Skip to content

README.md

Latest commit

 

History

History
100 lines (72 loc) · 3.71 KB

README.md

File metadata and controls

100 lines (72 loc) · 3.71 KB
Raw

Doppler Correspondence



Doppler Correspondence is a framework for establishing correspondence in 4D radar odometry

Resources

  • 📄 Paper PDF
    • Doppler Correspondence: Non-Iterative Scan Matching With Doppler Velocity Based Correspondence, Jiwoo Kim, Geunsik Bae, Changseung Kim, Jiwoo Lee, Woojae Shin, Hyondong Oh, [RSS] 2025.
  • 🎬 Presentation Video

Doppler Correspondence Demo

4D Radar Scan Matching With Doppler Correspondence

🍎 Tutorial (Python)

Brief overview of how Doppler Correspondence works and how it differs from closest-point correspondence from ICP. An demo is available in the following folder: 📂 example/example.ipynb

Table of Contents

  • Closest-Point Correspondence

  • Doppler Correspondence

  • ICP vs. Doppler Correspondence (non-iterative)

  • Doppler Correspondence + RANSAC

  • Odometry Example With ICP, Doppler Corr, Doppler Corr + RANSAC

🍎 Closest Point Correspondence vs Doppler Correspondence

Visualization ICP correspondence and Doppler Correspondence:

3D Zoom

ICP Animation Doppler Animation

Error Comparsion

Error Comparison

This figure shows the translation error distribution of both methods evaluated on the Loop1 sequences from the NTU4DRadLM dataset.
You can choose the correspondence strategy via the lambda_doppler parameter in config/config.yaml:

  • lambda_doppler: 0.0 → Pure Closest Point (ICP)
  • lambda_doppler: 1.0 → Full Doppler Correspondence
  • lambda_doppler between 0.0 ~ 1.0 → Hybrid approach, which often yields the best results by balancing geometric and Doppler cues.

🍎 Usage

Dependencies

This package requires the following dependencies:

  • Eigen3 ≥ 3.3
  • PCL (Point Cloud Library)
  • ROS (Noetic)

We have tested this package on Ubuntu 20.04 and Intel(R) Core(TM) i7-14700KF.

ROS (C++)

# Clone the repository inside your catkin workspace
cd ~/catkin_ws
mkdir -p src && cd src
git clone https://github.com/Tars0523/Doppler_Correspondence.git
# Build the workspace
cd ~/catkin_ws
catkin build
# Source the environment
source devel/setup.bash
# Launch the node with RViz
roslaunch Doppler_Correspondence run.launch

📊 Dataset

Experiments in this repository are conducted using the NTU4DRadLM 4-D Radar-centric Multi-Modal Dataset.
Many thanks to Jun Zhang et al. for releasing this challenging dataset to the community.
https://github.com/junzhang2016/NTU4DRadLM