Fatigue Monitoring Of Train Drivers Based On Deep Learning

With the rapid development of railway transportation in my country,the operation of train drivers has changed from traditional train control to long-term monitoring of real-time train operation information.Monotonous driving operations and long-term driving operations will significantly increase the degree of fatigue and drowsiness of train drivers.The largest cause of train accidents is the fatigue driving of train drivers.Therefore,it is particularly important to reduce train accidents caused by fatigue driving and improve the driving safety of train drivers.Firstly,the impact and harm caused by fatigue of train drivers are analyzed,and the importance of fatigue detection during train operation is clarified.The previous research results are summarized.It is determined that this thesis takes the facial features of train drivers as the object to study,extracts facial fatigue parameters,and analyzes and judges whether the train drivers have fatigue driving.Secondly,an attention double enhancement algorithm is proposed.In view of the influence of illumination on the image,in a low-light environment,it is easy to reduce the detection rate of the image by directly detecting the captured video image.Using the attention double enhancement network,its attention subnet,noise subnet,enhancement subnet and enhancement subnet,these four subnets can simultaneously improve the brightness of the image and complete the denoising effect.The experimental results show that the recognizability of the image can be improved by processing the image in advance,the adaptability of the system to the illumination change can be enhanced,and the detection accuracy can be improved.Thirdly,perform face detection and face key point location on the image after low-light enhancement processing.This thesis combines the face detection algorithm based on multi-task convolutional neural network with the face key point detection algorithm based on regression tree.Locating the 68 key points of the face.The three features of the driver’s eyes,mouth and head pose are analyzed.For eye feature extraction,the eye feature aspect ratio algorithm is used to calculate the eye parameters,so as to obtain the PERCLOS value and as the basis for judging fatigue;for mouth feature extraction,the mouth aspect ratio algorithm is used to calculate the mouth parameters.For the influence of the thickness of the lips,the key points of the inner contour of the mouth are selected to calculate the MAR value,and yawning is used as the basis for judging fatigue;for the extraction of head features,the pitch angle offset in the Euler angle in the three-dimensional space is selected as 20%.Determine the basis for fatigue.Finally,multi-feature fatigue detection is used to judge the driver’s driving state.The accuracy of the existing single fatigue feature extraction is low.A multi-feature fatigue detection are designed in this thesis.The fatigue detection algorithm that fuses the fatigue features of eyes and mouth with fuzzy reasoning can obtain a graded fatigue state.The total fatigue detection time is 44.3ms,which meets the real-time requirements.If it is detected that the driver is in different levels of fatigue,it will give corresponding prompts and alarms.