Research On Road Height Limit Detection Technology Based On 4D Radar And Vision

Obstacle detection technology in front of vehicle is one of the important research directions of advanced driving assistant system(ADAS)and driverless system.At present,the commonly used millimeter wave radar and vision fusion detection schemes are based on the common 3D radar,which can not detect the height information of the target.This makes the current road safety detection scheme unable to detect the height limit pole,culvert and other targets.In view of the problems in the detection of target height information,this paper proposes a road height limit detection technology based on 4D radar and vision.In this scheme,4D radar is used to detect the height information of the target in front of the vehicle,and the fusion detection method of vision camera and millimeter wave radar is used to realize the effective detection of the height information of the target in front of the vehicle.From the technical level,this paper uses the millimeter wave radar to collect the state information of the target;then,based on the point cloud information detected by the millimeter wave radar,constructs the region of interest of the target;finally,unifies the radar and vision parts effectively,uses the visual detection algorithm to detect the target in the region of interest,and realizes the effective fusion of the millimeter wave radar and vision.The main contents of this paper are as followsFirstly,the working principle of millimeter wave radar is described,and the key technology of improving the angular resolution of millimeter wave radar MIMO is emphasized.Secondly,the rules of primary selection of effective targets are analyzed,and the invalid signals detected by radar are deleted to reduce the amount of calculation in the follow-up work.Finally,for the consistency test of effective targets,cubature Kalman filter algorithm is selected to detect effective targets We're tracking.Secondly,it introduces the classification of visual detection and recognition using deep learning algorithm,and selects fast r-cnn as the visual recognition algorithm of this paper after comparison;then,it focuses on the working principle of fast r-cnn,introduces the network framework of fast r-cnn and vgg16 network model;finally,it verifies the reliability of the target recognition algorithm by building a deep learning experimental platform.Thirdly,the fusion model of millimeter wave radar and vision sensor is established,the spatial fusion of sensor is completed by using coordinate transformation relationship,and the temporal fusion of millimeter wave radar and vision camera is realized by using interval frame sampling method;moreover,in the process of spatial fusion,due to the lens distortion of camera,Zhang Zhengyou calibration method is used to obtain the internal and external parameters and distortion of camera After completing the fusion between sensors,the target detected by millimeter wave radar is projected to the image coordinate system to construct the region of interest.Finally,the target detection is carried out in the region of interest by using the visual detection algorithm,and the experimental verification of millimeter wave radar and visual fusion detection is completed.