摘要
疲劳检测系统中的人眼检测误检率较高。为解决该问题,提出一种基于代价敏感支持向量机(CSVM)的人眼检测方法。对图像进行预处理,利用Gabor滤波器提取人眼特征向量,并使用主成分分析法实现降维,采用CSVM训练得到人眼和非人眼分类器,从而验证眼睛候选区域。实验结果表明,该方法能降低误检率,提高分类器的可靠性。
In order to improve the accuracy of eye detection in driver fatigue detection system, this paper presents an eye detection method based on Cost-sensitive Support Vector Machine（CSVM）. The feature vectors of eyes are extracted by Gabor filter after preprocessing, and its dimensionality is reduced by Principal Component Analysis（PCA）. The resultant feature vectors are used to train a CSVM to classify the eye and no eye, so as to verify the candidate eye area. Experimental results show that the proposed method can reduce the error rate and improve the reliability of the classifier.
出处
《计算机工程》
CAS
CSCD
2013年第9期254-257,276,共5页
Computer Engineering
基金
国家自然科学基金资助项目(60905034)
浙江省自然科学基金资助项目(Y1080950)
关键词
人眼检测
特征提取
拒识代价
支持向量机
敏感分类器
eye detection
feature extraction
rejection cost
Support Vector Machine（SVM）
sensitive classifier
