Drowsiness And Distracted Driver Detection Based On Deep Learning

With the development of deep learning and other technologies,intelligent assistant driving system can not only reduce the driver’s many operations in the driving process,but also guarantee the driver’s safety.In order to reduce the frequency of traffic accidents,it is very important to detect the driver’s fatigue and remind the driver when he is tired.How to perceive the behavior and state of drivers quickly,accurately and cheaply to further promote safe driving is a hot topic in the field of intelligent driving at home and abroad.In this paper,the method of fatigue detection is based on the trained face detection model,which uses the facial sign algorithm to detect the landmarks of eyes and mouth,and comprehensively judges the driver’s fatigue degree by combining the calculated PERCLOS value and the number of yawns.Through the experiment,the accuracy of the blink detection is 89%,the mouth opening is not less than 0.45,and the F1-Measure of yawn detection is 0.9 when the number of continuous frames is not less than 30.In this paper,the detection speed of the model is 0.41 seconds per image,and the size of the model is 13 MB.In this paper,a head posture detection model based on deep learning is designed,and multiple feature fusion algorithms are used to analyze the driver’s fatigue or attention.It can be seen from the experiment that the MAE of the head posture model designed in this paper is 9.005,and the actual number of fatigue detected is 17,which can accurately detect the characteristics of fatigue and distraction.In terms of detection speed,the CPU test result is 35 fps,GPU test result is 130 fps,and the model size is 6MB.According to the types of driving attitude,this paper designs a driving attitude classification network.The experimental results show that the accuracy of this model is 98%,the running speed of a single picture is 0.17 seconds,and the model size is 22.5MB.In conclusion,the fatigue detection method in this paper can give an accurate warning,and has good robustness to gender,race,light and whether to wear glasses.Not only the model size,speed and accuracy are optimized,but also the fatigue and driving attitude are combined for comprehensive detection.