Research On Driver’s Visual Fatigue Detection Based On Deep Learning

In the recent years,according to the relevant statistics of the road safety department,fatigue driving poses an increasing threat to road driving safety.About 20% of traffic accidents are caused by fatigue driving every year.Fatigue driving has become one of the important factors threatening traffic safety.Therefore,this thesis studies and designs a real-time detection system for driver fatigue state,which can provide timely and accurate alarm when the driver has fatigue characteristics,and effectively prevent the occurrence of traffic accidents.This thesis proposes a driver fatigue detection method based on Perclose and visual feature.In order to improve the effect of night recognition,this thesis uses a high-definition night vision camera to obtain the driver’s face image in real time.In order to eliminate the influence of uneven illumination on the subsequent algorithm,firstly,the input image is enhanced and filtered.Then,the face detection is realized by using the multi task convolutional neural network algorithm.After the face position is determined,in order to eliminate the influence of large occlusion and posture change on the feature points location,a projection mapping initialization method is proposed in this thesis,and the multi-view cascading regression algorithm is used to realize the face feature point location.Then the eye region is intercepted based on the face feature point location to realize the accurate eye location,and the image of the eye region is sent to the lightweight eye state recognition network to distinguish the human eye state.Finally,the fatigue state of the driver is judged according to the proportion of the time of eye closure and the time of eye closure per unit time.If the ratio exceeds the set threshold,a voice alarm will be given and the alarm information will be displayed on the TFT screen.In terms of hardware design,according to the actual needs of the system,such as large amount of data,fastly reading and transmission speed data,complex algorithm and so on,this thesis selects NT96687 as the core processor of the hardware system.The system is mainly composed of video acquisition module,NT96687 core processor,image display module and voice alarm module.This thesis not only completes the transplant of the algorithm,but also optimizes the algorithm through the instruction set of XM4 and parallel acceleration technology to ensure the real-time performance of the system.Finally,the test results of NT96687 show that the system can accurately identify the state of human eyes in the daytime or at night,and effectively detect the fatigue state of the driver.It not only has the high detection accuracy and reliability,non-contact,low power consumption and other characteristics,but also can meets the real-time application requirements of the vehicle,which has a certain application value.