基于粒子滤波的小波特征跟踪方法研究  

Wavelet Feature Tracking Method Based on Particle Filter

作　　者：姚剑敏[1] 孙俊喜[1] 孙中森[1] 宋建中[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所,吉林长春130033

出　　处：《计算机仿真》2006年第1期94-95,304-305,共4页Computer Simulation

摘　　要：该文提出了基于粒子滤波的小波特征跟踪方法。粒子滤波基于蒙特卡罗模拟方法来实现递推贝叶斯滤波,是一种实用的后验概率求解方法。文中研究了目标的Gabor小波网络表示,用一定数量的小波构成一个集合来表示目标特征,各小波的参数由优化方法来确定。构建了基于粒子滤波的跟踪框架,每个粒子表示一种Gabor小波网络的可能形式,并计算与当前图像的相似度。粒子权值与相似度成正比,目标状态的后验概率由粒子加权表示。与传统的“峰值”跟踪方法不同,粒子滤波具有“多峰”的跟踪形式,并结合对光照、噪声不敏感的小波表示形式,具有较强的抗局部遮挡能力。Wavelet feature tracking method based on particle filter is proposed. Particle filter aims to implement Bayesian filter based on Monte Carlo simulation, and it is a practical method for computing posterior PDF （ probability density function）. Gabor wavelet network （GWN） is used to describe the features of the object. The description is robust to illuminate variation and noise. The parameters of the wavelets are computed by optimization procedure. The tracking framework is based on particle filter and each particle figures one kind of possible GWN. The weights of the particles axe proportional to the correlation values and the posterior PDF of the object is denoted by the sum of the weighted particles. Different from the＂ peak -tracking＂ based method, tracking with particle filters is robust as a result of multimodality. Combined with GWN description, the method has strong ability to track under local obstruction.

关 键 词：粒子滤波 小波网络 视频跟踪 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]