Target tracking by multiple cues based on genetic particle filter

This paper represents a mutate resample method to modify the particle impoverishment problem, which increase the diversity of particles to ensure a better estimate of posterior density. A precise observation model is necessary to track target robustly and accurately, which integrate the intensity and gradient cue base on the characteristics of the target image sequence. An adaptive fusion method is proposed that a log likelihood ratio of sample densities from target and background is computed. Finally, we use a model updating strategy to change observation template appropriately. Experiments show that the modified algorithm has a better tracking performance, which can deal with the occlusion and severe background interference in the tracking process.