Abstract: The traditional parse representation based visual tracking mainly uses the grayscale feature of the target to construct the sparse representation model. Since grayscale feature is sensitive to the change of illumination, which may reduce the robustness of the target tracking in complex scenarios. The multisource data based visual tracking can significantly improve the visual tracking robustness, but how to effectively fuse different dimensions, different types of multi-source target features become the key issues to be solved in the multisource data fusion. This paper proposes a target state and grayscale feature fusion based sparse representation method for robust visual tracking. The proposed method can effectively fuse two features with different dimensions through using the kernel sparse representation model to explore the relation between target state and grayscale. The proposed method can improve the accuracy of the visual tracking in complex scenarios.

Key words: sparse representation, multisource data fusion, target tracking

摘要： 传统的基于稀疏表示的目标跟踪方法主要利用目标的灰度特征构建稀疏表示模型。由于灰度特征对光照变化敏感，这会影响目标跟踪在复杂场景下的鲁棒性。基于多源数据融合的目标跟踪可以明显提升目标跟踪鲁棒性，但如何有效融合不同维度，不同类型的多源目标特征成为基于多源数据融合的目标跟踪所要解决的关键问题。提出了一个基于目标状态以及灰度特征的稀疏表示目标跟踪方法。所提出的方法可通过基于核函数表示的稀疏表示模型，在探究目标状态以及灰度特征相关性的基础上，将两种不同维度的特征进行有效融合，提升目标跟踪在复杂场景下的鲁棒性。

关键词: 稀疏表示, 多源数据融合, 目标跟踪