Build a 3D point-cloud map

What you will do

Starting from the trajectory and scans of the FAST-LIO2 lab, you will produce a clean, bounded map suitable for localisation (lesson M4-T13) and for planning.

Before you start

• Output of the lab “Run FAST-LIO2 on a dataset” (registered scans or the saved PCD map).
• Open3D or PCL and CloudCompare for inspection; the voxel filter of lesson M4-T6.

Steps

1. Load the saved map and print point count, extent and file size.
2. Voxel-filter at 0.1, 0.2 and 0.5 m and record the point count and file size for each; choose the resolution that meets a map-size budget you define (for example 50 MB) and the accuracy your mission needs.
3. Remove statistical outliers and, where your data allow, dynamic points (points seen only once or in few scans).
4. Inspect in CloudCompare for double walls, floating points and holes; list defects with their likely cause (timing, extrinsics, dynamics).
5. Save the final map with a text record of the sensor, software versions, configuration and voxel size, ready to be versioned (lesson M7-T19).

# with Open3D (pip install open3d); API as in the Open3D documentation for your version
import open3d as o3d
pcd = o3d.io.read_point_cloud("map.pcd"); print(pcd)
for v in (0.1, 0.2, 0.5):
    d = pcd.voxel_down_sample(voxel_size=v); print(v, len(d.points))
clean, idx = d.remove_statistical_outlier(nb_neighbors=20, std_ratio=2.0)
o3d.io.write_point_cloud("map_clean.pcd", clean)

How to know it worked

• A table of voxel size against point count and file size.
• A cleaned map within your size budget with a defect list.
• A configuration record stored with the map.

If it does not work

References

• Open3D documentation
• PCL documentation
• FAST-LIO repository (map saving option)

Record your progress