用于目标跟踪的自适应粒子滤波算法

Adaptive Particle Filtering for Efficient Object Tracking

结合粒子滤波算法,提出了一种能够根据目标运动特征自动确定粒子数的自适应目标跟踪算法。为了准确表示后验概率密度,粒子滤波通常使用大量的粒子。当运动预测准确时,用少量的粒子就可以准确估计概率密度函数。提出的算法利用描述概率密度所需的粒子数和运动估计准确程度之间的关系,自动确定粒子滤波所需的粒子数及其提议分布,提高了粒子的使用效率,避免了由于使用过多粒子而增加跟踪器计算量的问题。实验结果表明该算法可以有效地估计出进行目标跟踪所需的粒子数目。

An adaptive object tracking algorithm based on particle filters which could determine the number of the particles according to the object dynamic characteristic was proposed. Particle filters utilize many particles to approximate the posterior probability density function （pdf）. However, when the prediction of the object movement is accurate, a few of particles is required to represent the posterior pdf. The proposed algorithm determines the number of particles and proposal density based on the relationship between the numbers of particles required to represent the posterior pdf and accuracy of the movement estimation. The proposed algorithm improves the ej^ciency of the particles and alleviates the increment of computational complexity caused by using too many particles. The experimental results show that the proposed algorithm can determine efficiently the number of the required particles under the condition of accurate tracking.