基于全局与局部结构反稀疏外观模型的目标跟踪算法*

为了提高稀疏表示跟踪模型性能,提出基于全局与局部结构反稀疏外观模型的目标跟踪算法(GLIS).首先采用反稀疏表达方式一次求解优化问题,计算所有粒子权重以提升算法实时性.然后,提出基于联合判别相似度图(JDS map)排名机制以提升算法鲁棒性,将候选目标分块并分别计算加权稀疏解,联结不同权重的局部块为整体并计算其稀疏解.最后采用联合机制将2种稀疏解合并为JDS map.在跟踪过程中,采用双重模板更新机制更新目标模板及权重模板.实验表明,在复杂环境下,文中算法仍然可以准确跟踪目标.

Object Tracking via Global and Local Structural Inverse Sparse Appearance Model

To improve the performance of sparse representation based trackers, an object tracking method based on global and local structural inverse sparse appearance model is proposed. Firstly, an inverse sparse representation formulation is proposed to compute the weights of all particles by solving one optimization problem and this is conducive to improving the real-time performance. Then, a ranking mechanism based on joint discriminative similarity map(JDS map) is designed to improve the robustness. The formulation block candidates are divided into several pitches and the weighted sparse solutions are computed respectively. Next, these pitches are concatenated with different weights and meanwhile the sparse solution of each particle is computed. A combination mechanism is proposed to unite two sparse solutions as the JDS map. During the tracking, the object template and weight template are updated using a double-template updating strategy. Experiments demonstrate that the proposed algorithm is robust for benchmark video sequences under complicated conditions.