A Saliency Based Tracking Method Via Two-stage Sampling

Visual tracking is a challenging task within the field of computer vision due to the real-world complicated scenarios, the appearance changes of the target and the uncertainty of the target motion. To design a robust visual tracking algorithm, a discriminative appearance model is a must as well as a good search strategy to predict the target. This paper focuses on the abrupt motion problem of the target to build a robust visual tracking framework.In order to come up with a good solution to abrupt motion tracking in complicated scenarios, the visual saliency detection is firstly introduced into the Wang-Landau Monte Carlo (WLMC) framework to provide a prior for the global search; Then combined with the Interactive Markov Chain Monte Carlo (IMCMC), a two-stage sampling method is designed to improve the accuracy of the target search. Main contribution of this paper can be summarized as below:1、Firstly, a saliency-based Wang-Landau Monte Carlo(WLMC) tracking method for abrupt motion problem is proposed. In this scheme, the visual salience is introduced as prior into the WLMC-based tracking algorithm. By dividing the spatial space into disjoint sub-regions and assigning each sub-region a saliency value, a prior knowledge of the promising regions is obtained; then the saliency value of sub-regions is integrated into the Markov Chain Monte Carlo acceptance mechanism to guide effective states sampling which results in the success of capture the target.2、Secondly, a saliency based tracking method via two-stage sampling is proposed Considering the abrupt motion sequence contains both abrupt and smooth motions, a two-stage sampling model is built. In the first stage, the model detects the motion type of the target. According to the result of the first stage, the model chooses either the Saliency-based WLMC method to track abrupt motions or the IMCMC method to track smooth motions of the target in the second stage. The algorithm efficiently addresses tracking of abrupt motions while smooth motions are also accurately tracked, and thus help to achieve precise sampling.The proposed algorithms are compared with alternative tracking methods. According to the experimental results, the saliency-based WLMC tracking method outperforms against state-of-the-art algorithms on abrupt motion tracking; the saliency based tracking method via two-stage sampling efficiently addresses tracking of abrupt motions while smooth motions are also accurately tracked. Compared to5state of the art single object tracking algorithms, our method tracks the object with stable performance and smaller center position error.