摘要
为解决传统学习疲劳检测方法在复杂场景下准确率与实时性不足的局限,文章设计了一种混合检测算法,旨在协同利用面部特征和上下文信息。通过改进YOLOv8-LSTM模型以提取多模态特征,并整合头部姿态、环境光照及时间序列分析。在YawDD公开数据集和自制数据集上的评估显示,该算法的查准率与查全率分别达到96.22%和98.08%。值得注意的是,在暗光、强光等挑战性环境下,算法仍能维持90%以上的识别准确率,较现有传统方法性能提升15%~20%。
In order to solve the limitations of insufficient accuracy and real-time performance of traditional learning fatigue detection methods in complex scenes,this paper designed a hybrid detection algorithm,which was designed to jointly use facial features and context information.By improving the YOLOv8-LSTM model,multimodal features are extracted,and head pose,environmental illumination and time series analysis are integrated.Experimental results on YawDD public dataset and self-made dataset show that the precision and recall of the proposed algorithm are 96.22%and 98.08%,respectively.It is worth noting that in challenging environments such as low light and high light,the proposed algorithm can still maintain a recognition accuracy of more than 90%,which is 15%~20%higher than the performance of the existing traditional methods.
作者
赵君君
ZHAO Junjun(Gansu Institute of Mechanical&Electrical Engineering,Tianshui,Gansu 741000,China)
基金
2024年甘肃机电职业技术学院校级科研项目:大模型助力面部特征分析的学生课堂学习状态检测系统(GSJD2024A01)。
