基于改进YOLOv7的疲劳驾驶检测算法研究

Research on Fatigue Driving Detection Algorithm Based on Improved YOLOv7

疲劳驾驶是一种常见的危险驾驶行为,准确检测驾驶员是否处于疲劳驾驶状态对于道路安全至关重要。提出一种基于YOLOv7的疲劳驾驶检测算法,该方法使用改进的YOLOv7模型识别驾驶员的眼部、嘴部疲劳状态,再通过计算眼睑纵横比(Eye Aspect Ratio,EAR)、嘴巴纵横比(Mouth Aspect Ratio,MAR)、头部3个欧拉角来判别驾驶员是否疲劳驾驶。实验结果表明,使用MobileOne轻量化的网络作为主干网络,使模型检测速度达到每秒71帧,在Neck部分引入卷积块注意力模块(Convolutional Block Attention Module,CBAM)和Focal EIoU Loss损失函数,在基本不影响速度的情况下使检测精度达到95.35%。本方法在公开数据集NTHU-DDD上取得了较好的疲劳检测效果,可应用于实际场景中的实时安全监测。

The issue of drowsy driving,a common yet perilous behavior,has garnered widespread attention.Therefore,accurately detecting whether a driver is in a drowsy state is crucial for road safety.This paper presents a drowsy driving detection algorithm based on YOLOv7.The approach utilizes an enhanced YOLOv7 model to identify the fatigue status of a driver's eyes and mouth.It further determines whether the driver is drowsy by computing the Eye Aspect Ratio(EAR)and Mouth Aspect Ratio(MAR)while also considering the three Euler angles of the head.Experimental results demonstrate that using the lightweight MobileOne network as the backbone,the model achieves a detection speed of 71 frames per second.By introducing the Convolutional Block Attention Module(CBAM)in the Neck section and the Focal EloU Loss function,the model attains a detection accuracy of 95.35% without significantly compromising speed.The proposed method exhibits promising drowsy driving detection performance on the publicly available NTHU-DDD dataset and can be applied to real-time safety monitoring in practical scenarios.