基于笛卡尔乘积字典的稀疏编码跟踪算法

Sparse Coding Visual Tracking Based on the Cartesian Product of Codebook

为了提高基于稀疏编码的视频目标跟踪算法的鲁棒性,该文将原始稀疏编码问题分解为两个子稀疏编码问题,在大大增加字典原子个数的同时,降低了稀疏性求解过程的计算量。并且为了减少1?范数最小化的计算次数,利用基于岭回归的重构误差先对候选目标进行粗估计,而后选取重构误差较小的若干个粒子求解其在两个子字典下的稀疏表示,最后将目标的高维稀疏表示代入事先训练好的分类器,选取分类器响应最大的候选位置作为目标的跟踪位置。实验结果表明由于笛卡尔乘积字典的应用使得算法的鲁棒性得到一定程度的提高。

In order to improve the robustness of the visual tracking algorithm based on sparse coding, the original sparse coding problem is decomposed into two sub sparse coding problems. And the size of the codebook is intensively increased while the computational cost is decreased. Furthermore, in order to decrease the number of the1？-norm minimization, ridge regression is employed to exclude the intensive outlying particles via the reconstruction error. And the sparse representation of the particles with small reconstruction error is computed on the two subcodebooks. The high-dimension sparse representation is put into the classifier and the candidate with the biggest response is recognized as the target. The experiment results demonstrate that the robustness of the proposed algorithm is improved due to the employed Cartesian product of subcodebooks.