基于驾驶员面部时序数据的疲劳驾驶检测算法

Fatigue Driving Detection Algorithm Based on Facial Sequential Data of Drivers

中文关键词:  迁移学习  疲劳驾驶  疲劳检测  ResNet  LSTM

英文关键词:transfer learning  fatigue driving  fatigue detection  ResNet  LSTM

中文摘要:

      针对传统疲劳驾驶检测方法识别准确率低、泛化能力差的问题，提出了一种基于CNNs和LSTM的端到端可训练网络，检测驾驶员的疲劳状态。根据驾驶员面部特征点提取ROI，将在其他计算机视觉任务上表现较好的深度网络迁移到疲劳检测任务中，并结合LSTM处理时序数据的能力，提出一种新的疲劳检测网络，该网络能够读入视频流中的时序数据并检测出驾驶员的疲劳状态。实验证明所提方法和模型在公开数据集中具有较高的识别准确率，并且在不同的数据集间具有很好的泛化能力，对于减少路面车祸、保障人身安全具有很重要的意义。

英文摘要:

      In order to solve the problems of low recognition accuracy rate and poor generalization ability of traditional fatigue driving detection methods,an end-to-end trainable network based on CNNs and LSTM is proposed to detect the fatigue state of drivers.According to the driver’s facial landmarks,ROI of fatigue is extracted.Next other deep network which performs well on other tasks of computer vision is used to transfer to the problem.Combining with the ability of LSTM to process time series data,a new fatigue driving detection network is proposed.The proposed network can read the time series data in the video stream and detect the fatigue state of the driver.Experiments prove that the method and model have a high recognition accuracy rate on public data set,and have a good generalization ability between different datasets.It has significance meaning for reducing road accidents and ensuring personal safety.