实时视频图像中的人脸检测与跟踪

视频图像目标检测与跟踪是远程协作系统中感兴趣的研究课题之一。文中提出了一种协同系统中视频序列图像人脸检测及实时跟踪的方法。该方法根据用户选定的目标(如人脸)的颜色分布特点，用多幅训练样本图像建立人脸肤色模型，然后根据该模型和人脸特征对待检测的彩色图像进行分割与匹配，从而确定候选区域是否人脸。在视频图像跟踪中用此方法可实现人脸的实时检测跟踪，为了提高跟踪速度，提出了改进的基于运动预测的快速跟踪法。该方法充分利用运动连续性规律，能较好地处理多干扰目标同时出现的情形。实验表明所提出的方法执行效率高，检测跟踪正确率高．对有旋转的非正面人脸图像也有较好的适应性。

Real-time Face Detection and Tracking in Video Sequences

Object detection and tracking in video sequences is becoming one of the research focuses in the field of remote collaborative system. This paper presents a real-time face detection and tracking method that can be applied in CSCW system. According to skin-color distribution in the color space, this method develops a statistical skin-color model through interactive sample training. With this model, all candidate regions are segmented. The presence or absence of a face in each region is verified by means of feature matching. Real-time detection and tracking can be achieved by using this method in video sequences. In order to speed up tracking, the traditional method is improved by adding motion prediction, which uses the rules of continuity and works efficiently when several disturbing objects appear simultaneously. Experiment results show the proposed method has not only high speed and efficiency, but also robust performance to head rotation.