The Crowd Movement Significance Test Based On Flow Field Topology

With the rapid development of artificial intelligence,the development of intelligent monitoring technology has been paid more and more attention by researchers.For monitoring the crowd behavior of video detection and analysis is the core content of intelligent monitoring system,its main task is from continuous changes in the image detection and extract the movement characteristics of the crowd,and realize the description and analysis of crowd motion state.How to detect the significance of the movement of the crowd is the focus of this study.At present,the analysis of crowd behavior is mainly divided into two categories: the micro-analysis of the individual in the crowd and the macroscopic analysis of the overall behavior of the crowd.Due to the large number of microscopic methods and the mutual obstruction between individuals,there is a large error.Therefore,this paper proposes a method to detect the significance of crowd movement based on flow field topology.Its specific contents include the establishment of crowd motion flow field,the calculation of flow field topological structure and the clustering of motion trajectory.First,the video is divided into frame-by-frame images,each pixel in the image is viewed as a moving particle,and the optical flow method is used to extract the speed of particle motion.Then,the clustering method is used to find the velocity vector of the moving particle in the video cycle to obtain a constant crowd motion flow field.Secondly,according to the method of particle advection in the crowd movement flow field,the particle trajectory information is obtained,and then the particle trajectory is obtained.The information establishes a topological model of the flow field and uses this to obtain significant information including critical point information and boundary information.The critical point is determined by the two parameters of divergence and curl,and the boundary information is calculated using Lagrange's coherent structure;Finally,the trajectory is similarly clustered by using three different mainstream clustering algorithms.Then by analyzing the advantages and disadvantages of the three clustering methods,the method of density clustering combined with hierarchical clustering is used to divide thetrajectories of crowds.Combining saliency information with trajectory information enables topological analysis of crowd flows.This article has experimented with multiple sets of videos in different scenes in the UCF database.The results show that the algorithm can effectively detect the movement state and significance area of the crowd.