彩色序列图像中的人脸检测与跟踪  

Face detection and tracking in color image sequences

作　　者：李久贤[1] 夏思宇[1] 袁晓辉[1] 夏良正[1] 

机构地区：[1]东南大学自动控制系,南京210096

出　　处：《计算机应用研究》2007年第8期189-192,共4页Application Research of Computers

基　　金：浙江省长三角联合攻关项目(2005E60007)

摘　　要：提出了一种彩色视频序列图像中的人脸检测与跟踪方法。该方法将人脸检测与人脸跟踪有效地结合在一起,采用Condensation滤波跟踪算法对区域进行跟踪,在跟踪过程中提出引入基于支持向量机的人脸置信度,样本的置信度随时间进行更新,人脸检测的结果基于置信度的后验概率。同时,该方法对Condensation滤波跟踪算法作了改进,在跟踪过程中采用了基于Metropolis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对人脸自由运动的跟踪,且精度较高。实验结果表明,该方法有效地解决了复杂背景中人脸姿态变化情况下的人脸检测与跟踪问题,与静态人脸检测相比有更好的检测效果。Human face detection and tracking received extensive attention because of the potential applications in many fields, such as video coding, surveillance and human-computer interface. The aim of this paper was to detect and to track human face in color image sequences effectively. An approach to simultaneous detection and tracking in video images was presented. The approach was based on posterior estimation using conditional density propagation. A face identity score was obtained for each image position and at several scales and posed based on support vector machine classifier. The identity scores were propagated over time using Condensation filter. Additionally, a simple and efficient sampling method based on the Metropohs algorithm and an adaptive dynamical model were used in Condensation algorithm for robust face tracking. Experimental results show that this approach could detect and track human face＇ s free movement in videos containing complex backgrounds, and be more effective compared with still-to-still face detection.

关 键 词：模式识别 人脸检测 人脸跟踪 Condensation滤波 支持向量机 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]