基于粒子滤波的人脸跟踪算法  

Human face tracking algorithm based on particle filter

作　　者：杨向波[1] 朱虹[1] 邓颖娜[1] 刘薇[1] 张曈[1] 

机构地区：[1]西安理工大学自动化与信息工程学院,西安710048

出　　处：《计算机工程与应用》2008年第23期209-211,共3页Computer Engineering and Applications

摘　　要：人物跟踪技术是目前智能监控系统的核心方法之一,针对人脸运动的非线性非高斯的特点,引入粒子滤波算法来进行运动预测估计,抵抗遮挡干扰。同时,根据人脸结构特点,提出了一种分块颜色直方图,用以描述人脸的特征。并且根据预测精度对预测过程中目标运动速度和过程噪声方差进行自适应更新。实验结果表明,在人脸的旋转,肤色和部分遮挡影响下跟踪精度较高,抵抗光照环境变化,以及人脸大小变化等的鲁棒性较强。Human tracking technique is one of core method in intelligence surveillance system.The human face moving is non- linear and non-gauss,particle filter algorithm is proposed for motion estimation,which is efficient even in occlusion situation.According to the face structure,a blocking color histogram is proposed to describe the human face character.The speed of object moving and variance of process noise are update adaptively according to estimate precision.Experimental results show that the tracking precision is high even under face rotation,complexion and part occlusion.It has strong robustness to resist illumination change and human face size change.

关 键 词：人脸跟踪 粒子滤波 分块颜色直方图 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]