一种改进的毫米波雷达聚类算法

针对毫米波雷达数据均匀性差,数据量小,噪点多等问题,提出一种基于DBSCAN (density-based spatial clustering of applications with noise)的雷达自适应聚类算法.改进算法能够根据K近邻距离和目标反射截面自适应调整聚类半径.首先给出一种聚类半径根据K近邻距离动态调整的机制:目标第K个近邻的距离与阈值相比较,以确定阈值半径取值.再提取雷达提供的目标反射截面,基于该值计算目标假象半径作为聚类半径的补充量.实现根据目标反射截面与数据稀疏程度自适应聚类的效果.将改进算法与不同参数的DBSCAN聚类算法在真实雷达点云数据进行实验对比.相较于选取合适参数的DBSCAN算法,改进算法能够更好适应毫米波雷达点云特征,对行人目标识别准确率提高4.18％,对车辆目标识别准确率提高5.63％.

An improved clustering algorithm for Willimeter Wave radar

A radar adaptive clustering algorithm based on DBSCAN algorithm is proposed aiming at the problems of poor uniformity, small amount of data and more noise points. The improved algorithm can adjust the clustering radius adaptively, and the clustering radius is associated with K-nearest neighbor distance and the target reflection cross section. Firstly, a clustering radius dynamic adjustment mechanism based on K nearest neighbor distance was presented: the distance of the target''s K nearest neighbor was compared with the threshold value to determine the value of threshold radius. Then the target reflection cross section provided by the radar was extracted and the target illusion radius was calculated as the supplement of the cluster radius based on this value. The adaptive clustering is achieved according to the reflection cross section of the target and the sparsity of the data. The improved algorithm was compared with DBSCAN clustering algorithm with different parameters in real radar point data. Compared with the DBSCAN algorithm with appropriate parameters, the improved algorithm can better adapt to the characteristics of millimeter wave radar points. The accuracy of pedestrian target recognition is improved by 4.18%, and the accuracy of vehicle target recognition is improved by 5.63%.