基于蒙特卡罗方法的目标跟踪  

Object Tracking with Monte Carlo Methods

作　　者：张海青[1] 李厚强[1] 

机构地区：[1]中国科学技术大学电子工程与信息科学系,合肥230027

出　　处：《中国图象图形学报》2008年第5期937-943,共7页Journal of Image and Graphics

基　　金：国家自然科学基金委员会与微软亚洲研究院联合资助项目(60672161)

摘　　要：为了更鲁棒和快速地进行目标跟踪,在基于粒子滤波的目标跟踪方法的启发下,提出了一种新的基于蒙特卡罗方法的目标跟踪方法。该方法首先运用蒙特卡罗技术对下一帧目标可能出现的位置和尺度进行抽样;然后计算各抽样与参考目标的相似度;最后通过估计目标状态来获得跟踪目标。实验表明,该方法无需目标运动信息,特别适用于目标灵活运动时的跟踪,与现有的算法相比,不仅算法实现简单,同时有较好的鲁棒性和通用性。For tracking object more robustly and rapidly, a novel approach for object tracking using Monte Carlo method is presented, which is motivated by the method of tracking with particle filter. In this approach, the locations and scales of the candidates in the next frame are sampled through Monte Carlo technique. Then the similitude degree between the samples and the reference objects are calculated. At last, the object states are estimated and the object under tracking is attained. This method does not need the motion information of the object, so it is especially good at tracking flexible object. The method is easier to implement than other present methods. Furthermore, it is proved to be quite robust and versatile by the numerical experiments.

关 键 词：目标跟踪 蒙特卡罗方法 贝叶斯滤波 粒子滤波 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]