Research On Classification And Early Warning Of Fatigue Detection Based On Deep Learning

With the rapid growth of the number of cars,traffic safety problems are becoming more and more prominent.Among them,fatigue driving will lead to the decline of human body functions and seriously endanger road safety.Therefore,based on modern information and artificial intelligence technology,the research and development of fatigue detection and classification early warning system has important academic value and application prospects.Aiming at the typical characteristics of fatigue driving,this paper solves the problem of state recognition and intelligent early warning of fatigue driving based on convolutional neural network.The main research contents are as follows:First of all,in order to solve the problem of low face detection rate in complex scenes,for the samples in the face database,techniques such as illumination compensation,filtering,enhancement processing,gamma correction,logarithmic transformation and other techniques are used to preprocess the image to improve the accuracy of the image.The detectability of the useful information of the image,avoiding the useless information in the image to interfere with the detection result.Secondly,based on the MTCNN(Multitask Cascaded ConvolutionalNetworks,MTCNN)network structure,this paper makes a series of improvements to complete the face detection and localization tasks,and also solves the problem of the current driver’s attitude change,complex lighting and occlusion.In order to solve the problem of poor real-time performance of face detection,this paper introduces the lightweight network structure MobileNet to improve the detection speed,and changes the pooling layer to full convolution operation to improve the network detection performance,and the MTCNN structure is redesigned.build.Based on the above method,the selected4000 pictures of the Celeb A face data set are pre-trained.The experimental results show that the Miou index of the improved network is finally stable at 95.2%,which is about 2% higher than that of the original network.The final convergence value of Loss Compared with the original algorithm,it is stable at 0.18%.The improved Mobile-MTCNN detection algorithm is significantly better than the original network in detection accuracy.Finally,in order to solve the problem that a single fatigue index is difficult to accurately assess the driver’s fatigue state,this paper uses the cascade regression tree face key point detection algorithm to locate the driver’s face,and proposes a fusion of blink frequency,eye closure times,and yawning.The five fatigue judgment index methods of frequency,PERCLOS value and head posture estimation are used to normalize the index values collected in real time,and the fatigue state is divided into five states: awake,critical,mild fatigue,moderate fatigue and severe fatigue.The wx Form Builder development tool is used to build a visual monitoring and early warning platform,which can monitor the driver’s status in real time and realize the driver’s fatigue driving graded early warning.