Robust visual tracking via a compact association of principal component analysis and canonical correlation analysis

摘要

We propose a novel correlation-based incremental tracking algorithm based on the combination of principal component analysis (PCA) and canonical correlation analysis (CCA), which called Principal Component-Canonical Correlation Analysis (P3CA) tracker. We utilize CCA to evaluate the target goodness, resulting in more robust tracking than using holistic information, especially in handling occlusion. PCA is adopted to solve the Small Sample Size (3S) problem and reduce the computation cost in the generation of CCA subspace. To account for appearance variations, we propose an online updating algorithm for P3CA tracker, which updates the PCA and CCA cooperatively and synchronously. Comparative results on several challenging sequences demonstrate that our tracker performs better than a number of state-of-the-art methods in handling partial occlusion and various appearance variations.