Research On Data Association And Fusion Algorithm Of Multi Millimeter Wave Radars

With the rapid development of autonomous driving technology,the use of radars for environmental perception have become a research hotspot.Because millimeter wave is longer than light wavelength.Millimeter wave has a stronger ability to penetrate objects,its survival and reliability are unmatched by other sensors.Millimeter wave radar plays a very important role in vehicle environmental perception.The detection range of a single millimeter wave radar is limited and the measurement obtained are inaccurate,so multiple radars are used to detect the environment.When radar is used to measure the environment,the number of targets in the environment are unknown.Thus it is impossible to determine whether the measurement data are generated by the targets or obstacls.Furthermore,it's necessary to use data association technology to associate the measurement data obtained by millimeter wave radar with the historical targets,and to update the status of the historical targets.The original output data of the millimeter wave radar are measurable data at the point cloud level,which can output measurable data at the target level.This paper studies these two types of data separately.When studying data association tracking algorithm of point cloud data,x,y,z distance values of the point cloud data is used as the input of the clustering algorithm.Improved Kernel density estimation's K-means clustering algorithm is used to cluster point cloud data.The simulation results show that the improved method can increase the correct number of clustering target judgments.The feature information of the clustering point cloud cluster are used,and the feature information are used in the data association algorithm.Global error minimum data association algorithm is proposed to associate point cloud data,and then Kalman filtering is used to update the state of the point cloud targets.When the data association,tracking and fusion at the target level measurement data are studied various measurement information is fully considered.The degree of membership function is used to calculate the degree of subordination between each modal measurement information of the candidate measurement and the corresponding modal information of historical targets.In order to know the importance of each modal measurement in association matching of the candidate measurement,both the subjective and objective weight calculation methods is used to weight the measurement information of each modal of the candidate measurement.The subjective and objective weight of each modal measurement information are combined to obtain the combined weight value.The degree of subordination is weighted by the combined weight value,the comprehensive similarity matrix can be obtained between historical targets and candidate measurement.The principle of maximum comprehensive similarity is used to match the historical targets and candidate measurement.The simulation results show that the accuracy of data association using combined weighting method is slightly higher than that using single weighting method.If a historical target is successfully correlated with the measurements obtained by multiple radars,D-S evidence theory is used to perform data fusion on the historical target.Kalman filtering is used to update historical targets state values.The algorithms proposed are simulated and analyzed,and vehicle experiments are carried out on the data association,tracking and fusion algorithm at the target level.The simulation and experimental results indicate the algorithm proposed can get good result in measurement information association and the targets tracks.