Research And Implementation Of Fatigue Driving Detection Method Based On Deep Learning

The rapid development of the road transportation industry has facilitated people's daily lives,but the huge economic losses and casualties caused by frequent traffic accidents cannot be ignored.Research results show that one of the main causes of traffic accidents is fatigue driving.According to a report from the National Center for Statistics and Analysis,drivers can avoid 60% of traffic accidents if they can make correct operations half a second before a traffic accident.Therefore,by detecting the driver's fatigue state and promptly reminding the driver,has the great practical significance to reduce the occurrence of traffic accidents caused by fatigue driving and protect the life and property of the driver and passengers.The research on fatigue driving detection has attracted widespread attention from researchers at domestic and abroad,and has also achieved more research results.The detection method based on machine vision has become an important method for detecting fatigue driving due to its non-contact and real-time advantages.However,this method has two major problems: it is not adaptable to light changes,and it is difficult to balance real-time and accuracy.The fatigue driving detection method proposed in this paper gradually detects the face area and facial key points information through digital image processing and machine vision technology,extracts the driver's eye and mouth areas,and determines the driver's head posture based on the facial key points information.Then,the driver's eye and mouth states are accurately identified through the convolutional neural network technology in deep learning,and finally the fatigue characteristics of the eyes,mouth and head are combined to determine the driver's fatigue.The main work of this article is as follows:1.Aiming at the impact of light on the image,two digital image processing methods,reference white and histogram equalization,are used to supplement the light collected in the image under abnormal light conditions.Median filtering is used to reduce the noise of the image and simple digital image processing methods,such as image graying,pre-process images.It is verified through experiments that the image preprocessing method proposed in this paper can improve the recognizability of the device imaging,enhance the system's adaptability to light changes,and improve the accuracy of detection.2.Face detection and facial key points detection are the basis for studying and analyzing fatigue driving.Considering the real-time and accuracy of detection,a face detection algorithm based on directional gradient histogram is used to locate the face area,and a regression tree is used.The method accurately locates 68 key points of the face in the detected face region,and performs head pose estimation and extracts eye and mouth region images through the key points information.3.Using the obtained images of the eye and mouth regions as training data,the effects of different parameters in the two convolutional neural networks on the accuracy of the network model were studied through experiments,and the parameters with higher accuracy were selected to construct the convolutional neural network.After training the obtained convolutional neural network model can determine the open and closed state of the eye and the open and closed state of the mouth.4.For the existing fatigue detection methods,a single fatigue characteristic is often used,and the fatigue state is not fully characterized and the adaptability to complex environments is not strong.By calculating the parameters of eye fatigue,mouth fatigue and head posture,the fatigue state of the driver is detected by combining multiple fatigue parameters.The experimental results show that the fatigue driving detection method based on deep learning proposed in this paper has a strong adaptability to the influence of light,and can achieve higher accuracy under the premise of ensuring real-time performance.