一种残差网络辅助的驾驶员视线定位方法  

A Residual Network Assisted Driver Eye Gaze Positioning Method

作　　者：靖丽君 李长勇[1] 庄力骏 JING Li-jun;LI Chang-yong;ZHUANG Li-jun(School of Mechanical Engineering,Xinjiang University,Xinjiang Urumqi 830047,China)

机构地区：[1]新疆大学机械工程学院,新疆乌鲁木齐830047

出　　处：《机械设计与制造》2023年第11期220-223,共4页Machinery Design & Manufacture

基　　金：新疆维吾尔自治区自然科学基金(2016D01C060)。

摘　　要：针对目前视线定位方法在驾驶员头部姿态和光照变化时,易出现定位精度低的问题,提出了一种残差网络辅助的驾驶员视线定位方法。首先在基础的残差网络模型上,改进残差网络结构来控制参数数量,采用激活函数FReLU代替ReLU,解决负值消失的问题;其次构建数据集,预训练检测模型,利用模型对驾驶员进行人脸检测,在人脸检测基础上使用dlib库提取面部特征点并计算头部姿态;最后根据头部姿态和虹膜中心位置信息融合确定视线方向。实验结果表明该方法降低了头部运动姿态和光照变化对视线定位精度的影响,能够实现头部活动范围直径在22cm内的视线定位,视线定位误差为0.7°。In order to solve the problem that the current eye gaze positioning method tends to have low positioning accuracy when the driver’s head posture and illumination change,a residual network assisted driver eye gaze positioning method was proposed.Firstly,on the basis of the basic residual network model,the residual network structure was improved to control the number of parameters,and the activation function FReLU was utilized to replace ReLU solve the problem of the disappearance of negative values.Secondly,a data set pre-training detection model was constructed,and the model was used to detect the driver’s face.On the basis of face detection,the dlib library was used to extract facial feature points to calculate the head posture.Finally,eye gaze was determined according to the fusion of the head posture Angle and the calculated pupil position information.The experimental results show that this method reduces the influence of the head movement posture and illumination changes on the eye gaze positioning accuracy,and can achieve eye gaze positioning within the diameter of the head movement range within 22cm,and the eye gaze positioning error is 0.7°.

关 键 词：头部姿态 视线定位 残差网络 激活函数 人脸检测 

分 类 号：TH16[机械工程—机械制造及自动化] TP391.41[自动化与计算机技术—计算机应用技术]