基于人脸关键点的疲劳驾驶检测研究  

Research on Fatigue Driving Detection based on Face Key Points

作　　者：黄家才[1] 旷文腾 毛宽诚 

机构地区：[1]南京工程学院自动化学院,江苏南京211167 [2]河海大学能源与电气学院,江苏南京211100

出　　处：《南京工程学院学报（自然科学版）》2017年第4期8-13,共6页Journal of Nanjing Institute of Technology(Natural Science Edition)

基　　金：国家自然科学基金项目(61104085);江苏省自然科学基金项目(BK20151463)

摘　　要：为解决疲劳驾驶检测中人眼状态识别的难点,提出一种基于人脸关键点的疲劳检测方法.首先从摄像头实时采集驾驶员图像;然后经过Ada Boost方法检测出图像中最大的人脸,并完成人眼定位和人眼睁闭状态判断;最后使用PERCLOS算法判别驾驶员的疲劳状态.其中人眼睁闭状态判定采用基于Dlib库的关键点检测方法,通过计算上下眼睑的距离判断人眼睁闭状态.采集了1 000组人眼状态的数据之后进行分析,在闭眼时99%的状态值都小于2,而眼睛全睁时99%的状态值都大于2.4,当状态阈值设置为2时基于人脸关键点的检测算法正确率最高.试验结果表明,基于人脸关键点的检测算法在头部倾斜等姿态下和光线不好的情况下仍能很好地检测出驾驶员的疲劳状态,具有较强的鲁棒性.To solve the problem with difficulty in recognizing conditions of human eyes paper proposes an approach based on facial key points. Firstly, images from cameras are captured the largest face of the images is detected by using AdaBoost face detection method, eyes’ position is located, and judgment of whether the eyes are open or closed are completed; finally, PERCL0S （percentage of eyelid closure over the pupil over time） algorithm is applied to determine drivers’ fatigue state. Judgmentofeyes’stateisacquiredbymeansofDlib-library- based key point detection method, to calculate the distancc between upper and lower eyelids to determineeyes are open or closed. Analysis of the data of 1000 groups of human eyes shows that 99% of the state values are less than 2 in the closed eyes, 99% of the state values greater than 2. 4 when the eyes are fully open; when the state threshold is set to 2, the detection rate based on face key points is the highest. The resulon face key points is highly robust, which can detect drivers’ fatigue state even under the light.

关 键 词：疲劳检测 人眼状态 Dlib 人脸关键点 

分 类 号：TP274.4[自动化与计算机技术—检测技术与自动化装置]