Multiagent Adaptive Points Cloud Clustering For Millimeter Wave MIMO Radar

Multiple-input multiple-output (MIMO) millimeter wave (mm Wave) radar is one of the indispensable sensor for automatic driving, which enables the high-accurate environment sensing. Different with the classical radar system, each target is now detected as much reflection points in MIMO mmWave radar system. To further classify different targets and reduce the tracking overhead, it is necessary to cluster the detection points belonging to the same physical object, whilst traditional algorithms fail to provide the robust clustering performance of common moving targets of MIMO mmWave radars. In this work, we propose a novel multi-agent adaptive points cloud clustering (MAPCC) scheme, which aims to improve the points clustering accuracy and robustness and only requires the clustering radius. Our new proposed MAPCC method firstly divides four epsilon neighborhoods of each detection point based on the clustering radius, and meanwhile develops one strategy to calculate the corresponding similarity. On this basis, we then move each detection point to the epsilon neighborhood with largest similarity, and finally update the label. As illustrated by the numerical simulations and real experiments, our new MAPCC method achieves more accurate and robust point cloud clustering performance comparing to traditional algorithms.