基于Adaboost算法的人眼检测中样本选择研究

Research of Samples Selection in Eye Detection Based on Adaboost Algorithm

人眼检测在表情识别和人脸识别中起着非常重要的作用,是驾驶员疲劳检测的基础。采用了基于Adaboost算法的人眼检测的方法,训练阶段中的样本选择是Adaboost算法的关键,分析和讨论了训练阶段不同特征的正、负样本对最终检测结果的影响,提出了一种新型的负样本选择方法,并实验得到了各种样本训练生成的分类器对人脸库的检测率和误检率,得出用去除眼睛部分余下的人脸作为负样本训练出来的分类器能有效降低误检率,为以后的眼睛分类器训练提供了实验依据。

Eyes detection plays a very important role in the facial expression recognition and face recognition,it is basic of driver fatigue detection. Use a method based on Adaboost algorithm for eyes detection. ,Samples selection is the key for Adaboost algrithm. Different characteristics of the positive and negative samples in the training phase are analyzed and discussed. A novel feature of the negative samples is proposed, generate the detection rates and the false rate of different classifers to face database,and figure out that classifer which is generated by the remaining part of face after removing eyes as negative samples can reduce false rate effectively. It provides the experimental basis for training eye classifier in the future.