Research On Visual Group Targets Tracking Technology

Target tracking is widely used in the field of production and life,both in military and civil applications.When tracking the migrating animals,birds and UAVs in the airport,it is necessary to automatically track these targets as a whole and obtain each individual's trajectory.Therefore,the research of group target tracking technology has a broad application prospect.However,in the process of group target tracking,there are many interferences,large scale range,many kinds of targets and complex background,so the task of tracking faces a series of challenges,which makes the research of group target tracking technology need a higher theoretical basis.Aiming at these problems and difficulties,this paper takes these highly similar moving multi-target as the object,and studies the tracking technology for group target.This paper constructs and labels a group target tracking data set.The whole data set includes 2411 frames and 13109 targets.It covers a variety of different target types,different scales,different backgrounds and different target motion modes.The data set also includes the target deformation,target occlusion,small target,scale change and other factors that are not conducive to stable tracking interference.Aiming at the problems of poor tracking performance of static templates and the vulnerability of dynamic templates based on the results of previous frame tracking to background pollution,this paper proposes an iterative template design idea.This method uses time series information and target response degree to update the target template automatically and dynamically in the tracking process,so that the adaptability of the template in the tracking process is enhanced.In this paper,we combine the proposed template design method with the Siam FC network and carry out experiments.It is found that the new iterative template and the static template have different advantage state domains.In view of this discovery,this paper further proposes a fusion template design method,using different template response,adaptive dynamic template and static template fusion design.The experimental results show that the fusion template improves the performance of group target tracking,and has better effect under different interference factors.To solve the problem of mutual interference between adjacent targets in the tracking process of group targets,a graph similarity evaluation index is proposed in this paper.Then this paper combines the evaluation index,fusion template and Siam FC,and proposes a group target tracking algorithm based on the graph similarity constraint,which improves the tracking performance by using the graph similarity constraint between frames.The results of ablation experiments show that the performance of group target tracking can be effectively improved by adding similarity constraints.