彩色视频序列图像中的人脸跟踪方法  

Face Tracking in Color Image Sequences

作　　者：夏思宇[1] 夏良正[1] 金立左[1] 

机构地区：[1]东南大学自动控制系,南京210096

出　　处：《中国图象图形学报》2006年第9期1249-1254,I0001,共7页Journal of Image and Graphics

基　　金：浙江省长三角联合攻关项目(2005E60007)

摘　　要：针对彩色视频序列图像的人脸检测,提出了一种基于肤色的人脸跟踪方法。该方法首先在Hsu提出的肤色模型基础上,采样一种自肤色分割算法来提取复杂背景下人脸的肤色特征,与传统的采用固定肤色模型的检测算法相比,该方法具有更好的检测效果;然后,在人脸跟踪过程中采用Condensation滤波跟踪算法,并对算法做了两点改进,即在跟踪过程中采用基于M etropolis算法的重采样方法以及自适应的动态模型,实现了复杂背景下的人脸自由运动的跟踪,并从各种影片中截取了彩色视频序列图像进行了测试实验。实验结果表明,该方法有效地解决了复杂背景下人脸自由运动、光照变化及部分遮挡的问题,且精度较高。Human face tracking has received extensive attention because of its potential applications in many fields, such as video coding, surveillance and human-computer interface. The aim of this paper is to track human face in color image sequences effectively. A method using human skin feature integrated with Condensation （conditional density propagation） algorithm is proposed. At first, this paper presents a method of self-skin segmentation based on Hsu R L＇ s skin color model, to detect human skin in color image correctly and effectively. Compared with conventional skin color model methods, this method has more precise detection performance in varying lighting condition. Additionally, this paper presents two improvements for Condensation algorithm： a simple and efficient sampling method based on the Metropolis algorithm and an adaptive dynamical model are used in Condensation algorithm for robust face tracking. Experimental results show that in complex backgrounds and occlusion case, this approach could track human face＇ s free movement.

关 键 词：彩色视频序列 人脸跟踪 Condensation滤波 自肤色分割 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]