基于多特征在线模板更新的鲁棒目标跟踪算法

Robust object tracking based on online update of multi-feature template

摘要:

在Mean-shift算法框架下提出一种基于多特征在线模板更新策略的鲁棒目标跟踪算法．首先,针对目标与背景色彩相似引发的跟踪漂移现象,提取照度不变性色彩特征与旋转不变性LBP纹理特征提取算法,并通过引入BWH算法实现多特征融合;其次,在传统的Mean-shift算法收敛条件上增加了直方图相似度校验,以避免陷入局部最优解．此外,还提出了基于直方图差异空间分布图的遮挡现象检测算法,从而提升了模板在线更新算法的准确性．实验结果表明,本文方法对于复杂动态场景、遮挡现象以及目标自身形变具有较强的鲁棒性和较高的准确性．

Abstract:

This paper proposes a robust object tracking algorithm under the Mean-shift framework based on the online update strategy of multi-feature template. At first, to solve the drift problem caused by cluttered backgrounds, the illumination invariant color features and the rotation invariant LBP texture feature were extracted and were combined together with the BWH algorithm. Secondly, in addition to the traditional convergence condition of Mean-shift algorithm, a histogram similarity checking step was presented against the local optima problem. Besides, occlusion detection algorithm based on spatial distribution of the histogram difference was proposed to enhance the precision of the template update. Experimental results showed that the proposed tracking algorithm is robust and accurate against cluttered dynamical background, occlusion and the object deformation.