Research On Driving Behavior Analysis Based On Machine Vision And Edge System Design

Nowadays,while traveling by car brings convenience to people,it also increases a lot of pressure on traffic safety.Among them,fatigue behavior caused by long-term driving and irregular distracted driving behavior are the main reasons for frequent traffic accidents.Therefore,this paper mainly studies the driving behavior analysis and recognition method based on machine vision and designs a real-time edge system.The main work is as follows:First,this paper studies a driving behavior recognition method based on transfer learning.This part is mainly divided into research on driving behavior recognition methods and technical research on image enhancement technology,as well as experimental design and analysis on public datasets.The research on the recognition method of driving behavior mainly includes the theoretical research of convolutional neural network and the research of mainstream models,including Alex Net,VGGNet,Google Net,Res Net and other models,to explore the feasibility of machine vision methods in driving behavior analysis.Secondly,this paper studies the identification and analysis of fatigue behavior.This part mainly includes the research on the driving fatigue monitoring method and the design and implementation of the fatigue behavior recognition system.Among them,the research on driving fatigue monitoring method mainly includes HOG-based face detection and face key point monitoring,and finally uses the PERCLOS principle to identify fatigue behavior.Based on the method of fatigue detection,this paper designs a real-time recognition system for fatigue behavior.In complex scenes,the recognition of eyes closed and yawning behaviors is realized,and the driver’s fatigue state is comprehensively judged.Thirdly,t his paper studies the identification and analysis of driving distraction behavior.This part includes the collection of the NJUPT dataset,the research on the selection of the dataset,the training and evaluation of the model,and the analysis of multi-weight driving behavior.In the research part of data set collection and data set selection,the same model VGG16 is used to train on different data sets,and finally the data set combination with the strongest generalization ability is selected.After the image dataset is preprocessed,multiple models are tested and evaluated on the dataset,and the results of balancing accuracy and inference speed are combined.Shuffle Net V2 is selected as the model used by the edge system and the model is improved to avoid distraction The overall behavior recognition rate increased from 82% to 86.3%.Finally,this paper implements a machine vision-based edge system for driving behavior.It mainly includes two parts: system design and system function test.The system design part mainly introduces the design and implementation of hardware and software.The system function test part realizes the function test in the actual environment,and conducts real-time detection of the driver’s behavior,reaching a correct rate of more than 80%,which basically meets the requirements of realtime identification.