基于特征显著性的均值漂移鲁棒目标跟踪

在均值漂移算法框架下，提出基于目标显著性的特征融合与在线模板更新策略，实现复杂动态环境下的鲁棒跟踪.通过目标区域与背景区域的特征对比定义了特征显著性测度.提出了基于特征显著性的色彩空间选择以及基于Gabor小波稀疏编码的纹理特征提取算法.通过特征显著性加权实现参考直方图模板的初始化，并在此基础上针对遮挡现象与目标自身形变的区别设计了在线模板更新策略.实验结果表明，本文方法与其他跟踪算法相比具有较强的鲁棒性和较高的准确性.

Meanshift Robust Object Tracking Based on Feature Saliency

This paper proposes the saliency-based feature fusion and template updating strategy in the meanshift framework to realize robust tracking in dynamic complex environment. Firstly, A feature saliency measurement is defined as the contrast between object and background. The optimal color space based on feature saliency and the texture feature based on Gabor wavelet sparse coding are extracted. The reference histogram template initialization is weighted by feature saliency and an online template updating strategy is presented for occlusion and object deformation. Experimental results show that the proposed model is more precise and more robust compared with up-to-date competing models.