基于自适应粒子滤波器的物体跟踪

Object Tracking Based on Adaptive Particle Filter

利用分类概念及粒子滤波理论,提出了一种基于自适应粒子滤波器的物体跟踪算法。将Boosting算法引入粒子滤波器,构建了自适应粒子滤波器,该方法首先利用背景信息和目标信息建立特征分类器,将分类器的输出结果作为粒子滤波系统观测的重要信息,进行粒子权值的计算,并在跟踪过程中不断更新特征分类器,从而自适应地更新粒子的权值。实验结果表明,该算法可以根据背景信息的不同自适应地选择特征,对于存在遮挡、形变及背景干扰等情况,依然可以很好地对目标进行稳定跟踪。

An object tracking algorithm based on adaptive particle filter is proposed in this paper. Boosting algorithm is introduced into particle filter algorithm, and adaptive particle filter is constructed. Features classifiers are constructed utilizing object information and background information, and the outputs of these classifiers taken as important information of observations of particle filter are used to calculate particles＇ coefficient. Also, these classifiers are updated during tracking in order to update particles＇ coefficient adaptively. The experiment result shows that the tracking algorithm we proposed can adaptively select features for tracking utilizing different background information, in applications such as existence of covering, appearance changed, clutter in the background and illumination changing. The objects can be tracked stably.