全局与局部分块联合的目标跟踪算法

Target tracking algorithm via local-based and global-based collaborative modeling

基于相关滤波器(Correlation filter,CF)的目标跟踪算法因其高效率而引起了人们越来越多的兴趣,但这类算法对部分遮挡和形变十分敏感,可能导致最终跟踪失败。针对这一问题,该文将自适应互补模型引入到基于多分块的跟踪框架中,联合全局模型与局部分块模型来应对严重遮挡问题,并设置单独的快速尺度估计模块获得尺度信息。在跟踪基准数据集OTB2013上的实验表明,该文算法可以有效应对跟踪过程中的遮挡和形变问题,在保证实时性的同时提高目标跟踪精度。

Correlation filter has drawn increasing interest in target tracking due to its high efficiency,however,it is sensitive to partial occlusion and irregular deformation,which may result in tracking failure finally.To address this problem,this paper introduces the model complementary estimation into the part-based tracking framework,and combines the global and local model to deal with the severe occlusion problem.Also,this paper utilizes a separate fast multi-scale estimate method to obtain the information about scale variations.A large number of comparison experiments on the tracking benchmark datasets OTB2013 can demonstrate that the proposed algorithms perform favorably against the effect of target deformation and occlusion.In addition,our methods achieve the real-time online tracking while obtaining high accuracy.