Design Of Fatigue Driving Detection System Based On Facial Feature

According to the statistics of the Ministry of Public Security,there were 240 million cars and 409 million drivers in China by the end of 2018,with an average annual growth of 30.12 million.Along with the rapid development of road traffic,the casualties and property losses caused by road traffic accidents have also increased year by year and now it is ranked first in the world.Related studies have found that 20% to 30% of traffic accidents are caused by fatigue driving.Therefore,it is significant to design a reliable and real-time fatigue driving detection system to ensure road traffic safety.In this thesis,a reliable real-time fatigue driving detection system is designed.Through the research on the practical application environment of the fatigue driving system and the driver's driving habits,it is found that there are great limitations in the common fatigue driving detection technology at home and abroad.For example,in the actual working environment of the system,the effectiveness of the existing fatigue driving detection system in detecting the fatigue driving state is still insufficient due to the influence of many factors,such as the difference in driver skills,driving habits,complex road traffic environment and the acceptability of the driver to the detection system.In this thesis,a reliable and high-performance fatigue driving detection system with good real-time performance will be designed.The method of multi-fatigue feature fusion is applied to fatigue detection.The method is also validated and applied in this thesis.The details of this thesis are as follows:(1)The driver's face area detection and tracking method combined with the MTCNN face detection algorithm and the DSST video single target tracking algorithm is proposed in this thesis.Common algorithms can't detect face occlusion and head rotation very well.So the MTCNN face detection algorithm with the best effect is selected to locate the driver's face.The DSST algorithm,which has the best target tracking effect in video at present,is used for face tracking and the continuous and stable tracking of driver's face area is realized.(2)The face feature point alignment algorithm based on regression tree set is proposed.Commonly used facial feature point localization methods are based on region segmentation or statistical-based methods,and most of the algorithms are to calculate the positioning and opening degree of human eyes and mouth respectively,with complex steps and low accuracy.After a lot of literature review,a new method is applied to the fatigue driving system in this thesis which integrates human eye/mouth detection and locates the eye/mouth position in a very short time(milliseconds).(3)A new fatigue driving discriminating strategy is proposed.The characteristics of the behavioral characteristics of the driver's fatigue state and the currently recognized characteristics with high degree of fatigue are studied.Fatigue driving discrimination is carried out by using multiple features such as PERCLOS value,blink frequency,yawn frequency,nodding frequency and head posture change.According to the definition of PERCLOS,a new eye discrimination index(eye aspect ratio)and a new mouth discrimination index(mouth aspect ratio)are proposed to discriminate fatigue driving.The Fusion detection of multi-features is realized.When a certain index fails,fatigue state can still be detected normally,which has a certain robustness.(4)A fatigue driving detection system based on multi-feature fusion is designed.According to the analysis of the actual working environment of the system,using the targeted software and hardware platform,the algorithm studied in this thesis is realized by using the targeted software and hardware platform.The system has been validated on the platform of a scientific and technological company.A large number of experimental results show that the system designed in this thesis has good real-time performance,effectiveness and robustness.It can achieve good results in practical application environment and improve road traffic safety.