基于改进YOLOv8n-Pose的疲劳驾驶检测  

Fatigue driving detection based on improved YOLOv8n-Pose

作　　者：蔡忠祺 林珊玲 林坚普 吕珊红 林志贤[1,2] 郭太良 CAI Zhongqi;LIN Shanling;LIN Jianpu;LÜShanhong;LIN Zhixian;GUO Tailiang(School of Advanced Manufacturing,Fuzhou University,Quanzhou 362251,China;Fujian Science and Technology Innovation Laboratory for Photoelectric Information,Fuzhou 350116,China)

机构地区：[1]福州大学先进制造学院,福建泉州362251 [2]中国福建光电信息科学与技术实验室,福建福州350116

出　　处：《液晶与显示》2025年第4期617-629,共13页Chinese Journal of Liquid Crystals and Displays

基　　金：国家重点研发计划(No.2021YFB3600603)。

摘　　要：针对目前驾驶员疲劳检测算法存在检测过程复杂、参数多、精度低、运行速度慢等问题,提出了一种基于改进YOLOv8n-Pose的轻量级模型。该模型优化了YOLOv8n-Pose的结构。首先,在模型主干网络中,引入Ghost卷积减少模型参数量和不必要的卷积计算。其次,引入Slim-neck融合主干网络提取的不同尺寸特征,加速网络预测计算。同时在颈部网络添加遮挡感知注意力模块(SEAM),强调图像中的人脸区域并弱化背景,改善关键点定位效果。最后,在检测头部分提出一种GNSC-Head结构,引入共享卷积,并将传统卷积的BN层优化成更稳定的GN层,有效节省模型的参数空间和计算资源。实验结果显示,改进后的YOLOv8n-Pose相较于原始算法,mAP@0.5提高了0.9%,参数量和计算量各减少了50%,同时FPS提高了8%,最终的疲劳驾驶识别率达到93.5%。经验证,本文算法在轻量化的同时能够保持较高的检测精度,并且能够有效识别驾驶员状态,为车辆边缘设备的部署提供有力支撑。Aiming to address the issues of complex detection processes,numerous parameters,low accuracy,and slow execution speed in current driver fatigue detection algorithms,we propose a lightweight model based on an improved YOLOv8n-Pose.This model optimizes the structure of YOLOv8n-Pose.Firstly,Ghost convolution is introduced into the backbone network to reduce the number of model parameters and unnecessary convolution computations.Secondly,a Slim-neck is introduced to fuse features of different sizes extracted by the backbone network,accelerating network prediction calculations.Additionally,an occlusion-aware attention module(SEAM)is added to the neck part to emphasize the facial region in images and weaken the background,improving keypoint localization accuracy.Finally,a GNSC-Head structure is proposed in the detection head part,which incorporates shared convolution and optimizes the BN layers of traditional convolution with more stable GN layers,effectively saving model parameter space and computational resources.Experimental results show that compared with the original algorithm,the improved YOLOv8n-Pose increases mAP@0.5 by 0.9%,reduces parameter count and computational cost by 50%,and increases FPS by 8%.The final fatigue driving recognition rate reaches 93.5%.Verified through experiments,this algorithm maintains high detection accuracy while being lightweight and effectively recognizes driver status,providing strong support for deployment on vehicle edge devices.

关 键 词：疲劳驾驶检测 深度学习 YOLOv8n-Pose 轻量化 注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]