基于粒子滤波和Galerkin法的改进目标跟踪算法

Improved object tracking algorithm based on particle filter and Galerkin's method

在粒子滤波框架下,估计的准确性受到建议分布选取的影响很大。传统的粒子滤波通常采用系统转移概率作为建议分布,但传统的建议分布选取方法由于没有考虑新的观测信息,因此不能产生准确的估计值。为此采用一种叫做Galerkin法的数学工具去构造建议分布,依据该方法构造的建议分布相对传统的方法提高了粒子滤波估计的准确性。同时,在新的跟踪算法框架中,将颜色模型和形状模型进行自适应的融合,并提出了一种新的模型更新方法,提高了目标跟踪的稳定性。实验结果证明了该跟踪算法的有效性。

In the particle filter framework, estimation accuracy strongly depends on the choice of proposal distribution. The traditional particle filter uses system transition probability as the proposal distribution without considering the new observing information; therefore, they cannot give accurate estimation. A new tracking framework applied with particle filter algorithm was proposed, which used Galerkin＇s method to construct proposal distribution. This proposal distribution enhanced the estimation accuracy compared to traditional filters. In the proposed framework, color model and shape model were adaptively fused, and a new model update scheme was also proposed to improve the stability of the object tracking. The experimental results demonstrate the availability of the DrODosed algorithm.