PCA-Based Appearance Template Learning for Contour Tracking

摘要

A novel method is proposed in this paper to model changes of object appearance for object contour tracking. Principal component analysis is utilized to learn eigenvectors from a set of the object appearance in our work, and then the current object appearance can be reconstructed by a linear combination of the eigenvectors. To extract the object contour, we perform covariance matching under the varia-tional level set framework. The proposed method is tested on several sequences under large variations, and demonstrates that it outperforms current methods without updating the appearance template.