基于改进粒子滤波算法的人眼跟踪方法  

Eye tracking method based on improved particle filtering algorithm

作　　者：郭君斌[1] 郭晓松[1] 雷磊[1] 薛冰[1] 

机构地区：[1]第二炮兵工程学院,西安710025

出　　处：《仪器仪表学报》2010年第8期1720-1725,共6页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(60675019)资助项目

摘　　要：针对粒子滤波计算量大、粒子匮乏等问题,提出了2种策略来改进粒子滤波算法,一方面基于Galerkin投影法产生良好的粒子滤波参考分布函数;另一方面将BP神经网络引入粒子滤波,提出了一种提高粒子多样性的重采样策略。在此基础上,将改进的粒子滤波算法应用于人眼跟踪问题,运用颜色和纹理复合的观测模型,实现了人眼的准确跟踪。实验结果表明,改进算法有效提高了粒子滤波的估计精度和运算速度,避免了粒子退化和样本匮乏现象。To resolve the crux problems of particle filtering （PF）,such as great computational workload and sample impoverishment,two strategies are utilized to improve PF algorithm.One is using Galerkin projection method to produce proposal distribution function （PDF）;the other is utilizing back propagation neural network in particle filtering.A resampling algorithm is introduced,which can improve particle diversity and avoid degeneracy phenomena.Then the improved PF algorithm is applied to eye tracking.The likelihood model of PF algorithm combines color and texture and implements precise eye tracking.Experiment results show that the improved algorithm effectively enhances the estimation accuracy and computation speed,and can avoid the degeneracy of particles and impoverishment of samples effectively.

关 键 词：参考分布 重采样 人眼跟踪 粒子滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]