Effects of FMCW radar point-cloud pre-processing on ICP matching performance

Abstract

Radar-based sensing in the mmWave spectrum spans a wide range of applications beyond the traditional automotive industry, such as indoor environment sensing for robotics, smart home systems, and industrial applications. The single-chip FMCW radar enables environment perception while preserving privacy and is reliable even in vision-denied environments. Mobile agent localization, mapping and ego-motion estimation are areas where the usage of such sensor has been increasing steadily. Various approaches for localization and ego-motion estimation were explored in the past, including the Iterative Closest Point (ICP) algorithm. In this paper, we examine the effect of filtering radar targets based on range and the vertical coordinate in an indoor setting, where ghost targets emerging from multipath propagation are a common occurrence. As the number of detected targets from a single-chip radar is usually low, we evaluate the effect of z-densification method on the radar data. We investigate the effect of the proposed techniques on the matching performance of the ICP in position as well as orientation.