians in the next few frames can be reasonably predicted.This paper mainly studies the problem of insufficient extraction of target image feature information of multi-objective passenger flow tracking algorithm under the subway crowded scene,and studies the practical DeepSort algorithm which is more suitable for the scene in this paper.In this paper,the DeepSort algorithm is divided into three modules,which are the passenger flow image feature extraction module,the passenger flow target feature selection module and the position prediction module of the future frame of the passenger flow target.This paper mainly improves the algorithm of the passenger flow image feature extraction module of the DeepSort algorithm.The main contributions are as follows:Starting from the three main evaluation indexes of accuracy rate,recall rate and AP,the detection effects of YOLOv3 subway passenger flow target detection algorithm and Faster R-CNN subway passenger flow target detection algorithm on the subway passenger flow data set were compared respectively.With residual number,number of false alarm and calculated based on this data base multiple comprehensive index as the main body of evaluation standard,compares the DeepSort metro passenger flow target tracking algorithm and Da Siam RPN metro passenger flow target tracking algorithm on the metro passenger flow data set tracking effect,have chosen to perform better under this scenario target detection algorithm and target tracking algorithm.Aiming at the problem that DeepSort algorithm is insufficient in image feature extraction in the subway passenger flow trajectory tracking scene,based on the SE module(ECA-Net)which adds a 1D fast convolution layer of size k,the interaction information between local channels is extracted without reducing the vector dimension.The ECA-Net module is embedded into the Res Net50 network,and the network is used as the backbone network of the target detection algorithm to complete the target detection task together.In the case of subway passenger flow tracking,the target detection algorithm based on ECA-Net was introduced into the DeepSort multi-target tracking algorithm,and a DeepSort subway passenger flow tracking algorithm based on channel attention mechanism was proposed,which effectively solved the problem that the DeepSort algorithm was not sufficient in extracting subway passenger flow features.The improved algorithm was tested and evaluated on the subway passenger flow data set,and all the comprehensive indicators were improved.The results strongly show that the improved pedestrian re-recognition algorithm model has stronger robustness in terms of pedestrian feature expression and can achieve higher recognition accuracy.