Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DenserRadar: A 4D millimeter-wave radar point cloud detector based on dense LiDAR point clouds

Zeyu Han, Junkai Jiang, Xiaokang Ding, Qingwen Meng, Shaobing Xu, Lei He, Jianqiang Wang

TL;DR

This work tackles the sparsity and noise of 4D mmWave radar point clouds by generating dense 3D occupancy ground truth from stitched multi-frame LiDAR data and training DenserRadar to produce denser, more accurate radar detections. The DenserRadar network employs a Doppler-aware 3D U-Net backbone with a cross-attention module and a weighted dice–focal loss to enhance reconstruction of radar occupancy, supervised by the dense LiDAR ground truth. Experiments on the K-Radar dataset demonstrate improved density and accuracy over CFAR-based and learning-based baselines, validating the efficacy of dense supervisory signals. The approach has practical implications for autonomous driving perception in challenging conditions and opens avenues for further enhancement via diffusion models and downstream task integration.

Abstract

The 4D millimeter-wave (mmWave) radar, with its robustness in extreme environments, extensive detection range, and capabilities for measuring velocity and elevation, has demonstrated significant potential for enhancing the perception abilities of autonomous driving systems in corner-case scenarios. Nevertheless, the inherent sparsity and noise of 4D mmWave radar point clouds restrict its further development and practical application. In this paper, we introduce a novel 4D mmWave radar point cloud detector, which leverages high-resolution dense LiDAR point clouds. Our approach constructs dense 3D occupancy ground truth from stitched LiDAR point clouds, and employs a specially designed network named DenserRadar. The proposed method surpasses existing probability-based and learning-based radar point cloud detectors in terms of both point cloud density and accuracy on the K-Radar dataset.

DenserRadar: A 4D millimeter-wave radar point cloud detector based on dense LiDAR point clouds

TL;DR

This work tackles the sparsity and noise of 4D mmWave radar point clouds by generating dense 3D occupancy ground truth from stitched multi-frame LiDAR data and training DenserRadar to produce denser, more accurate radar detections. The DenserRadar network employs a Doppler-aware 3D U-Net backbone with a cross-attention module and a weighted dice–focal loss to enhance reconstruction of radar occupancy, supervised by the dense LiDAR ground truth. Experiments on the K-Radar dataset demonstrate improved density and accuracy over CFAR-based and learning-based baselines, validating the efficacy of dense supervisory signals. The approach has practical implications for autonomous driving perception in challenging conditions and opens avenues for further enhancement via diffusion models and downstream task integration.

Abstract

The 4D millimeter-wave (mmWave) radar, with its robustness in extreme environments, extensive detection range, and capabilities for measuring velocity and elevation, has demonstrated significant potential for enhancing the perception abilities of autonomous driving systems in corner-case scenarios. Nevertheless, the inherent sparsity and noise of 4D mmWave radar point clouds restrict its further development and practical application. In this paper, we introduce a novel 4D mmWave radar point cloud detector, which leverages high-resolution dense LiDAR point clouds. Our approach constructs dense 3D occupancy ground truth from stitched LiDAR point clouds, and employs a specially designed network named DenserRadar. The proposed method surpasses existing probability-based and learning-based radar point cloud detectors in terms of both point cloud density and accuracy on the K-Radar dataset.
Paper Structure (19 sections, 8 equations, 4 figures, 3 tables)

This paper contains 19 sections, 8 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. 4D mmWave Radar in Autonomous Driving
  4. 4D mmWave Radar Point Cloud Generation
  5. Methodology
  6. Overview
  7. Ground Truth Generation
  8. The DenserRadar Network Architecture
  9. Input 3D Convolution
  10. 3D U-net-based Backbone
  11. Output 3D Deconvolution
  12. Cross-attention Mechanism
  13. Loss Function
  14. Experiments
  15. Experimental Setup
  16. ...and 4 more sections

Figures (4)

  • Figure 1: The overview of the whole algorithm.
  • Figure 2: The ground truth generation pipeline.
  • Figure 3: The DenserRadar network architecture. The size of each cube approximately represents its relative resolution.
  • Figure 4: The qualitative point cloud comparisons between our DenserRadar algorithm and the CA-CFAR algorithm, accompanied by images and dense 3D occupancy ground truth point clouds for reference. Each arrow in the figures represents a length of 10 meters.