基于毫米波雷达的近邻目标聚类和跟踪  

Neighboring target clustering and tracking based on millimeter-wave radar

作　　者：张春杰[1] 赵佳琦 陈奇 ZHANG Chunjie;ZHAO Jiaqi;CHEN Qi(School of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学信息与通信工程学院,黑龙江哈尔滨150001

出　　处：《应用科技》2024年第5期219-227,共9页Applied Science and Technology

基　　金：黑龙江省自然科学基金联合引导项目(LH2020F019)。

摘　　要：为解决毫米波雷达在对多目标跟踪时目标近邻聚类失败导致的目标数目低估和跟踪精度下降问题,提出一种基于概率假设密度(probability hypothesis density,PHD)滤波器的量测集联合划分方法。利用带噪声密度空间聚类(density based spatial clustering of applications with noise,DBSCAN)算法对采集到的量测集进行初步划分。通过PHD滤波器的预测值判断初步划分的点云簇是否存在重叠簇。针对重叠簇,利用滤波器预测值改进高斯混合模型(Gaussian mixed model,GMM)聚类算法并进行子划分。在仿真和实际环境中进行算法测试,仿真结果表明,所提算法能正确划分并跟踪近邻的目标,相比其他算法具有更好的跟踪精度。实测结果进一步验证了该算法能够成功识别近邻目标数量并跟踪,具有一定的工程实践意义。To address the issue of underestimated number of targets and decreased tracking accuracy caused by the failure of neighboring target clustering in millimeter-wave radar-based multi-object tracking,a measurement set joint partitioning method based on the Probability Hypothesis Density(PHD)filter is proposed.Initially,the collected measurement set is preliminarily partitioned using the Density Based Spatial Clustering of Applications with Noise(DBSCAN)algorithm.The predictive values of the PHD filter are then employed to determine the presence of overlapping clusters in the preliminary partitioned point clouds.For overlapping clusters,an improved Gaussian Mixed Model(GMM)clustering algorithm is applied,leveraging the predictive values of the filter,to perform sub-partitioning.Algorithm testing was conducted in both simulation and real-world scenarios,with simulation results indicating that the proposed method could accurately divide and track neighboring targets,exhibiting superior tracking precision compared to other algorithms.The experimental results further confirm the algorithm's capability to successfully identify and track the number of nearby targets,highlighting its practical significance in engineering applications.

关 键 词：毫米波雷达 扩展目标 多目标 概率假设密度 带噪声密度空间聚类 联合划分 近邻目标 高斯混合模型聚类 

分 类 号：TN953[电子电信—信号与信息处理]