Design Of Embedded Drinving Anti-fatigue System Based On Facial Feature

According to statistics from the public security department,as of the end of 2019,the number of motor vehicles in the country reached 348 million,and the number of motor vehicle drivers reached 435million.However,the vigorous development of the transportation cause has led to an increasing number of traffic accidents.One of the important reasons is the fatigue of drivers.Therefore,the design of a real-time and stable driver fatigue prevention system is of significant importance for their lives and property.Through the analysis of many anti-fatigue driving systems in the actual driving environment,it is found that the current fatigue driving detection system is easily influenced by the driver’s operating habits,driving skill proficiency,the complexity of the traffic road and other practical factors,and there is still room to improve its real-time performance and accuracy.The paper gives a design proposal for an embedded driving anti-fatigue system based on facial feature detection,an ensemble intelligent vision platform and STM32,which is tested in real driving environments.The research for the thesis is as follows:(1)Firstly,a deep learning algorithm based on MTCNN(Multi-task Convolutional Neural Networks)is implemented in the Caffe framework for face detection and multi-critical point localization and trained to generate a caffe model.(2)Then the dynamic face images are collected in real time based on Tencent’s VisionSeed intelligent vision platform,and after the conversion,optimization,deployment and fatigue detection algorithm of the caffe model generated by training on the PC side,the face images with the face frame and 98 key points labeled are output via the serial port on the one hand,and the structured parameters of the eye aspect ratio(EAR)and mouth aspect ratio(MAR)that can characterize the driver’s fatigue state calculated from the face key points on the other hand.(3)Finally integrated STM32 as the core controller based on VisionSeed calculation output EAR and MAR parameters as the main input for real-time voice alarm,so as to design and implement an embedded driving anti-fatigue system that can collect real-time face images and facial feature recognition.Tested by the actual driving environment,the system’s fatigue detection rate is above 15fps,and the whole system power consumption is below 10W,with real-time,low power consumption and the ability to adapt to complex environments.