智能视频监控中基于肤色信息的人脸检测算法研究  

Research on human face detection algorithm based on skin color information in intelligent video surveillance

作　　者：黄知超 张鹏 赵华荣 赵文明 HUANG Zhichao;ZHANG Peng;ZHAO Huarong;ZHAO Wenming(School of Mechanical and Electrical Engineering,Guilin University of Electronic Technology,Guilin 541004,China)

机构地区：[1]桂林电子科技大学机电工程学院,广西桂林541004

出　　处：《现代电子技术》2018年第7期58-61,66,共5页Modern Electronics Technique

基　　金：广西教育厅项目(2015JGA202);研究生创新项目(2016YJCX107)~~

摘　　要：针对智能视频监控中人脸检测受复杂环境以及多姿态人脸的影响,采用一种基于肤色特征与Ada Boost算法相结合的方法,提取两种算法各自优点并加以优化,其主要思想是利用肤色特征建立肤色模型,选出含有人脸预检测肤色区域,进行人脸样本训练,提取人脸样本Haar特征,进行弱分类器训练,利用迭代的方法,再将不同的弱分类器组合成强分类器,最后形成级联分类器,运用级联分类器检测含有人脸的肤色区域。实验结果表明,该方法不仅提高了智能视频监控中人脸检测的效率和准确性,而且具有较好的鲁棒性。Since the human face detection in intelligent video surveillance is influenced by complex environment and multi-pose human face,a method based on skin color feature algorithm and AdaBoost algorithm is adopted to extract and optimize the respective advantages of the two algorithms.According to the main thought of the method,the skin color feature is used to estab-lish the skin color model,and select the pre-detecting skin color area containing human face for human face sample training.The Haar feature of human face sample is extracted to train the weak classifiers.The iterative method is used to combine different weak classifiers into a strong classifier,so as to form a cascade classifier,with which the skin color region containing human face is detected.The experimental results show this method can improve the efficiency and accuracy of human face detection in intelligent video surveillance,and has strong robustness.

关 键 词：人脸检测 肤色特征 AdaBoost算法 弱分类器训练 级联分类器 视频监控 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP319.4[电子电信—信息与通信工程]