基于肤色与Haar-like扩展集的驾驶员人脸检测算法  

Face Detection Algorithm Based on Skin Color and Haar-like Features Extension Set

作　　者：葛小凤 GE Xiao-feng(School of Electronic Information Engineering,China West Normal University,Nanchong 637009,China)

机构地区：[1]西华师范大学电子信息工程学院,四川南充637009

出　　处：《软件导刊》2020年第8期30-34,共5页Software Guide

基　　金：西华师范大学青年教师科研项目(416829)。

摘　　要：针对疲劳驾驶预警系统中人脸检测准确率低、误检率高的问题,提出一种基于肤色与Haar-like扩展集的驾驶员人脸检测算法。首先根据驾驶员人脸肤色在YCbCr色彩空间的聚类性、脸部特征及驾驶环境,筛选人脸肤色作为候选区;然后在传统基于Haar-like特征的AdaBoost算法中,加入两组新的符合人脸特征分布的Haar-like特征进行驾驶员人脸检测。以MIT人脸库和拍摄的驾驶员人脸图像作为训练与检测样本,与传统AdaBoost算法进行对比实验。结果表明,该算法对正面人脸和侧面人脸(倾斜角度小于45°)检测准确率分别提高1.25%和5.00%,误检率降低2.81%和4.50%,人脸检测准确率得到较大提高。Aiming at the problems of low accuracy and high false detection rate of face detection in fatigue driving warning system,this paper proposes a driver face detection algorithm based on skin color and Haar like extension set.Firstly,according to the clustering of drivers’face skin color in YCbCr color space,facial features and driving environment,human skin color is selected as candidate re⁃gion;then,based on AdaBoost algorithm of traditional Haar like feature,two new Haar like features are added to detect the driver’s face.The MIT face database and the driver’s face images taken as training and detection samples are compared with the traditional Ad⁃aBoost algorithm.The experimental results show that the detection accuracy of the algorithm is improved by 1.25%and 5.00%respec⁃tively for front face and side face(tilt angle is less than 45°),and the false detection rate is reduced by 2.81%and 4.50%,which proves the face detection accuracy is greatly improved.

关 键 词：人脸检测 HAAR-LIKE特征 肤色分割 ADABOOST算法 级联分类器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]