一种结合空间信息和稀疏字典优化的目标跟踪算法

针对复杂场景中目标表观变化引起的跟踪漂移问题, 提出一种新的基于稀疏表示的目标跟踪算法. 该算法通过稀疏性和空间相关性正则约束得到一种优化的目标代价函数, 利用拉格朗日对偶理论和加速近端梯度方法完成字典优化, 并利用最大池化理论和空间金字塔方法得到降维的且包含更多空间信息的目标模板系数和候选样本系数. 实验结果表明, 所提出的算法在背景干扰、光照变化、形变、运动模糊、严重遮挡等多种复杂场景中都能取得较为鲁棒的跟踪效果.

An object tracking algorithm combining spatial information and sparse dictionary optimization

Aimming at the problem of tracking drift caused by object appearance change in complex scene, a novel object tracking algorithm based on sparse representation is proposed. An optimized objective cost function is designed with the sparsity and spatial correlation regularization constraint. The Lagrange dual theory and the accelerate proximal gradient approach are used to complete the dictionary optimization. By using the maximum pooling theory and the spatial pyramid method, the coefficients of the object template and candidate samples with the reduced dimension and more spatial information are obtained. Experimental results show that the proposed algorithm can perform robust tracking effect in a variety of complex scene, such as background clutters, illumination variation, deformation, motion blur, heavy occlusion, and so on.