A Non-contact Fatigue Detection Method Based On Blood Flow

In recent years,the problem of fatigue has attracted more and more attention,especially in the field of driving.The increasing number of traffic accidents due to fatigue driving has caused great harm to people's lives,health,safety and property.How to detect fatigue effectively and accurately has become a hot topic for many researchers.Therefore,it is important to study fatigue detection methods.Traditional fatigue detection methods are limited to contact detection methods.Compared to contact fatigue detection methods,non-contact fatigue detection methods have the advantages of non-invasive methods,ease of use and low cost.At present,most non-contact fatigue detection methods are based on facial or limb behavioral characteristics.When the behavioral characteristics are obvious,these methods have a higher accuracy of judgment,but when the behavioral characteristics are absent,the result is usually unsatisfactory.This article makes a study of fatigue detection based on facial blood flow signals rather than behavioral characteristics.This article uses face video data set as the research object,focusing on the non-contact physiological fatigue detection method,hoping to improve the accuracy of fatigue detection.This article mainly consists of 4parts:First,the background of human fatigue and the target of research are introduced.Secondly,this article presents the basic theory of blood flow signal and blood flow imaging technology.Then,based on the non-contact image photoplethysmographymethod,this study proposes a blood flow-based equivalent mapping atlas,fusing the blood flow equivalent atlas with face images.Finally,we apply this method combining with deep convolutional neural networks on a 60-person UTA-RLDD(UTA Real-Life Drowsiness Dataset)fatigue video data set and obtain good results.In this study,a variety of convolutional neural networks were compared and analyzed.In the two-class detection task of fatigue and non-fatigue,the higher accuracy rate was 88%.The results show that the method of this study has a good effect on the problem of fatigue detection,and provides a novel method for fatigue detection.