Driver Fatigue Detection Based On Edge Computing

Fatigue driving is one of the important causes of traffic accidents.The incidence of traffic accidents caused by fatigue driving is increasing.Timely warning of drivers’ fatigue state has important practical significance.In this thesis,the relevant algorithms of driver fatigue state detection are studied,and based on embedded platform,a fatigue driving detection system is designed to remind drivers to take a rest in time.First of all,according to the overall requirements of the system,the work flow of the system is determined,and the overall architecture of the system is designed.On this basis,the hardware and software design of the system is completed.The hardware is based on the embedded development board imx6 ull as the core,and the OV5640 camera and ATK-7084 display module are configured.The driver of related modules is written,and the application is designed based on QT/Embedded.Secondly,the driver fatigue state detection algorithm based on facial features is studied,aiming at the current mainstream algorithm MTCNN in face feature point detection of low efficiency shortcomings were improved,using hierarchical task,into the cavity convolution,space separation convolution,group convolution,average pooling and improved XCeption module.A lightweight convolutional neural network,Stretch-Net,is designed for face feature point detection,which reduces data redundancy,speeds up the convergence speed and enhances the sparse computing capability of the network,thus improving the computing efficiency.In addition,a multi-feature SVM classification model was constructed to judge the fatigue state according to the classification results of multiple frames,which was used to solve the problem that the original fatigue classification method was too simple.The trained network is deployed to the edge computing device for the detection of driver fatigue state.Finally,the modules of the system were tested,including feature point detection and fatigue state identification.The experimental results show that the system designed in this project meets the early needs of users,and has good safety,stability and reliability,and the system has high detection accuracy and practical value,which can be widely used in the driving field.