首页|期刊导航|高师理科学刊|基于多源特征融合的驾驶员疲劳检测

基于多源特征融合的驾驶员疲劳检测OA

Driver drowsiness detection based on multi-source feature fusion

中文摘要英文摘要

针对传统单一特征疲劳驾驶检测系统在光照突变、部分遮挡等复杂场景下适应性弱、准确性与鲁棒性不足,且存在漏检误检、精度与实时性难以平衡的问题,提出了一种非接触式多源特征融合检测方法.该方法融合眼部、嘴部、头部姿态等多源特征,采用"Haar级联分类器快速粗定位+Dlib点模型精细检测"的混合框架,融入动态阈值补偿、仿射校正、双线程优化等策略.实验表明,该方法的人脸检测召回率大于等于95%,疲劳检测F1-Score大于等于0.90,单帧处理延迟小于等于80 ms,能够较好地平衡精度与实时性,对提高道路交通安全具有重要意义.

To address the issues of weak adaptability,insufficient accuracy and robustness,missed and false detections,and the difficulty in balancing precision and real-time performance in traditional single-feature fatigue driving detection systems under complex scenarios such as sudden illumination changes and partial occlusion,a non-contact multi-source feature fusion detection method is proposed.This method integrates multi-source features such as eye,mouth,and head posture,and adopts a hybrid framework of"Haar cascade classifier for fast coarse positioning+Dlib-point model for fine detection",incorporats strategies such as dynamic threshold compensation,affine correction,and dual-thread optimization.Experiments show that the face detection recall rate of this method is greater than or equal to 95%,the fatigue detection F1-Score is greater than or equal to 0.90,and the single frame processing delay is less than or equal to 80 ms,which can achieve a good balance between accuracy and real-time performance,has great significance to improve road traffic safety.

魏芳波;姚世佳;樊希;袁治;罗奇迹

汉江师范学院物理与电子工程学院,湖北十堰 442000汉江师范学院物理与电子工程学院,湖北十堰 442000汉江师范学院物理与电子工程学院,湖北十堰 442000汉江师范学院物理与电子工程学院,湖北十堰 442000汉江师范学院物理与电子工程学院,湖北十堰 442000

信息技术与安全科学

疲劳检测多源特征融合Haar级联分类器眼睛纵横比眼睑闭合百分比

drowsiness detectionmulti-source feature fusionHaar cascade classifiereye aspect ratioPercentage of Eyelid Closure(PERCLOS)

《高师理科学刊》 2026 (8)

33-38,6

湖北省省级大学生创新创业训练计划项目(S202510518045)汉江师范学院重点科学研究项目(XJ2025A09)

10.3969/j.issn.1007-9831.2026.08.005

评论