空间信息码本和粒子滤波相结合目标跟踪算法

摘要

In order to avoid tracking failure caused by the errors easily accumulated in particle filter tracking process,we proposed an algorithm particle filter tracking,it is based on spatial-information CodeBook background modelling.First,in the process of codebook background modelling,it combines the CodeBook of targets’ pixels and the CodeBook of its 8-neighborhood pixels to overcome the interference of background noise and to get precise foreground targets;Then,it extracts the weighted kernel function colour features of foreground regions as the prior distribution of initial state of particle filter,meanwhile imports the weighted improvement of position information to resampling progress.Experimental results show that this algorithm reduces the sampling errors caused by particle emission,and can effectively track targets under complex background as well,moreover,it is superior to the standard particle filter algorithm in real-time property and accuracy.%针对粒子滤波跟踪过程中容易积累误差引起跟踪失败的问题，提出一种基于空间信息码本背景建模的粒子滤波跟踪算法。首先，在码本背景建模过程中将目标像素码本和周围8邻域像素码本进行融合，克服背景噪声干扰得到精确的前景目标；然后提取前景区域的核函数加权颜色特征作为粒子滤波初始状态先验分布。同时在重采样过程中引入位置信息加权改进。实验结果表明，该算法减少了粒子发散引起的采样误差，且能够在复杂背景下对目标进行有效跟踪，在实时性和准确性上优于标准粒子滤波算法。