Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

The Finer Points: A Systematic Comparison of Point-Cloud Extractors for Radar Odometry

Elliot Preston-Krebs, Daniil Lisus, Timothy D. Barfoot

TL;DR

This paper addresses the problem of how radar point-cloud extraction quality affects ICP-based odometry in autonomous driving. It conducts a comprehensive, controlled comparison of 13 extractors (signal and spatial) using an ICP pipeline across two on-road FMCW radar datasets totaling about 176 km. The findings show that the simple K-strongest extractor delivers the best overall performance, with substantial improvements over the average and the results highly sensitive to extractor tuning. The work highlights the significant impact of front-end processing on odometry accuracy and provides practical guidance for extractor selection to improve radar-based navigation in real-world conditions.

Abstract

A key element of many odometry pipelines using spinning frequency-modulated continuous-wave (FMCW) radar is the extraction of a point-cloud from the raw signal. This extraction greatly impacts the overall performance of point-cloud-based odometry. This paper provides a first-of-its-kind, comprehensive comparison of 13 common radar point-cloud extractors for the task of iterative closest point based odometry in autonomous driving environments. Each extractor's parameters are tuned and tested on two FMCW radar datasets using approximately 176km of data from public roads. We find that the simplest, and fastest extractor, K-strongest, is the best overall extractor, consistently outperforming the average by 13.59% and 24.94% on each dataset, respectively. Additionally, we highlight the significance of tuning an extractor and the substantial improvement in odometry accuracy that it yields.

The Finer Points: A Systematic Comparison of Point-Cloud Extractors for Radar Odometry

TL;DR

This paper addresses the problem of how radar point-cloud extraction quality affects ICP-based odometry in autonomous driving. It conducts a comprehensive, controlled comparison of 13 extractors (signal and spatial) using an ICP pipeline across two on-road FMCW radar datasets totaling about 176 km. The findings show that the simple K-strongest extractor delivers the best overall performance, with substantial improvements over the average and the results highly sensitive to extractor tuning. The work highlights the significant impact of front-end processing on odometry accuracy and provides practical guidance for extractor selection to improve radar-based navigation in real-world conditions.

Abstract

A key element of many odometry pipelines using spinning frequency-modulated continuous-wave (FMCW) radar is the extraction of a point-cloud from the raw signal. This extraction greatly impacts the overall performance of point-cloud-based odometry. This paper provides a first-of-its-kind, comprehensive comparison of 13 common radar point-cloud extractors for the task of iterative closest point based odometry in autonomous driving environments. Each extractor's parameters are tuned and tested on two FMCW radar datasets using approximately 176km of data from public roads. We find that the simplest, and fastest extractor, K-strongest, is the best overall extractor, consistently outperforming the average by 13.59% and 24.94% on each dataset, respectively. Additionally, we highlight the significance of tuning an extractor and the substantial improvement in odometry accuracy that it yields.
Paper Structure (22 sections, 4 equations, 2 figures, 1 table)

This paper contains 22 sections, 4 equations, 2 figures, 1 table.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Radar Odometry
  4. Signal Extractors
  5. Spatial Extractors
  6. Extractor Comparisons
  7. METHODOLOGY
  8. Odometry
  9. Units of Power
  10. Signal Extractors - CFAR Variants
  11. Signal Extractors - Non-CFAR Variants
  12. Spatial Extractors
  13. EXPERIMENTAL RESULTS
  14. Dataset
  15. Fixed Parameters
  16. ...and 7 more sections

Figures (2)

  • Figure 1: Visualized point-clouds of the best- and worst-performing extractors and their percentage average translation errors. The top and bottom rows illustrate extractor performance in the same geometric environment under the differing noise characteristics of the two datasets collected using the Boreas platform, which is depicted in the centre. Gray indicates raw radar data, while red represents the extracted points.
  • Figure 2: Generalized CFAR extractor schema. Diagram is zoomed in on a given reference window capturing cells $x_1$ to $x_N$.