目标跟踪中的改进Monte Carlo滤波算法

摘要

传统的Monte Carlo滤波算法在目标跟踪过程中存在严重的采样贫瘠问题,这直接导致了样本集的退化.为了解决这个问题,提出一种改进的Monte Carlo滤波算法.在样本集建立阶段,采用基于视觉机制的方法建立样本集合,使得样本集在与中心距离较近的地方密集,在离中心较远的地方稀疏,这样的样本集合建立方法能够更准确地反映人眼对事物的感知;在样本集传播阶段,获得一个区分样本优劣的阈值,将样本集合分为优劣两种,用重采样的方法对优样本集合采样,采样半数样本,用随机抽样的方法补充其余半数样本,实验结果表明,这种方法可以很好地解决样本退化的问题.%Traditional Monte Carlo filter for target tracking has the problem of sample barren, which leads to samples degraded directly. To solve this problem, an improved Monte Carlo filter has been proposed. During the initial stage, the samples are built based on vision method, more samples which are near the center are got, and fewer samples which are far from the center are got. During the samples transmission process, half of the samples are re-sampled. The other half of samples are transmitted directly. Experimental results show that this method deals with sample degraded problem well.