基于人脸肤色和特征的实时检测跟踪算法.pdfVIP

第 34 卷第 4 期 光电工程 Vol.34, No.4 2007 年 4 月 Opto-Electronic Engineering April, 2007 文章编号：1003-501X(2007)04-0044-05 基于人脸肤色和特征的实时检测跟踪算法 向桂山，王宣银，梁冬泰 ( 浙江大学 流体传动及控制国家重点实验室，浙江 杭州 310027 ) 摘要：针对实时视频中的运动人体目标，提出一种基于人脸肤色和特征的快速人脸检测和跟踪方法。首先运用帧 差法和形状信息检测出视频中的运动人体目标；然后在 YC C 色彩空间中，根据肤色的色度聚类特性建立 Gaussian b r 模型，分割出肤色区域，去除噪声后，结合人脸的几何特征和器官独有的颜色特征滤除非人脸肤色区域，准确定 位人脸在图像中的位置；利用控制策略驱动摄像机，根据人脸信息使人的头肩部位始终处于视频图像的中心，从 而实现运动跟踪。为了增强系统对光线变化的适应性，提出了适当的 Gaussian 模型参数更新策略。实验结果表明， 该算法能够适应复杂背景下的人脸检测，具有速度快、准确率高、鲁棒性好的特点，实现了运动人脸的可靠跟踪。 关键词：运动检测；肤色模型；人脸检测；目标跟踪 中图分类号：TP391 文献标识码：A Real-time detection and tracking algorithm based on the color and character of human face XIANG Gui-shan ，WANG Xuan-yin ，LIANG Dong-tai ( The State Key Lab of Fluid Power Transmission and Control, Zhejiang University, Hangzhou 310027, China ) Abstract : For fast detecting and tracking human face in real-time video, a method based on the color and character of face was proposed. Moving human-body objects were extracted from video according to temporal differencing and shape information firstly, and then the skin area was segmented based on its Gaussian model established due to the clustering character of skin color in YC C color space. After noise was suppressed, the non-face skin area was eliminated according b r to the geometric characters of face and the color characters of organs in face, and then the face was located in image. According to the face location, the head and shoulder part were centered in video by driving camera using cont