| Fatigue can make drivers drowsy when driving,which is one of the important reasons for traffic accidents.Therefore,if we can monitor the driver’s real-time status and give a warning when he is sleepy,we can effectively prevent the occurrence of traffic accidents.The existing drowsiness detection technologies can be roughly divided into three categories:(1)Detection methods based on vehicle parameters.(2)Detection methods based on physiological parameters.(3)Detection methods based on visual features.The detection methods based on visual features only needs on-board camera to monitor the driver’s driving state.Compared with the method based on physiological parameters,the driver will not be if.Moreover,the detection technology based on visual features directly monitors the driver’s facial expression,which is more reliable and less affected by the driver’s driving habits than the detection technology based on vehicle parameters.Therefore,this paper uses the technology based on visual features to detect the driver’s drowsiness,and judges whether the driver has drowsiness in the process of driving through the driver’s facial expression and head movement.Traditional methods use hand-made features to extract facial information,but the feature extraction process is complex,depends on specific tasks,and has poor portability.Therefore,this paper uses deep learning to extract facial features,and proposes two sleep detection schemes based on CNN(Convolutional Neural Network).(1)This paper proposed a Drowsiness Action Recognition Model(DARM).This model is based on a 3D Convolutional Neural Network,which is able to identify four actions,including three drowsy actions and one non-drowsy actions.Then,optical flow is input into the model to further extract the motion information,which improves the accuracy of sleepiness detection.Finally,the NTHU-DDD(National Tsinghua University Driver Drowsiness Detection)dataset is used to verify the performance of the model.The experimental results show that the accuracy rate reaches 86.6%,which is better than the existing methods.(2)This paper presents a detection method based on TEA(Temporal and Aggregation)network.This method firstly uses MTCNN(Multi-task Cascade Convolutional Neural Network)to extract facial features and remove redundant information such as background.The extracted image is then entered into the TEA network for driver sleepiness recognition.The network contains both the Me(Motion Modeling)block and the MTA(Multiple Temporal Aggregation)block.It is very convenient for time information modeling.Compared with similar algorithms,it has better performance. |
