基于深度学习的分层多尺度疲劳驾驶检测

Hierarchical Multi-Scale Fatigue Driving Detection Based on Deep Learning

摘要:

针对驾驶员疲劳检测的问题，建立了深度HM-LSTM网络实现基于面部特征的疲劳识别。使用LOTR模型实现面部空间特征的提取。考虑到驾驶员疲劳是一个循序渐进的过程，定义8种时序疲劳特征，使用堆叠的HM-LSTM在短时间序列以及长时间序列上捕捉疲劳度语义特征，从而在早期就准确识别出驾驶员疲劳迹象，通过低成本的视觉方法，进行可靠的驾驶员异常行为监测。使用公开数据集与自制D1-DDB数据集实现了模型的训练与系统的集成。

Abstract:

To address the issue of driver fatigue detection， this paper develops a deep HM-LSTM network to achieve fatigue recognition based on facial features. The LOTR model is used to extract facial spatial features. Considering that driver fatigue is a gradual process， it defines eight temporal fatigue features and uses stacked HM-LSTM to capture fatigue semantic features over both short and long time series， in order to accurately identify signs of driver fatigue at an early stage. This approach provides reliable monitoring of abnormal driver behavior through low-cost visual-based methods. Model training and system integration are conducted by using publicly available datasets and self-constructed D1-DBB datasets.