Research On Multi-density Clustering Algorithm Of Millimeter Wave Radar

As a common sensor,millimeter-wave radar is gradually becoming the preferred sensor for automotive radar and obstacle avoidance radar due to its large detection range,high detection accuracy,and strong anti-interference ability.At present,automobiles and security manufacturers in various countries are stepping up research on millimeter-wave radar.As the carrier frequency of the millimeter-wave radar gradually migrates from 24 GHz to 77 GHz,the resolution of the millimeter-wave radar is getting higher and higher,and the number of reflected echoes of the same target also increases.Therefore,it is necessary to adopt a certain clustering strategy to simplify the number of targets and facilitate subsequent tracking and processing.On the other hand,the clustering algorithm can eliminate noise and reduce the interference of irrelevant interference points on target detection.With the maturity of data mining,there have been many developments and improvements in clustering algorithms.However,there are relatively few studies on clustering algorithms for millimeter wave radar features,and there is a lot of research space in this field.Based on this,this paper studies the clustering algorithm of millimeter wave radar,which mainly includes the following parts:1.Study the data distribution characteristics of millimeter-wave radar detection targets and obtain the distribution laws of detection targets.This paper introduces the DBSCAN algorithm,a general scheme for density clustering,and verifies the effectiveness of the clustering effect of the DBSCAN algorithm on common uniform data sets.In addition,the clustering performance of the DBSCAN algorithm under the distribution of multi-density data sets is discussed.2.Aiming at the shortcomings of the DBSCAN algorithm in a non-uniform,multi-density data set,a new algorithm developed based on the DBSCAN algorithm is introduced.The application of the data partitioning idea in the DBSCAN algorithm is introduced,and the method of converting global parameters into multiple local parameters is used to improve the shortcomings of the DBSCAN algorithm.A GDBSCAN algorithm combining the idea of grid clustering and density clustering was proposed;the idea of shared nearest neighbors was introduced.By transforming the Euclidean distance between targets into shared nearest neighbors,an improved SNN algorithm was proposed.,Improved the performance of the DBSCAN algorithm.3.Other density clustering algorithms for multi-density distribution are introduced.OPTICS algorithm and DPC algorithm also provide solutions for multi-density and uneven distribution.Here,in view of the shortage of DPC algorithm in data allocation and noise processing,an improved NN-DPC algorithm is proposed.4.Introduced two kinds of effectiveness evaluation methods commonly used for clustering results-external evaluation and internal evaluation.The algorithm under complex distribution environment is simulated and verified.Based on the results of the internal evaluation and the cluster simulation results of the data set,the above algorithms are summarized.