Adaptive fusion strategy based multiple-cue tracking

Multiple cue fusion based tracking is one of the most active research in tracking literature. In this paper, a novel adaptive fusion strategy is proposed for multiple cue fusion, base on two common used fusion rules: product rule and weighted sum rule. This strategy employs particle filtering technique, estimating second order moment of the weighted sample set and computing its Frobenius norm to denote how cues are reliable, and then switch the two fusion rules in time. In practice, the new fusion strategy shows more robustness than traditional single fusion rule.