自适应上下文感知的目标追踪方法

Adaptive context-aware target tracking method

针对相关滤波方法容易受到背景干扰导致跟踪漂移的问题,提出自适应上下文感知图像跟踪方法.为了减少背景干扰,选取离目标位置远的高响应区域为自适应上下文区域,赋予该区域自适应的低响应值.根据上下文区域与目标区域响应的相对差值,给上下文区域自适应的惩罚因子,使得该算法具有更好的鲁棒性.该算法在OTB2013、OTB2015及Temple-Color128标准数据集上都展现了优秀的跟踪性能,OTB2015的重叠率精度达到61.53%,超过大部分已有的优秀算法,特别是在背景混叠及部分遮挡的情况下有着更卓越的表现.该算法的平均跟踪速度为24.5帧/s,实时性较好.

An adaptive context-aware target tracking method was proposed aiming at the problem that the correlation filtering methods were easily interfered by background,which led to tracking drift.The high response area far from the target position was selected as the adaptive context area in order to reduce the background interference.Then the adaptive low response value was assigned to the area.The penalty factor was adaptively given to the context area according to the relative difference of response value between the context area and the target area,which made the algorithm more robust.The algorithm showed excellent tracking performance on OTB2013,OTB2015 and Temple-Color128 benchmark.The overlapping rate accuracy of OTB2015 was 61.53%,which was superior to most existing excellent algorithms.The algorithm performed better especially in the case of background clutter and partial occlusion.The average tracking speed of the algorithm was 24.5 frames per second,and the algorithm had a good real-time effect.