Driving Fatigue Detection Based On Deep Learning And Information Fusion

As the most unstable human factor in road traffic safety,driving fatigue not only causes huge loss of social property,but also threatens the personal safety of citizens at all times,and is one of the main causes of major traffic accidents,so how to effectively monitor and prevent driving fatigue plays a vital role in ensuring road traffic safety and maintaining social stability.Traditional driving fatigue detection methods analyze the driver’s state based on the fatigue characteristics of a single source,and most of these methods have problems such as high false detection rate and poor adaptability,and thus the robustness of traditional driving fatigue detection methods is poor.In contrast,information fusion technology can effectively utilize the characteristics of complementarity and redundancy of multiple source features,which can easily form a consistent description of the detection target.Therefore,this dissertation adopts the information fusion method to fuse fatigue indicators from multiple sources,so as to realize the judgment of driver’s status.Since the human face information can intuitively reflect the health condition of human body,the detection of driver’s face information is more beneficial to determine the driver’s fatigue status.Compared with the existing face detection models,the Hog feature-based face detection model used in this dissertation not only has a better effect on the detection of side faces,but also can better meet the requirements of the algorithm in terms of detection time.At the same time,considering that the change of driver’s expression can greatly interfere with the determination of fatigue status,this dissertation builds a deep learning-based smile detection model by improving the existing LeNet-5 neural network model to achieve the classification of smiles in face images and reduce the risk of fatigue misclassification.In addition,in order to reduce the adverse effects of head deflection on the judgment of facial status,this dissertation introduces a feature correction algorithm based on Euler angles,which can largely reduce the differences between individuals and improve the accuracy of driving fatigue detection.When the driver is in a fatigue state,the control ability of the vehicle will gradually weaken,and it is easy to produce the phenomenon of the vehicle deviating from the road,which may even lead to traffic accidents in serious cases,so the characteristics of detecting the yaw of the vehicle also have an extremely important role in preventing fatigue driving.In this dissertation,based on the geometrical characteristics of the lane lines in the near field of view road images,a detection model of lane based on straight lines is constructed,and the parameters of the model are calculated using the PPHT algorithm to obtain the straight-line equations of the left and right lane lines in the road images,and the concept of yaw difference of the vehicle is proposed according to the change of the tilt angle of the left and right lane lines,which is used to describe the yaw characteristics of the vehicle in the process of driving.Finally,this dissertation uses the improved PERCLOS,blink frequency,yawn frequency and vehicle yaw difference as the main evaluation indexes for driving fatigue detection,and uses a weighted average for fusion analysis,so as to detect the different degrees of fatigue generated by the driver.It is also found that the proposed multi-source feature fusion driving fatigue detection method has higher robustness than the traditional driving fatigue detection method.