Research On Driver Fatigue Detection Method Based On Deep Learning

With the continuous development of road traffic and the increasing number of car ownership,the incidence of traffic accidents remains high.The white-hot contradiction between the people,vehicles and roads brings both property damage and mental harm to society and their families,threatening the life and safety of every citizen at every moment.Currently,a large number of research institutes both at home and abroad focus on this field.Among them,the detection method based on neural networks,as an important development direction of machine vision,not only possesses the inherent advantages of non-contact and convenient deployment,but also highlights the high accuracy and adaptability sexual self-characteristics,has gradually been a driver of fatigue testing a new hot spot.The backward infrared acquisition subsystem used in this paper can expand the application of detection system and improve the sample setting to improve the generalization ability of deep neural network detection model.Based on the driver fatigue detection requirements,finished the work of feature point detection,eye area detection,nodding motion detection and facial status recognition.Using the eye and head status regarded as the basic dimensions to construct the facial information space,proposing a method combining convolution neural network and recurrent neural network for driver fatigue detection.Firstly,the driver’s face position and feature point.Based on the output of the network model,the eye region is quickly located according to the face prior information.Secondly,constructing a cascaded convolutional neural network to recognize the extracted eye information and detecting the head movement in combination with facial key points to output facial state data.Finally,serialized face state data in facial information space,designed LSTM network,detected and alerted driver fatigue.Experimental results show that the proposed method can accurately detect the driver’s facial state under experimental and vehicle conditions,and the accuracy of the fatigue detection algorithm reach 94%with an average detection time of 65ms...