Research On Driver Fatigue Detection Method Based On Convolutional Recurrent Neural Network

With the development of economy,the number of cars is increasing all over the world.At the same time,traffic accidents happen frequently,which seriously endangers people's lives and property safety and has a great impact on social stability.Research shows that fatigue driving is one of the main causes of traffic accidents.If drivers are reminded of fatigue state in time,it will effectively avoid the occurrence of traffic accidents.Therefore,fatigue driving detection has great research significance and social value.Fatigue detection based on visual features has attracted wide attention due to its low cost,non-invasive characteristics.Detecting face and locating facial feature point are key steps in the methods based on visual features.However,there are some situations where the driver wears sunglasses or the light varies greatly in the real driving process,which will greatly affect the accuracy of fatigue detection.To solve these problems,this thesis builds an infrared camera acquisition system to collect driver's face images,which can effectively adapt to various driving environments and reduce the interference of light.At the same time,a fatigue driving detection method based on convolutional recurrent neural network is proposed,which regards fatigue detection as image-sequence recognition.The main works include face detection and feature location,eye region extraction,design for an end-to-end fatigue driving detection network,etc.Firstly,the driver's face image is collected by infrared acquisition equipment.Secondly,face detection and feature point location are carried out by multi-task cascaded convolution neural network(MTCNN);And the corresponding eye image-sequence is obtained according to the geometric relationship between feature points;Then,based on the end-to-end convolutional recurrent neural network model(Fatigue Driving Recognition Network,FDRNet),the spatial-temporal features of driver's eye state are extracted for some time.Finally,the driver's driving state is judged by analyzing the correlation between adjacent eye frames.The experimental results show that the proposed detection method can also accurately extract eye features under poor light conditions or when the driver wears sunglasses.Compared with the typical fatigue detection method based on Convolutional Neural Network(CNN)and PERCLOS standard,this thesis achieves higher accuracy of fatigue driving detection and realizes the video-level prediction for driving state.