Research And Implementation Of Facial Fatigue Information Detection Method Based On Deep Learning

According to statistics from the Ministry of Public Security,the number of motor vehicles nationwide will reach 372 million in 2020,and the number of motor vehicle drivers will reach 456 million.Such a huge number of cars makes the frequency of traffic accidents high.According to the "China Statistical Yearbook" released by the National Bureau of Statistics in the past three years,the average annual number of deaths in traffic accidents in my country is 63,000,and the number of traffic accidents is 231,900..According to related investigations,more than 20% of traffic accidents are caused by fatigue driving.Therefore,designing and implementing an efficient fatigue detection method is of great significance to preventing fatigue driving.This paper uses computer vision technology to design and implement a reliable real-time fatigue detection method.This method is low in cost,high in efficiency,and has no contact with the driver,and will not affect the normal driving of the driver.At present,fatigue detection using computer vision technology at home and abroad mainly uses a single feature to make judgments.The number of blinks or yawns is used as the basis for judging fatigue.This detection method is too limited and only uses the number of times as the basis.Judgment basis,reliability is not high.In order to improve the reliability of the detection method,this paper adopts the method of multi-feature comprehensive judgment for fatigue detection.In addition to blinking and yawning as features,this paper proposes to use head posture features as the basis for judgment to determine whether there is nodding due to drowsiness In order to design and implement an efficient and reliable fatigue detection method.This paper uses the Ada Boost structure based on Haar-Like features to train the face detection tool,and proposes a multi-task cascaded convolutional neural network.This method is used to train the 30 key points of the face and realize the facial features of the driver.Continuous and stable positioning of points.At the same time,this article proposes a comprehensive judgment method incorporating head fatigue features.The head fatigue feature is mainly the behavior of nodding.The change in head posture is the increase in head depression angle.Therefore,the head posture information needs to be obtained during the judgment process.,The introduction of this feature provides a new idea for fatigue detection and,after research in this article,it is found that most of the fatigue detection methods characterized by blinking only use the number of blinks in a period of time as the basis for judging fatigue.This judgment strategy is too simple.Inspired by the PERCLOS judgment index,this article proposes a new judgment method to distinguish between normal blinks and fatigue blinks.The number of fatigue blinks is used as the eye fatigue feature,and the yawning mouth is integrated.The fatigue feature and the nodding head fatigue feature are comprehensively judged to realize the comprehensive detection of multiple features.When a feature fails,the detection method can still detect fatigue normally.