机电工程技术2025，Vol.54Issue(17)：41-45,87,6.DOI:10.3969/j.issn.1009-9492.2025.00039
基于MediaPipe-Attention-Mesh模型的面部多特征疲劳驾驶检测算法
Facial Multi-feature Fatigue Driving Detection Algorithm Based on MediaPipe-Attention-Mesh Model
摘要
Abstract
To address the limitations of existing facial feature localization algorithms,such as insufficient accuracy in landmarks detection and oversimplified features for fatigue state discrimination,a multi-feature fatigue driving detection algorithm based on MediaPipe-Attention-Mesh model is proposed.The method constructs a lightweight neural network architecture and introduces a channel-spatial dual-attention mechanism to enhance feature extraction in critical regions(e.g.,eyelids and lips),achieving precise localization of 478 facial landmarks.Temporal features,including blink frequency,sustained eye closure duration,and yawn frequency per unit time,are extracted to establish a fatigue-specific dataset.A random forest classifier is employed to determine fatigue states.Experimental results on the YawDD benchmark dataset demonstrate an accuracy of 95.7%(F1-score=0.952,AUC=0.983).Ablation studies reveal that the multi-feature fusion strategy improves AUC by 3.5%compared to single-feature approaches.This method provides an effective technical solution for real-time fatigue monitoring in complex driving scenarios.关键词
MediaPipe-Attention-Mesh/关键点检测/多特征融合/随机森林/疲劳判断Key words
MediaPipe-Attention-Mesh/key point detection/multi-feature fusion/random forest/fatigue assessment分类
信息技术与安全科学引用本文复制引用
薛珍珠,程豪,王权,王艳..基于MediaPipe-Attention-Mesh模型的面部多特征疲劳驾驶检测算法[J].机电工程技术,2025,54(17):41-45,87,6.基金项目
河南省高等学校重点科研项目计划(23B460005) （23B460005）
