Research And Application Of Pedestrian Multi-Target Tracking Algorithm Based On DeepSORT

For the DeepSORT algorithm,the description of pedestrian apparent feature information in the process of pedestrian tracking is not enough to meet the needs of actual complex scenes.In order to improve the tracking accuracy of DeepSORT algorithm and make it more robust,by improving the feature extraction network of DeepSORT algorithm,extracting multi-layer pyramid features,increasing the number of residual layers of the network,improving the breadth and depth of feature extraction,and verifying it on the public data set mot-16,the tracking accuracy of DeepSORT algorithm is improved by 4.1%.At the same time,in order to solve the IOU(Intersection over Union)distance measurement,when there is no overlap between the detection frame and the prediction frame,the IOU value is always 0,and the weight value between the detection frame and the prediction track frame cannot be judged.At the same time,when the detection frame and the prediction frame intersect,there will be different overlapping forms,and the IOU distance value is the same.In order to solve this problem,GIOU distance measurement is used to replace the IOU distance measurement in DeepSORT algorithm to improve the accuracy in the tracking process,and MOTP reaches 81.9.In order to solve the problem of pedestrian density statistics,this thesis proposes a pedestrian line collision counting method based on DeepSORT algorithm.This method first uses the detection algorithm to detect the position of the traveler target,and then combines the Deepsort algorithm to predict the trajectory through the Kalman filter algorithm to complete the tracking of the pedestrian target,keep the ID information unchanged,and judge the direction of the target passing through the counting line according to the trajectory and ID information of the target,This method can effectively calculate the number of people passing through the counting line from different directions.Indirectly count the number of people entering a place or passing through an intersection to achieve the purpose of tracking and counting.