Research On Road Target Classification And Recognition Technology Of Vehicular Radar

With the rapid development of artificial intelligence technology,self-driving vehicles have gradually become the focus of attention.Autonomous driving technology can not only improve the safety of road traffic,reduce traffic congestion,but also improve people’s travel experience.In order to perceive the road environment around the vehicle,autonomous vehicles need to be equipped with a large number of sensors,millimeter wave radar is one of them.The core task of using millimeter-wave radar to perceive the road environment is to search,intercept,and track road targets,and classify and identify targets and determine the threat level.The identification and determination results will be used as the decision-making basis for the automatic driving system to control the vehicle.This thesis mainly studies the classification and recognition technology of four types of common road targets using millimeter-wave radar.The four types of targets are nonmotor vehicles,large motor vehicles,small motor vehicles and pedestrians.The dissertation carried out the research on the classification and recognition technology based on the High Resolution Range Profile(HRRP)data and point cloud data of road targets.The specific research contents are as follows:(1)The method of acquiring target information by using millimeter-wave radar to transmit Linear Frequency Modulated Continuous Wave(LFMCW)was studied and radar data of road targets were collected.Firstly,the principle of LFMCW radar ranging,velocity measurement and angle measurement and signal processing flow are studied,and the performance parameters of LFMCW radar are deduced.After that,the experimental platform was built and the HRRP data and point cloud data of road targets were collected in different experimental scenarios,which provided data support for the classification and recognition method in the following text.(2)The classification recognition method based on road target HRRP is studied.Firstly,the magnitude sensitivity and attitude sensitivity of HRRP are analyzed,and the scale and statistical features of the measured target HRRP are extracted as the feature vector of the target.Afterwards,two classifiers were designed using support vector machine and fully connected neural network respectively.By inputting the HRRP of the target into the two classifiers,the average accuracy rates of 82.23% and 82.83% were achieved respectively.(3)The algorithm of millimeter wave radar point cloud clustering is studied.First,the characteristics of different target point clouds are compared using three views of the point cloud and the amplitude sensitivity and attitude sensitivity of the point cloud are analyzed.Afterwards,three clustering algorithms based on data density were studied and applied on the simulation data and measured point cloud data.By adjusting the input parameters of the algorithm and comparing the clustering effect,the optimal input of each algorithm was finally obtained.parameter.(4)The classification recognition method based on road target point cloud is studied.First,the generation method of point cloud three-view data is studied,and it is used as the input of multi-view convolutional network,and the average accuracy rate of 92.94% is obtained,which exceeds the previous two methods.After that,the Point Net++ network using point cloud raw data as input was studied,which further improved the effect of classification and recognition,achieving an average accuracy rate of 97.48%.