Research On Vision And Radar Data Fusion Technology For AEB System

The popularity of automobiles has greatly facilitated people’s travel,but uneven driving skills have caused hidden dangers to automobile safety and high traffic accident rates.As one of the main active safety measures,automatic emergency braking(AEB)system can automatically determine the safety status of the car and take corresponding measures when necessary to ensure driving safety.In this paper,the data fusion technology of vision and radar is investigated for the perception part of AEB,and the corresponding collision avoidance strategy is developed.Firstly,we study the target detection method based on millimeter wave radar.After studying the working principle of millimeter wave radar in detail,the nu Scenes dataset is selected as the data base of this paper for the research object of this paper;the raw data of millimeter wave radar is studied and read,and the method is designed to filter out the null targets,invalid targets and stationary targets among them,and to judge the validity of the filtered targets.Then we studied the forward target recognition method based on machine vision,and finally chose YOLOv3 algorithm as the target recognition algorithm after comparing various vision processing methods;combined nu Scenes,VOC2012 dataset and some custom images,combined into a self-built dataset and trained YOLOv3 algorithm accordingly,and finally verified the trained algorithm.The results show that the algorithm can effectively recognize the targets,with an accuracy of 91.2%for vehicle recognition and 90.6% for pedestrian recognition.Then the target detection algorithm based on the fusion of radar and vision data is studied,and the world coordinate system,radar coordinate system,camera coordinate system and pixel coordinate system are established respectively,and the coordinate conversion between each coordinate system is completed to realize the spatial data fusion of sensors;the time fusion of sensor data is completed according to the time stamps of sensor data;the target-level fusion is used,and the data are associated according to the Marxian distance and joint probability The results show that the fusion algorithm is 2.2% and 1.82% higher than the single vision detection algorithm for vehicle and pedestrian targets,respectively.Finally,the collision avoidance strategy of automatic emergency braking is studied.In view of the limitation that the traditional collision avoidance strategy cannot cope with the complex environment,a collision avoidance strategy based on the target motion trajectory is innovatively proposed,and the prediction of the target motion trajectory is completed by using Kalman filter.The experimental results show that the method can get the corresponding desired deceleration speed for different vehicle models,different targets and different vehicle speeds.The data fusion strategy of vision and radar and the automatic emergency braking collision avoidance strategy based on the target motion trajectory proposed in this paper have certain practical value for reducing the occurrence rate of traffic accidents and ensuring the safety of driving,and have certain theoretical reference value for the development of automatic emergency braking system of current vehicles.