Design Of Driver Fatigue Monitoring System Based On Deep Learning

As an important part of modern transportation tools,automobiles not only facilitate people’s daily travel but also bring traffic accidents.Among them,the accidents caused by fatigue driving account for about 21% of the number of car accidents every year.Therefore,it is of great social significance to design a monitoring system for early warning of the driver’s fatigued driving state.In this paper,a driver fatigue monitoring system is designed by analyzing the characteristics of the driver’s driving behaviour.The system is based on the YOLOv5 target detection algorithm under improved deep learning.This paper uses the Res2 Net structure to replace the residual components in the original network structure of YOLOv5 to make the model backbone.The network has a stronger ability to extract target features to improve the accuracy of detection;the samples of the analysis data set are small-scale targets,so the prediction head part of the original structure of YOLOv5 is pruned to eliminate large targets and medium targets so that The network is lightweight to improve the realtime detection speed of the system when the edge computing device is deployed.At the same time,this paper uses NVIDIA Jetson Nano as the development platform to develop a driver fatigue monitoring system.The system captures the driver’s face image through the Intel Real Sense D435 i camera.When the system determines that the driver has fatigue driving behaviour,the system uses the visual interface.Visually display the relevant information of the monitoring system,and provide early warning through the buzzer to help the driver drive safely.The detection model designed in this paper was tested in comparison with the publicly available dataset NTHU-DDD and a self-built dataset,using accuracy,precision and recall as test evaluation indicators.After comparative experiments,the improved detection framework shows good performance.Compared with the network structure before the improvement,the accuracy is increased by 1.2%,the accuracy rate is increased by 1%,the recall rate is increased by 1.1%,and m AP@0.5From 97.9% to 98.3%,the calculation amount of the model is reduced by 25%,and the amount of parameters is reduced by 14%.The driver fatigue monitoring system developed by this algorithm is verified by in-vehicle experiments.When the driver has a fatigue driving behaviour with increased eye closure frequency and increased yawning frequency,the system can timely detect and give an early warning to the driver,while monitoring Information can also be displayed simultaneously on the system’s visual interface.Experiments show that the system designed and developed in this paper runs smoothly,with high accuracy,robustness and real-time performance.