田全 王蕴哲

摘要:针对眉毛和眼镜框等区域对人眼定位造成较大影响的问题,根据人眼周围区域各部位灰度值变化特征,提出了一种新的人眼定位及开闭状态识别算法。该算法利用AdaBoost算法对人眼进行粗定位,通过图像水平积分法,根据积分极值点特征完成人眼特征的精确定位,实现人眼部位条状区域的分割。最后利用投影法得出人眼的开闭状态。实验结果表明,该文提出的方法能够有效准确的实现人眼部位的精确定位,对人眼开闭状态识别率能够满足实际需求,且对图像中的各种干扰因素具有较强的鲁棒性。
关键词:人眼定位;人眼状态识别; AdaBoost算法;水平积分投影法
中图分类号:TP31 文献标识码:A 文章编号:1009-3044(2018)03-0170-03
Abstract: Aiming at the problem that eyebrow frame and other areas have great influence on human eye location, a new eye location and state recognition algorithm is proposed Based on the change of gray value in different parts of the human eye. The algorithm uses AdaBoost algorithm to locate the human eye roughly, and completes the precise location of human eye Based on the integral extremum point feature by image horizontal integration method, and realizes the segmentation of human eye part. Finally, by using the projection method of opening and closing state of the eye. Experimental results show that the proposed method can accurately and accurately locate human eyes, and it can meet the actual demand of human eye opening and closing state, and is robust to all kinds of interference factors in images.
Key words: eye location; eye state recognition; AdaBoost algorithm; projection method
1 概述
在車辆辅助驾驶系统中,人眼作为人脸特征的关键特征,其定位精度和开闭状态判断的准确度直接影响系统对于驾驶员行车过程中分神或者疲劳状态检测的准确性[1]。目前人眼定位算法已较成熟,常用方法有以下几种:一是基于形状的人眼定位方法,如Hough变换法,利用瞳孔或虹膜的圆形特征,以眼睑或眼角的几何特征加以辅助,能够实现人眼的精确定位[2]。该方法常用于安检口身份验证及手机解锁等场合,通常需要指定被测者头部保持固定的姿势,同时,该方法对测量设备精度要求较高,相应的增加了测量成本,同时,在被测者半闭眼或者完全闭眼情况下,无法完成人眼的定位;二是基于模板匹配进行人眼定位,该方法利用选定的模板图像与待测图像各区域进行比较,选出相似度最高的区域作为匹配结果[3]。……