Robust object tracking based on structural local sparsity via a global L2 norm constraint

摘要

In the structural local sparse model, every candidate derived from the particle filter framework is divided into several overlapping image patches. However, in the tracking process, the structural characteristics of the target may change due to alterations in appearance, resulting in unstable pooled features and therefore drifting and false tracking. We propose a method to correct the changed part of the target using atoms in the patched dictionary by adding a global constraint. If the target is corrupted, this constraint term will weaken the influence of variation and strengthen the stability of the pooled features. Otherwise, the method is based on the whole target and will protect its spatial continuity. Both qualitative and quantitative evaluations on challenging benchmark image sequences demonstrate that the proposed algorithm has excellent tracking behavior, displaying robustness and stability with little drifting on a target with altering appearance and partial occlusion.