一种检测区域动态更新的目标跟踪算法

A Target Tracking Algorithm for Dynamic Update of Detecting Regions

TLD(Tracking-Learning-Detection)是针对视频中未知物体的长时间跟踪算法。当目标跟踪受到光照、遮挡和旋转等因素干扰时,TLD算法无法实现视频目标准确跟踪。针对TLD存在的问题,提出了一种改进的TLD动态更新检测区域的目标跟踪算法。采用压缩跟踪作为跟踪模块,并且增加了对压缩跟踪模块和检测模块运行效果的评估,根据评估结果来调整压缩跟踪模块和检测模块所占的权重,更准确预测目标的位置。在每次检测之前根据前一帧目标的位置和目标的历史运动速度选定一个动态更新的检测区域,缩短了检测模块和学习模块的运行时间。通过实验对原始TLD和改进后的TLD进行了对比测试,结果表明,改进后的TLD算法较原始TLD算法具备更高的跟踪成功率和更快的处理速度。

TLD(Tracking-Learning-Detection)is a long-term tracking algorithm for unknown objects in video.When the target tracking is interfered by factors such as illumination,occlusion and rotation,the TLD algorithm cannot achieve accurate tracking of the video target.Aiming at the problem of TLD,this paper proposes an improved target tracking algorithm of dynamic updating detection area for TLD.The compression tracking was adopted as the tracking module,and the evaluation of the operation effect of the compression tracking module and the detection module was added.And the weights of the compression tracking module and the detection module were adjusted according to the evaluation result in order to more accurately predict the location of the target.Before each detection,a dynamically updated detection area was selected according to the position of the previous frame and the historical moving speed of the target,which shortened the running time of the detection module and the learning module.The original TLD and the improved TLD were compared by experiments.The results show that the improved TLD algorithm has higher tracking success rate and faster processing speed than the original TLD algorithm.