毫米波点云生成论文 | 3D Point Cloud Generation with Millimeter-Wave Radar

Kun Qian, Zhaoyuan He, Xinyu Zhang
UCSD
Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies (ACM IMWUT)

原始论文地址: http://xyzhang.ucsd.edu/papers/KQian_UbiComp21_RadarPointCloud.pdf
Video地址：ACM SIGCHI官方频道

本文为毫米波点云生成论文 3D Point Cloud Generation with Millimeter-Wave Radar的阅读笔记, 原载于R.X. NLOS的博客
大量参考了作者的 Pre-recorded presentations for UbiComp/ISWC 2021, September 21–26
笔记难免存在问题，欢迎联系 981591477@qq.com 指正。

内容在CSDN、知乎和微信公众号同步更新

  • CSDN博客
  • 知乎
  • 微信公众号

文章目录

  • 毫米波点云生成论文 | 3D Point Cloud Generation with Millimeter-Wave Radar
    • 1 Introduction
    • 2 困难和解决方法
    • 3 实验与结果
    • 总结

1 Introduction

  • Sensors for Automomous Driving

    • Lidar, Camera and Radar
    • Radar is more robust against bad weathers
  • Limitations of MmWave Radar (2 main drawbacks)

    • Extremely low resolution

      ❌ due to its small form factor

      ❌ only generates intensity maps with strong reflection peaks

    • Blindness due to specular reflection

      ❌ specularly reflected by most objects

  • An Existing Solution: Non-coherent Imaging

    • Fusing measurements along the Radar’s moving trajectory

      ✅ To some extent, alleviate the specular reflection problem (by illuminating from diverse locations)

      ❌ The imaging resolution is still limited by the physical size of the antenna array

      ❌ Cannot be fundamentally improved through spatial sampling

  • MilliPoint: Coherent Imaging

    • Raw radar measurements are directly combined with SAR

      ✅ low-end vehicle radars

      ✅ coherently combines measurements of the radar

      ✅ generate dense and high-resolution 3D point clouds

2 困难和解决方法

  • Enabling SAR with Vehicle Radar
  • Challenge 1: SAR requires accurate tracking of the radar

    • 左图：uniform motion (位置已知)
    • 右图: Variable motion (位置未知)、
    • 结论: Without the knowledage of the radar’s location, the image of the object can be highly distorted