A Robust Monte Carlo Tracking Algorithm Based on Feature Adaptive Selection

We propose a novel Monte Carlo tracking algorithm which can work robustly under complex dynamic scenario. Firstly, for the problem that particle filter tracking framework is prone to make the tracking failure under complex background when the features have low discriminative abilities, we design a feature adaptive selection mechanism based on online Adaboost algorithm. This mechanism can choose the most discriminative features online. Secondly, considering that online Adaboost algorithm is easy to cause "drift" phenomena as well as the features in the candidate feature pool are not reliable, we propose a novel half-forgotten sample set update strategy and a brand-new construction mode for the candidate feature pool which is based on color and pyramid gradient orientation histogram feature. Experimental results show that our tracker is able to handle severe appearance change and recover from drifts in realistic videos. The algorithm proposed in this paper can track the objects accurately and reliably compared with other existing object tracking algorithm.