双层结构Adaboost健壮分类器用于人眼精确定位

Robust classifier based two-layer Adaboost for precise eye location

提出了一种双层结构的Adaboost分类器用于眼睛的定位检测和跟踪。双层眼睛分类器由训练的双眼区域和单眼区域的分类器级联构成一个强分类器。该算法较传统的YCbCr色度空间眼睛模板而言,对光照变化有更大的适应性。相对普通的Adaboost眼睛分类器,该算法保留了原有普通Adaboost分类器的高检测率,同时有效降低了眼睛的误检率。通过研究训练样本数,训练级数和Adaboost分类器误检率的关系,分类器训练效率得到提高。

A two-layer eye classifier for eye detection was proposed. Two layers, double-eye layer and single-eye layer, had been trained and cascaded into a strong one for eye detection. Two-layer classifier was more robust in illumination invariance eye detection compared with YCbCr space eye map algorithm. Also, it kept the same detection rate as the commonly trained Adaboost eye classifier with a much lower error detection rate. Relationship among stages, training sample number and error detection rate had been analyzed to facilitate the training procedure.