基于疲劳驾驶的人眼定位方法研究

Research on eye localization method based on fatigue driving

人眼定位是疲劳驾驶的研究关键。由粗到精,先后进行了人脸检测、瞳孔定位。针对现有方向梯度直方图人脸检测算法泛化能力不佳的问题,提出了一种基于信息熵加权的HOG特征提取算法,该算法将待分类的人脸特征进行信息熵阈值加权,形成新的HOG特征,然后通过支持向量机进行分类;针对现有瞳孔定位算法准确率不高的问题,提出了多算法协同工作的瞳孔定位方法,以自商图为基准寻找二值分割点,实现了眼部区域光照不变性,以灰度积分投影为依据实现了瞳孔精确定位。实验结果表明,该文提出的人脸检测算法在CelebA验证数据集准确率可达到98.26%,较传统识别方法有更高的准确率;而瞳孔定位算法也可达到令人满意的精确度,提高了瞳孔定位的准确性。

Eye positioning is the key of fatigue driving research. In this paper, from rough to fine, face detection and pupil positioning were carried out successively. To solve the problem of poor generalization ability of existing face detection algorithms based on histogram of directional gradient, this paper proposes a HOG feature extraction algorithm based on information entropy weighting. This algorithm can calculate the threshold weight of face features to be classified to form a new HOG feature, and then classify it by support vector machine. Aiming at the problem that the accuracy of pupil positioning is not high, this paper proposes a method of pupil positioning that works in collaboration with multiple algorithms, searches for binary segmentation points based on the self-quotient chart, realizes the illumination invariance of the eye region, and realizes the accurate pupil positioning based on the gray integral projection. Experimental results show that the face detection algorithm proposed in this paper can achieve 98. 26 % accuracy in CelebA validation data set, which is higher than the traditional recognition method. The pupil positioning algorithm can also achieve satisfactory accuracy, and improve the accuracy of pupil positioning.