一种基于多信息综合的人脸跟踪算法  

A Face Tracking Method Based on Multiple Information Combination

作　　者：冯春来[1] 朱虹[1] 李钦[1] 钱学明[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》2004年第23期98-101,共4页Computer Engineering and Applications

摘　　要：随着智能监视系统的发展需求,人脸跟踪作为智能监视系统中的一类特殊对象的跟踪方法,在跟踪时对环境的适应性要求很高。文章提出了一种基于多信息综合特征的,在变动场景下,对人脸进行跟踪的算法。为了使人脸跟踪在环境发生变化的情况下,可以进行有效跟踪,算法中用人脸区域内部的颜色信息、亮度梯度信息等获得综合的判别特征。在跟踪时,对于人脸的大小、朝向的变化以及光线变化,背景物等的干扰均有较强的鲁棒性。实验结果表明,算法在摄像机转动、仰俯、光线变化、人头晃动以及背景颜色分布与人脸近似等情况下均可完成有效跟踪。With the development of VR technology,face tracking as one of the methods in human computer interface technology,its effectiveness is a deep researching problem.This paper presents a multiple information combination based method which can be used successfully in continuously shifting scene.To track face effectively in the case of varying backgrounds,information such as color distribution,luminance gradient in face region is used to obtain comprehensive characters for discrimination.It's robust to the change of face scalar,direction and luminace and background disturbance.Experimental results show that the method can achieve effective face tracking under the condition of camera rotation in horizontal direction or camera translation in vertical direction,variety of illumination,movement of head,similar color distribution of background and face region.

关 键 词：人脸跟踪 颜色直方图 椭圆梯度 梯度直方图 智能监视系统 颜色分布信息 结构分布信息 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]