基于肤色与肤色矩实时视频人脸检测与跟踪  

Face detection and tracking based on skin color and skin color moment

作　　者：郑明恩[1] 管业鹏[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200072

出　　处：《计算机工程与设计》2008年第4期879-881,共3页Computer Engineering and Design

基　　金：上海市教委与教育发展基金曙光项目(04CX72);上海市教委青年基金项目(05AZ38)

摘　　要：提出了一种新颖、鲁棒,实时人脸检测与跟踪算法。该方法采用背景差分法提取运动区域,对运动区域利用肤色归一化RGB和HSV色彩模型的聚类性,得到人脸候选区域。利用人脸几何信息和孔洞信息对人脸候选区域进行验证。基于肤色矩特性,对人脸区域进行跟踪与预测。通过对不同背景条件下的人脸检测与跟踪,实验结果表明,所提算法不仅检测率高,且对光照、人脸姿态的变化具有较强的鲁棒性。基于480×360图像处理速度平均为25帧/秒,可满足系统实时性要求。Novel and robust algorithm for face detection and tracking is presented. Moving regions are extracted based on background subtraction at first. Candidate face regions are extracted according to skin-color normalized RGB and HSV color model based on the extracted motion region. Face information with certain size and some holes is used to verify the face candidates. The facial region is tracked and estimated based on skin color moment characteristics. The experimental results have shown that the proposed method not only attains high face detection rate, but also is robust to illumination and face pose variations. The processing speed is about 25f/s on an average across 480×360 images, which meets the demand of real time system for face detection and tracking.

关 键 词：人脸检测 跟踪 提取 背景 肤色 肤色矩 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]