Research On Fatigue Driving Detection Based On Deep Learning

In recent years,traffic accidents have occurred frequently,causing huge losses to society.Accidents caused by fatigue driving have been increasing year by year,which will seriously affect the life safety of the driver himself and others.Research on fatigue can generally be divided into two categories: one is subjective detection method;the other is objective detection method;for subjective detection,some scholars have proposed to collect related information of fatigue driving,and analyze the factors producing fatigue from the perspective of statistics.This method has the limitation of being affected by human factors and not being able to detect the fatigue of the driver in real time;based on objective detection,it can be divided into three categories: detection method based on driver’s physiological characteristics;detection method based on driver’s or vehicle’s motion state;detection method based on facial features.Some physiological scholars have proposed that there are differences in physiological indicators between fatigue and alertness.Through detecting the driver’s electroencephalogram and electrocardiogram and other physiological indicators,this method has the advantages of high accuracy and good robustness,but there are also shortcomings of difficult deployment of detection equipment and affecting driving experience.Therefore,this paper presents a fatigue driving detection algorithm based on deep learning.The work of this paper is as follows:(1)A modified MTCNN face detection optimization algorithm is proposed.According to the need of complex driving environment and quick location of face area,combined with depth separable convolution,the MTCNN network model is improved,and the network structure is simplified.In model training,various pictures with different driving scenarios,angles and light intensity are used.Compared with other face detection methods,the algorithm proposed in this paper has the advantages of fast detection speed and good robustness.(2)The proposed fatigue detection algorithm is based on facial feature points.In order to enable the algorithm to collect the driver’s fatigue features in real time,the face feature point algorithm has a certain detection speed.Based on the facial feature point detection algorithm PFLD,combined with GhostNet and re-parameterization idea,the model is lightened.At the same time,the output is predicted by adding 5 to the original 3 feature prediction layers,so that the model can improve the network inference speed while ensuring the detection accuracy of the facial feature point algorithm,thus meeting the need of real-time collection of fatigue driving.(3)In the process of building the fatigue feature model,the eye and mouth closure state of the driver is classified by support vector machine,instead of the original threshold segmentation method.This classification method takes into account the physical differences of different drivers and sets different classification standards for different individuals.(4)A multi-feature fusion fatigue judgment method is proposed.Through collecting eye fatigue features and mouth fatigue features,and analyzing the eye state within a certain period of time to obtain the driver’s real-time PERCLOS value,the fatigue of the driver is judged comprehensively through three fatigue features.Compared with single-feature detection methods,the multi-feature fusion method has the advantages of high accuracy and good robustness.