Forward Vehicle Information Recognition Based On Fusion Of Millimeter Wave Radar And Camera

This thesis takes the environmental perception in Advanced Driver Assistance System as the research object,and conducts the research of the information recognition of forward vehicles.Aiming at the deficiencies such as poor hardware stability,low real-time data,high cost,low software processing rate,and poor accuracy of the results of forward vehicle information identification with a single sensor,the vehicle based on the fusion of millimeter wave radar and camera Information recognition method is put forward,which has the advantages of high recognition accuracy and high recognition efficiency.The main research contents are as follows:(1)Screening of effective vehicle targets ahead based on millimeter wave radar.Firstly,the performance and characteristics of millimeter-wave radar are analyzed.Continental ARS-404 is selected as the experimental radar in this thesis.Secondly,the CAN message of the radar is analyzed by the radar analysis protocol to obtain the driving information of the road ahead.The support vector machine is used to train the vehicle prediction model,the vehicle targets are preliminarily selected,the valid targets were selected on the basis of the horizontal distance information and the vertical distance information,and the threshold is set to judge the valid targets.Finally,the proposed effective vehicle target screening algorithm is verified.The results of experiments show that the proposed algorithm can screen effective vehicle targets.(2)Detection of forward vehicles based on machine vision.Firstly,the basic principles of the algorithms Yolov3 and Yolov3-tiny are analyzed.On the basis of the network structure prototype of Yolov3-tiny,a scale is added,and Three-Scale Yolov3-tiny is proposed to achieve accurate positioning and classification of the preceding vehicles.Secondly,data are collected from different scenarios,training and test data sets are produced by using the Label Img gadget,and the K-Means++ algorithm is used to recalculate the Anchor for vehicle detection.The Three-Scale Yolov3-tiny is trained and tested by using the data set,and compared with the standard Faster-RCNN,Yolov3,and standard Yolov3-tiny algorithms.The results show that the average accuracy of the Three-Scale Yolov3-tiny proposed in this thesis is higher than the standard Yolo3-tiny,and can meet the real-time requirements.Finally,the Three-Scale Yolov3-tiny model obtained from training is verified statically and dynamically.(3)Forward vehicle information recognition based on fusion of millimeter wave radar and machine vision.By using the camera’s linear imaging model,Zhang’s calibration method is adopted to calculate the camera’s internal parameters and distortion parameters,so as to achieve the spatial fusion of the millimeter-wave radar and camera.By using multithreading to collect road data,the millimeter-wave radar and camera are integrated in time.On account of the recognition of the same target detected by the camera and the millimeter wave radar,the Kalman filter is used to predict the relative speed and distance of the leading vehicle missed by the millimeter wave radar.Finally,the fusion algorithm of this thesis is verified on the road.The experimental results show that under good weather and with street lights at night,the recognition rates of the vehicle information by using the fusion algorithm in this thesis are 94.7% and 90.8%,which are higher than those by using single millimeter radars :83.2% and 84.3%.