Singular value decomposition based particle filter for tracking in complex environment

Tracking in complex environment is an important and challenging problem. Most of the existing methods adopt low level features derived from objects appearance. In contrast to this mainstream approach, this paper introduces a new method considering image as matrix then employing singular value as the features. For evaluating the likelihood between template and the candidate, which is weight update in particle filter, we offer Bhattacharyya coefficient based metrics. We compare our proposed algorithm with typical color histogram based particle filter in two challenges: confusing color in background and illumination change. Experiment results show that our approach achieves better tracking accuracy and robustness.