摘要
在制造业工人疲劳检测领域,由于工人头部姿态频繁变化,传统疲劳识别方法在大姿态下往往精度下降。为此,提出一种基于3DDFA-V2的人脸对齐面部疲劳检测网络(FAFDNet)。首先利用改进的Retinaface面部检测器快速捕捉人脸,然后采用3DDFA-V2模型拟合出面部三维关键点及头部姿态坐标轴。基于预测的头部姿态Z轴,将眼睛与嘴巴区域的关键点投影到人脸前向平面,以消除大偏航角对关键点位置的影响;接着根据投影后关键点计算眼睛纵横比(EAR)和嘴巴纵横比(MAR),依此再计算眼睑闭合率PERCLOS和打哈欠频率YR来判断工人疲劳程度。在3个数据集上的实验结果表明,所提出的算法可以准确计算识别面部偏航角(0°-90°)范围内的眼睛和嘴巴状态,进而检测面部疲劳状态。
In the field of fatigue detection for manufacturing workers,traditional fatigue recognition methods often experience a decline in accuracy under large head posture changes due to the frequent changes in workers'head postures.To address this issue,this paper proposes a 3DDFA-V2-based face alignment facial fatigue detection network(FAFDNet).First,an improved Retinaface facial detector is used to quickly capture the face.Then,the 3DDFA-V2 model is employed to fit the three-dimensional key points of the face and the head posture coordinate axes.Based on the predicted Z-axis of the head posture,the key points of the eye and mouth regions are projected onto the forward plane of the face to eliminate the impact of large yaw angles on key point positions.Subsequently,the eye aspect ratio(EAR)and mouth aspect ratio(MAR)are calculated based on the projected key points,and the eyelid closure rate(PERCLOS)and yawning frequency(YR)are further calculated to assess the worker's fatigue level.Experimental results on three datasets demonstrate that the proposed algorithm can accurately calculate and identify eye and mouth states within a facial yaw angle range of 0°-90°,thereby detecting facial fatigue states.
作者
崔金磊
吴锦华
陈科伟
CUI Jinlei;WU Jinhua;CHEN Kewei(School of Mechanical Engineering and Mechanics,Ningbo University,Ningbo Zhejiang 315211,China;Zhejiang Wanfeng Tech-nology Development Co.,Ltd.,Shaoxing Zhejiang 312500,China)
出处
《佳木斯大学学报(自然科学版)》
2025年第10期129-131,156,共4页
Journal of Jiamusi University:Natural Science Edition
