基于稀疏表示的目标跟踪算法

Target tracking algorithm based on sparse representation

为了解决目标跟踪忽视背景信息的问题,运用非常稀疏的矩阵对目标与背景样本提取低维Haar特征,并将候选目标在过完备字典中进行稀疏表示,用块正交匹配追踪的方式对稀疏表示进行求解,通过残差对目标作最大似然估计,提高跟踪效果。实验结果表明,在光照变化、遮挡、复杂背景等情况下,基于稀疏表示的目标跟踪算法具有较高的鲁棒性,且在跟踪速度上有所提升。

In order to solve the defects of tracking algorithm which ignores the background information, a very sparse matrix is used to get low dimensional Haar feature of target and background, and the candidate target can be sparse representation in the over complete dictionary. Block orthogonal matching pursuit method is used to solve the sparse representation and get residual, then the maximum likelihood estimation is achieved. Experimental results show that the tracking method based on sparse representation still has very high robustness in illumination, occlusion and complex background conditions, and the tracking speed is improved.