文献类型：期刊文献

中文题名：基于多尺度池化卷积神经网络的疲劳检测方法研究

英文题名：Driver’s fatigue detection system based on multi-scale pooling convolutional neural networks

作者：顾王欢[1];朱煜[1];陈旭东[1];郑兵兵[1];何林飞[1]

机构：[1]华东理工大学电子与通信工程系

年份：2019

卷号：36

期号：11

起止页码：3471

中文期刊名：计算机应用研究

外文期刊名：Application Research of Computers

收录：CSTPCD;;北大核心:【北大核心2017】；CSCD:【CSCD_E2019_2020】；

基金：上海市科学技术委员会科研计划资助项目(17DZ1100808,17DZ1100803)

语种：中文

中文关键词：视觉特征分析;多尺度池化;卷积神经网络;疲劳检测;人脸检测

外文关键词：visual feature analysis;multi-scale pooling;convolutional neural network;fatigue detection;face detection

摘要：针对视觉特征分析疲劳检测问题,设计了一种级联深度学习的检测系统结构,并提出基于多尺度池化的卷积神经网络疲劳状态检测模型。首先通过深度学习模型MTCNN进行人脸检测,提取出眼睛和嘴巴区域;针对眼睛和嘴巴的状态表征和识别问题,提出一种基于ResNet的多尺度池化模型(MSP)对眼睛和嘴巴状态进行训练;实时检测时,将眼睛嘴巴区域通过训练好的卷积神经网络模型进行状态识别,最后基于PERCLOS和提出的嘴巴张合频率(FOM)对驾驶员进行疲劳判定。实验结果表明,该算法具有较高的检测准确率,同时满足实时性要求,且对复杂环境具有较高的鲁棒性。
This paper proposed a hierarchical convolutional neural network model with multi-scale pooling for vision-based fatigue detection system.The first step is face detection and extraction of eye and mouse regions by deep learning model:MTCNN.In order to solve the problem of characterization and recognition of eye and mouth regions,this paper proposed a multi-scale pooling model(MSP)based on ResNet to train different states of eye and mouth.In real-time detection,the system recognized the states of eye and mouth by the pre-trained convolutional neural network model.Finally,it detected fatigue through the PERCLOS and the frequency of open mouth(FOM).The experimental results show that the proposed algorithm has high detection accuracy,real-time performance and high robustness to complex environments.