Robust Visual Tracking via Multiple Kernel Boosting with Affinity Constraints

We propose a novel algorithm by extending the multiple kernel learning framework with boosting for an optimal combination of features and kernels, thereby facilitating robust visual tracking in complex scenes effectively and efficiently. While spatial information has been taken into account in conventional multiple kernel learning algorithms, we impose novel affinity constraints to exploit the locality of support vectors from a different view. In contrast to existing methods in the literature, the proposed algorithm is formulated in a probabilistic framework that can be computed efficiently. Numerous experiments on challenging data sets with comparisons to state-of-the-art algorithms demonstrate the merits of the proposed algorithm using multiple kernel boosting and affinity constraints.