Hacker News Reader

Comment by nimrody | Hacker News Reader

Beautiful! Thanks for doing this work.

Can you explain a bit on the process that was used to create this? How do you determine your position inside the pyramid precisely?

(would make for a great dungeon-style game :)

View on HN

I've used the BLK 360 before so some comment here. Essentially you scan at various points within the volume of interest, e.g. you walk around, place the scanner, scan, repeat. Each scan takes a few minutes at high angular resolution. Leica provides software called Register (or Cyclone) to match the scans together. Because you generally don't have GPS (the BLK doesn't have a GPS onboar and GPS is anyway much more inaccurate than the scan resolution - metres versus millimetres), the software has to do some kind of feature matching to stitch the scans. You get "links" between adjacent scanning points, and then you do a big optimisation pass to combine the scans.

This scan matching is by far the most difficult and time consuming bit. Probably OP had to manually align the scans as a first pass and then the software takes over using some algorithm like ICP (iterative closest point).

This is still only "internal" (i.e. scans are correct relative to each other, but you don't know where the full scan is) and you'd have to combine with an external reference point to geo-locate in the world. Doesn't really matter for this because you're just viewing the pyramid on its own, and you're not overlaying on a map. If you were, then usually what you have to do is take several ground control point (GCP) that are known with high accuracy and then reference that in the scan. You could geo-reference these using an RTK GPS or something, but it's quite difficult to get world coordinates at the millimetre scale and it rarely matters if you're that precise as long as the scan itself is consistent.

This video from Leica shows the full workflow for a typical use case (scanning a house with indoor and outdoor points). Note the point where they link inside and outside, around 16 mins in https://www.youtube.com/watch?v=AV0LPKowOXU

Complete newbie here, but wouldn't it be possible to affix some marks on the walls to help with the stitching afterwards?

When working with LIDAR scanning systems (Total systems etc) that generate a point cloud, this is exactly what you'd do. Affix a target or fiducial (a pattern you can apply to a surface that shows up in the scan, either because it's very reflective or because it has a high contrast pattern on it), and use those as reference points to stitch your clouds together.

yeah, this is how things used to be done maybe 10y ago. Spheres (good for 360 symmetry) or checkerboard stickers. Problem is, nobody will probably allow You to put stickers onto the interior of the Great Pyramid.

These days there's a shift towards automatic "stitching" (scan registration) on an accompanying device (tablet or a laptop).

Here's a demo of Trimble X7 (about 3y old product). Full disclosure - I worked on part of this. Not sure how much I can go in detail, but the video shows a pretty good basic demo of how this works:

https://www.youtube.com/watch?v=PApO60lOhlg&ab_channel=Surve...

Awesome, thanks for sharing. We use cyclone for 3D CMS scans of open holes at the mine I work at. Cool to see other use cases. Were you using 3DReshaper before the rebranding?

Thanks for visiting! I'll put together a blog post, but it was similar to the process in my previous work here: https://blog.mused.org/digitizing-luxor-temple-a-virtual-fie...

Similar except I captured with Matterport Capture and then downloaded and aligned data captured with BLK after in Cyclone so that we'd have both sets.

I had spent few weekends in my attempt to automate the process of generating 3D model from pictures. The underlying issue and concern folks have with matterport is their lock in. You are free to upload images to it but not allowed to self host or use their resource to generate the 3D dollhouse.

The regular way requires to feed the photo into multiple layer of software. Example, generate a point cloud, create a mesh from it, clean up the mesh and port it a Fanwood to be consumed (unity. Unreal, etc) this is all manual.

I am genuinely curious how the folks at matterport are able to do it with next to no human input after feeding it 2d pictures.

Reply on news.ycombinator.com