In this paper, a saliency based tracking method via two-stage sampling is proposed for abrupt motions. Firstly, the visual salience is introduced as a prior knowledge into the Wang-Landau Monte Carlo (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 values of sub-regions are integrated into the Markov chain Monte Carlo (MCMC) acceptance mechanism to guide effective states sampling. Secondly, considering the abrupt motion sequence contains both abrupt and smooth motions, a two-stage sampling model is brought up into the algorithm. 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 double-chain MCMC method to track smooth motions of the target in the second stage. The algorithm effciently addresses tracking of abrupt motions while smooth motions are also accurately tracked. Experimental results demonstrate that this approach outperforms the state-of-the-art algorithms on abrupt motion sequence and public benchmark sequence in terms of accuracy and robustness.%针对运动突变目标视觉跟踪问题,提出一种基于视觉显著性的两阶段采样跟踪算法。首先,将视觉显著性信息引入到Wang-Landau 蒙特卡罗(Wang-Landau Monte Carlo, WLMC)跟踪算法中,设计了结合显著性先验的接受函数,利用子区域的显著性值来引导马尔可夫链的构造,通过增大目标出现区粒子的接受概率,提高采样效率；其次,针对运动序列中平滑与突变运动共存的特点,建立两阶段采样模型。其中第一阶段对目标当前运动类型进行判定,第二阶段则根据判定结果采用相应算法。突变运动采用基于视觉显著性的WLMC 算法,平滑运动采用双链马尔可夫链蒙特卡罗(Marko chain Monte Carlo, MCMC)算法,以此完成目标跟踪,提高算法的鲁棒性。该算法既避免了目标在平滑运动时全局采样导致精度下降的缺点,又能在目标发生运动突变时有效捕获目标。实验结果表明,该算法不仅能有效处理运动突变目标的跟踪问题,在典型图像序列上也具有良好的鲁棒性。
展开▼
