Research On Multi-scale Target Tracking Method

With the advent of the intelligent era and the rapid development of the computer industry,target tracking technology has become more and more prominent in security systems,unmanned driving,and traffic monitoring,attracting a large number of scholars and experts to invest in research.Correlation filtering algorithms have attracted much attention in recent years because of their superiority in taking into account tracking speed and accuracy.In actual application scenarios,object tracking faces difficulties such as target diversity and background complexity.Among them,the scale change of the target is a key factor that affects tracking performance.Based on SAMF and DSST proposed for the scale problem,this paper is dedicated to seeking further breakthroughs in the problem of object scale changes.The main tasks include as follows:(1)A step-by-step tracking algorithm is proposed based on SAMF.The algorithm is improved on the framework of SAMF,and the scale of the object is estimated by the scale pool method.For the problem of SAMF tracking where the scale changes too fast and the deployment of calculations in the area to be detected without discrimination,a step-by-step approach is adopted.In the first step,a smaller scale pool and a larger scale interval are used for tracking,and then the obtained target information is used as the input of the next step.In the second step,a finer-grained scale pool and a smaller area to be detected are used to get accurate information about the object.(2)A local uniform tracking algorithm based on DSST is proposed.Under the framework of DSST's separate tracking of the target's position and scale,the second-step scale estimation takes the position tracking result of the first step as the criterion,which may lead to the error accumulation.So the idea of local uniform tracking is adopted.On the basis of the target position information obtained in the first step,the proposed method is used to perform uniform tracking in a small area to be detected.(3)The algorithm proposed in this paper is verified on the improved OTB and VOT.Experiments show that the method in this paper has a certain degree of improvement in both the success rate and accuracy compared with two benchmark methods in the scene of scale changes,which proves the scale strategy proposed in this paper does really have a good effect on the scale change problem.In other scenes such as object disappearance,lighting changes,fast motion,etc,the proposed method in this paper also has a better performance than the benchmark method,which improves the overall performance of the tracker.(4)According to the algorithm of this paper,a target tracking system is designed to realize the tracking of specific targets.The system includes three parts:data operation,target tracking and file operation.