基于积分直方图的快速粒子滤波跟踪算法

基于直方图的粒子滤波已成功地用于解决计算机视觉中的目标跟踪问题,但是,在观测似然计算上的低效限制了它们的实时应用。针对该问题,提出了一种快速的粒子跟踪方法。其建立在积分直方图技术的基础上,使得每个候选样本的观测似然能够由少量的查找表运算有效地计算出来。该方法使用了大量的粒子以确保鲁棒性,同时确保具备实时跟踪的能力。实验结果表明该方法在计算效率上优于通常的粒子滤波跟踪方法。

A fast particle filter tracking method based on integral histogram

Histogram-based particle filters for visual tracking have obtained success in many challenging tasks, but the computational inefficiency in evaluating the observation likelihood limits their use for real-time applications. A fast particle tracking method is proposed to this problem building on the integral histogram technique, with which the likelihood for each candidate can be computed efficiently by a few table lookup operations. As a result, the proposed method can employ a large number of particles to ensure the robustness while achieving the ability of tracking in real-time. Experimental results show that the method is computationally superior to the conventional particle tracking method.