Research On Pedestrian Multi-Target Tracking Algorithm Based On Video Surveillance

In recent years,target tracking algorithm has been widely used in vehicle,humancomputer interaction and monitoring.At present,the multi-target tracking algorithm of pedestrians has become a key research direction.In the process of pedestrian tracking,due to the fast movement of pedestrians,mutual occlusion between pedestrians and the influence of complex background such as weather and illumination,problems such as identity exchange,slow tracking speed and tracking loss are easy to occur during the tracking.Therefore,based on YOLO v5 and DeepSort algorithm,this paper optimized and integrated it to design a pedestrian multi-target tracking system under video surveillance.The main contents are as follows:(1)Aiming at the problem that YOLO v5 algorithm is not accurate enough for pedestrian detection and has poor robustness in complex environment,an improved YOLO v5 algorithm,CD-YOLO v5,is proposed.This algorithm firstly improves the CSP module in the backbone network to CSPA module which can retain more feature extraction ability.And add CA attention module to obtain more location information;A better clustering algorithm k-means ++ algorithm is used to replace the original K-Mean algorithm.For location loss,SIOU loss function is used to optimize the original CIOU loss function,and the effectiveness of the improvement is verified by experiments.(2)Aiming at the problems of identity transformation and occlusion in the tracking process,based on DeepSort algorithm,extended Kalman filter algorithm is used to replace the previous Kalman filter algorithm,and abnormal trajectory compensation algorithm is added to complete the optimization.By comparing the optimized algorithm with the unoptimized DeepSort algorithm and the one-stage target tracking algorithm JDE,it is proved that the improved algorithm can improve the accuracy of tracking and improve the influence of occlusion and other problems.(3)Based on the improved detector and tracker,a pedestrian multi-target tracking system is designed.For the possible interference problems such as environment,the CLAHE algorithm is firstly used for image preprocessing to improve the image clarity.Then the visualization experiment is carried out to prove that the pre-processing image input system can achieve better tracking effect.