Scholarlink

This paper proposes a fatigue detection method based on multi-modal feature fusion of ocular, oral, and head posture characteristics. A pre-trained robust facial landmark detection model is employed to achieve precise localization of 68 facial keypoints, with emphasis on constructing a three-dimensional feature space incorporating eye aspect ratio (EAR), mouth aspect ratio (MAR), and head Euler angles. A dynamic threshold method based on an improved PERCLOS criterion is proposed to quantify blink frequency and sustained closure duration. Spatial coordinate transformation establishes a mapping relationship between mouth aperture and yawning actions, while rotation matrices derived from head pose estimation enable precise measurement of nodding angular variations. A three-level fatigue classification model is developed according to the Stanford Sleepiness Scale (SSS), integrated with an entropy weight-based multi-feature fusion algorithm to enable comprehensive quantitative fatigue assessment.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on Fatigue Detection Based on Multi-feature Fusion with Deep Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Shaohua Li,
  • Zhiqiang Gao,
  • Weitong Ji,
  • DeXiang Yang,
  • Xuan Wang

摘要

This paper proposes a fatigue detection method based on multi-modal feature fusion of ocular, oral, and head posture characteristics. A pre-trained robust facial landmark detection model is employed to achieve precise localization of 68 facial keypoints, with emphasis on constructing a three-dimensional feature space incorporating eye aspect ratio (EAR), mouth aspect ratio (MAR), and head Euler angles. A dynamic threshold method based on an improved PERCLOS criterion is proposed to quantify blink frequency and sustained closure duration. Spatial coordinate transformation establishes a mapping relationship between mouth aperture and yawning actions, while rotation matrices derived from head pose estimation enable precise measurement of nodding angular variations. A three-level fatigue classification model is developed according to the Stanford Sleepiness Scale (SSS), integrated with an entropy weight-based multi-feature fusion algorithm to enable comprehensive quantitative fatigue assessment.