Global map is uniform but too sparse vs. local LiDAR scans — how to keep final global map dense (high-point sensor)?

[mahmood-ul-hasan]

Hi!
I’m mapping indoors with a high-point LiDAR (Airy; very dense raw scans). During run/visualization the local scans are dense, but the final global map looks uniform yet notably sparser than the local data. I’m aiming for a dense final global map that preserves as much of the original point density as feasible.

What I’m seeing

• Local LiDAR sweep (near real time): dense point clouds (as expected for this sensor).
• Final/global map after submapping + global mapping: uniform resolution (no near/far ramp anymore), but still much less dense than the local scans.

Question: which parameters (and stages) directly control the final global map density?
I’d like to confirm which settings actually cap or decimate points written into submaps / the global model (vs. params that only affect registration or internal voxel maps).

My current relevant config (abridged):

"sub_mapping": {
  "submap_downsample_resolution": 0.1,
  "submap_target_num_points": 50000,
  "max_num_keyframes": 15,
  "keyframe_update_strategy": "OVERLAP",
  "keyframe_voxel_resolution": 0.2,
  "keyframe_voxelmap_levels": 3
},
"global_mapping": {
  "submap_voxel_resolution": 0.2,
  "submap_voxel_resolution_max": 0.2,
  "submap_voxel_resolution_dmin": 1000000000.0,
  "submap_voxel_resolution_dmax": 1000000000.0,
  "submap_voxelmap_levels": 1,
  "randomsampling_rate": 1.0,
  "registration_error_factor_type": "VGICP_GPU"
}

Assumptions / attempts so far

• I disabled the distance-based coarsening in global_mapping (fixed voxel size; single level). The global map is now uniform, but still not as dense as the local scans.

• I suspect the effective density is primarily governed by sub_mapping post-processing, e.g.

  • submap_downsample_resolution
  • submap_target_num_points
  • max_num_keyframes (more frames per submap = more raw points)

• I also wonder if there is any final export/downsample step elsewhere that I might be missing.

Could you please advise on best-practice settings for dense global maps with high-point sensors?
Concretely:

1. Which parameters truly limit the number of points stored in each submap and thus visible in the final global map?

   • Is submap_downsample_resolution the main choke point?
   • Does submap_target_num_points hard-cap the stored points per submap (and should I set it to a higher value or -1)?

2. Is there any additional downsampling at the global mapping or export stage that I should disable/tighten?

Thanks a lot

[github-actions]

Please re-open the issue in the GLIM main repository.