Adaptive Probabilistic Tracking with Visual Saliency Selection Reliable Particles

It is very difficult to guarantee the stability and accuracy of object tracking in real-world scenarios. Although probabilistic tracking has become popular, there are two major problems: discriminative appearance modeling and reliable particles selection. To address these issues, this paper presents a visual saliency inspired approach that can capture the varying appearance characteristic of target and background, and select reliable particles during tracking. The global image signature descriptor, which can find the discriminant features for tracking, is used to model the appearances of target and background, as well each hypothetical observation, and can easily be embedded into the particle filter framework. The weight of each particle is estimated not only through saliency measurement implemented by Hamming distance between the target model and each hypothetical observation, but also through the background model and each hypothetical observation. The saliency selection reliable particles are used to estimate the target state. Finally, experimental results demonstrate the efficiency and effectiveness of the proposed method in presence of occlusion and large illumination variation, even non-target with similar features.