Research On Driving Behavior Recognition Technology Based On FMCW Radar

Cars have become one of the main means of mobility for people.Driving is an important part of our life.Every year,there are countless losses caused by road traffic accidents,and operation of the driver has always been one of the main causes of car accidents.Therefore,driving detection is playing a vital role in protecting people’s lives and property.With regard to research on driving detection,current methods are based on acquiring physiological information,vehicle state information,and head visual information to determine the state of the driver.However,these methods have shortcomings such as being aggressive to the driver,being susceptible to environmental influences,and personal privacy issues.In order to overcome these shortcomings and deficiencies,this article will use a radar system to analyze and identify driving behaviors.In this paper,a linear frequency-modulated continuous wave radar system is used to collect radar echo data of driving behaviors,which can be analyzed and processed to determine the kind of 7 common behaviors that may occur during driving.The main work of this article is as follows:1.The linear frequency modulation continuous wave radar system is introduced.Then the system structure and working principle are explained.And the last are the analyze of the radar echo signals of driving behavior and the pre-processing process.2.According to the Doppler information generated by the driver relative to the radar sensor,the radar echo signals are analyzed and processed from the time domain and the distance domain.The time Doppler and distance Doppler characteristics of the driving behavior are extracted.Finally,driving behaviors are classified with classifiers.3.Aiming at the difficulty of threshold selection and the tedious feature extraction during Doppler feature processing,a driving behavior recognition method based on convolutional neural network is proposed.First a time-frequency map data set of driving behavior is established.Then the network structure is built,which is trained and optimized with the time-frequency map data set.Finally a convolutional autoencoder is introduced to initialize the convolutional layer of the network,which solves the problem caused by small amount of the data set.And the result shows that the accuracy of driving behavior recognition is improved.4.The experiment for driving behaviors is designed to collect radar data,which is analyzed to classify seven driving behaviors.The experimental result shows that the time Doppler processing result is 93.7% and the distance Doppler’s is 91.3%.The classification accuracy of the improved convolutional neural network reaches 98.9% which effectively demonstrates the feasibility of the implementation of the system scheme.