基于深度学习的面部动作检测  

Facial Motion Detection Based on Deep Learning

作　　者：杜虓龙 余华平[1] DU Xiao-long;YU Hua-ping(College of Computer Science,Yangtze University,Jingzhou 434023,China)

机构地区：[1]长江大学计算机科学学院,湖北荆州434023

出　　处：《软件导刊》2021年第5期29-33,共5页Software Guide

摘　　要：对汽车驾驶员进行疲劳检测,需首先利用面部检测技术对驾驶员的面部动作进行识别,可以使用MTCNN与CNN相结合的深度学习方法完成面部动作检测。先利用MTCNN的3个模块将面部特征与关键点进行提取,再利用CNN对网络进行训练,以准确识别特定的面部动作。利用MTCNN与CNN相结合的方法,模型准确率达99%,并且实时检测的FPS平均在19帧左右。研究表明,使用MTCNN与CNN相结合的深度学习方法,可以及时、准确地对驾驶人进行面部动作识别,为下一步疲劳检测打下良好基础。In order to be able to perform fatigue detection on car drivers requires that the facial movements of the drivers are first identified using facial detection techniques.In this paper,a deep learning method combining MTCNN and CNN is used to complete facial motion detection.MTCNN’s three modules first extract facial features and key points,and then use CNN’s pair network training makes The network can accurately recognize specific facial movements.The use of MTCNN in combination with CNN has resulted in a 99%accuracy of the model and the average real-time detection of FPS is around 19.It can provide timely and accurate facial motion recognition to the driver,laying a good foundation for the next step in fatigue detection.

关 键 词：机器学习 面部识别 神经网络 行为检测 卷积神经网络 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]