基于卷积神经网络的车载疲劳驾驶检测系统的实现  

Implement of vehicle fatigue driving detection system based on convolutional neural network

作　　者：唐杰[1] 陈仁文[1] 余小庆[1] Tang Jie;Chen Renwen;Yu Xiaoqing(Mechanical Structural Laboratory, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China)

机构地区：[1]南京航空航天大学机械力学结构实验室

出　　处：《国外电子测量技术》2018年第1期116-121,共6页Foreign Electronic Measurement Technology

摘　　要：设计并实现一套基于Android平台的人脸疲劳检测系统。该系统利用Android设备的外接OTG接口摄像头获取驾驶员脸部视频,通过Android4.4之后自带的人脸检测算法快速检测人脸,然后通过深度卷积网络训练的模型进行人脸关键点定位,粗定位出人眼和嘴部范围,利用最大类间方差法Otsu对人眼及嘴部进行目标提取然后通过最小二乘法椭圆拟合人眼和嘴部轮廓,从而进行状态分析,最后利用PERCLOS原理实时的检测出疲劳状态。在板车上实验结果表明,该系统对光照具有较好的鲁棒性,并且可以极大的降低事故率,保障了驾驶员的安全。A face fatigue detection system based on Android platform was designed and implemented.The system uses Android equipment with OTG camera to get through the driver＇s face video,OpenCV face detection algorithm is used to detect face quickly.Then,the system uses the depth convolution networks to located the human eye and mouth coarsely and uses Otsu method to extract the target from the human eye and mouth,and then the contour of the human eye and mouth is fitted by the least square ellipse method,thus the state analysis is carried out.Finally,the system uses PERCLOS principle to detect fatigue state in real time.The experimental results on the vehicle show that the system has better robustness to illumination,and can greatly reduce the accident rate and ensure the safety of the drivers.

关 键 词：疲劳检测 人脸关键点定位 轮廓定位 状态分析 

分 类 号：TN911.73[电子电信—通信与信息系统]