search

ramdrop / autoplace

Implementation for the paper: AutoPlace: Robust Place Recognition with Single-chip Automotive Radar
104 stars 23 forks source link

Does this method work with 3d radar point clouds with height info? #7

Closed JzHuai0108 closed 1 year ago

JzHuai0108 commented 1 year ago

Dear Sir,

The paper experiments used radar data from nuScenes. What is your opinion on using your method for 4D radar point clouds? These point clouds have height info and often a 120 deg Horizontal FOV and 30 deg vertical FOV.

Thanks,

ramdrop commented 1 year ago

Hello, AutoPlace is primarily designed for 3D (XY+V) radar points clouds, the following factors should be considered when it is applied to 4D (XYZ+V) radar points:

  1. Feature extractor: a 3D CNN backbone might be needed.
  2. Dynamic points removal: build a 3D mathematical relations.
  3. RCSHR might be adopted as it is.
JzHuai0108 commented 1 year ago

After brief reading of autoplace paper, I agree with your thoughts. Thank you for the comments.