基于轻量化卷积神经网络的疲劳驾驶检测方法研究
Research on fatigue driving detection method based on lightweight convolutional neural network
刘鹏宇 1杨潞霞 1戴斌2
作者信息
- 1. 太原师范学院计算机科学与技术学院,晋中 030619
- 2. 运城学院计算机科学与技术系,运城 044000
- 折叠
摘要
交通安全问题与每个人的生活都密不可分.针对目前流行的检测模型缺乏实时性和准确性的问题,提出一种轻量化检测模型,对目标检测网络YOLOv4进行改进,主干特征提取网络使用MobileNet-V3.为使网络参数大幅度下降,通过采用深度可分离卷积来进一步减少参数量;通过给加强特征提取网络添加注意力机制,进一步优化检测速率和检测准确率,且利用一种数据增强方式来增加检测准确度.最后提取面部疲劳特征,由PERCLOS判断是否处于疲劳.通过实验得出检测准确率为92.90%,mAP指标为85.75%,参数量减少近95%,单帧检测速度是25.64 ms,可以平衡网络的准确性与实时性.
Abstract
Traffic safety issues are closely related to everyone's life.This article proposes a lightweight detection model to ad-dress the current lack of real-time and accuracy issues in popular detection models.The object detection network YOLOv4 is im-proved,and the backbone feature extraction network uses MobileNetV3.To significantly reduce the network parameters,a deep separable convolution is adopted to further reduce the number of parameters.By adding attention mechanisms to the enhanced fea-ture extraction network,further optimizing detection speed and accuracy.And by integrating Mosaic data augmentation methods,the detection accuracy of the model is ensured.Finally,the improved network in this article is used to extract facial fatigue features,and PERCLOS determines whether fatigue is present and outputs the results.Through experiments,it was found that the detection accuracy is 92.90%,the mAP index is 85.75%,the parameter quantity is reduced by nearly 95%,and the single frame detection speed is 25.64 ms,which can balance the accuracy and real-time performance of the network.
关键词
轻量化网络/疲劳驾驶/注意力机制/数据增强Key words
lightweight network/fatigue driving/attention mechanism/data augmentation引用本文复制引用
出版年
2024
NETL
NSTL
万方数据