Research On Pedestrian Detection And Tracking Algorithms In Shopping Malls Based On Deep Learning

With the improvement of social and economic level,there are more and more large shopping malls in life.Modernized shopping malls put forward more and more urgent needs for intelligent analysis of passenger flow,and it is vital to realize pedestrian detection and tracking in the shopping mall environment.Aiming at the problem of slow detection speed and poor tracking effect in pedestrian detection and tracking in the current shopping mall environment,based on the optimization of YOLOv3 and Deep-Sort algorithm,a multi-target pedestrian detection and tracking algorithm applied in the shopping mall environment is designed.Research work in the following three areas:(1)Aiming at the problem of inaccurate positioning and slow positioning speed of pedestrian detection algorithms in shopping malls.It is planned to propose an optimized algorithm based on YOLOv3.First,the Darknet-53 convolutional feature extraction network is simplified,and the Conv layer and the BN layer are merged in the network reasoning stage;secondly,for the pedestrian positioning problem,the DIo U loss function is used instead of the Io U loss function;finally,the K-means++ clustering algorithm is used to obtain The best anchors parameters.Three improvement strategies are applied to the YOLOv3 algorithm.It can be seen from the experimental results that the detection speed and detection accuracy have been improved after the optimization of the detection algorithm.(2)Aiming at the disadvantage of the poor robustness of the Deep-Sort algorithm,we plan to propose a multi-target pedestrian tracking algorithm that replaces the Kalman filter algorithm in the prediction phase with the unscented Kalman filter algorithm.It can be seen from the experiment that after the improvement,the number of ID jumps is reduced,and the MOTA index and MOTP index are also improved.(3)A set of pedestrian detection and tracking system in shopping mall environment has been preliminarily built.Aiming at the problem of video processing with poor background environment in practical applications,CLAHE image processing function is added to the system.The experiment shows that the accuracy of model detection after image processing Improved and improved the visual effect of pedestrian tracking.The pedestrian detection algorithm is combined with the tracking algorithm,and the actual tracking effect is tested.orithm is combined with the tracking algorithm,and the actual tracking effect is tested.The experimental results show that the optimized algorithm in this paper can detect and track multiple pedestrians in a shopping mall environment.This is of great significance for the intelligent analysis of passenger flow in modernized shopping malls.It has certain reference value for the development and application of multi-target pedestrian detection and tracking algorithms.