Adaptive multi-cue fusion for visual target tracking based on uncertainly measure

This paper presents a novel adaptive tracking algorithm that fuses multiple cues based on feature uncertainty measurement in the particle filter framework. We first introduce a self-adaptive multi-cue fusion strategy, which overcomes the drawbacks of the traditional product fusion and sum fusion strategies. Furthermore, the proposed strategy effectively sharpens the distribution of the fused posterior as well as makes the tracking results less sensitive to the noise. Then, based on the fact that tracking failure often happens in the cases of low discriminative abilities of the observed features, we define a new feature uncertainty measurement. The proposed uncertainty measurement is thereafter used to adaptively adjust the relative contributions of different cues to tracking. An extensive number of comparative experiments show that the proposed tracking algorithm is more stable and robust than the single feature, product fusion, and sum fusion tracking algorithms.