摘要：

针对视觉跟踪中的目标遮挡问题，提出一种基于稀疏表达的视觉跟踪算法。采用稀疏表达方法描述跟踪目标，构造基于Gabor特征的目标词典和遮挡词典，通过l1范数最优化求解稀疏表达系数。在粒子滤波框架下跟踪目标，根据稀疏表达系数判断遮挡，并利用重构残差更新遮挡情况下的粒子权重。在目标模板更新时，通过引入可靠性评价来抑制模板漂移。实验结果表明，该算法能够有效地跟踪处于遮挡状态下的运动目标，并对目标姿态变化以及光照变化具有较好的鲁棒性。

关键词: 目标跟踪, 粒子滤波, 稀疏表达, Gabor特征, 词典更新, 遮挡

Abstract:

A novel visual tracking algorithm based on sparse representation is proposed to solve the problem of occlusion. The tracked object is described using the sparse representation method, and the image Gabor-features are used to construct the object dictionary and occlusion dictionary. The optimal sparse coding coefficients are obtained via l1-norm minimization. The tracking algorithm is designed in a particle filtering framework. The occlusion is judged according to the distribution of nonzero values in sparse coding coefficients. Under occlusion, the particles’ weight is calculated based on the approximation residual of observation by sparse representation. A template reliability evaluation method is introduced to suppress the drift during the object dictionary update. Experimental results show that the proposed algorithm can handle occlusion efficiently, and be robust to pose and illumination variations.

Key words: object tracking, particle filtering, sparse representation, Gabor feature, dictionary update, occlusion

中图分类号: 

• TP391