Good evening all,
I’m going to visit a future office space to rent and we didn’t want to to bug with measurements and instead simply take a cellphone (Samsung S23 RE + Gimball) and film around the office space so we can create a point cloud and have the measurements and design/organize the office space.
My question is :
Is it possible to create a point cloud with a video from a Samsung S23 FE ?
Thank you in advance !
Trying to learn the limits of photogrammetry with WebODM
You will need to think about how you will reliably georeference the video and resultant images, and scale them appropriately.
Something like OpenCamera will write GPS positions to an SRT file, but getting reliable GPS data indoors is iffy at best.
What if the camera location were “faked”? Just compute relatively precise GPS locations, from a single “faked”, GPS location. Measure 2, 3 or 4 feet in a known direction, and compute a second(or and a third), relatively very accurate(?), GPS location, then take overlapping photos from those “fake”, GPS locations. It would just be a matter of the EXIF files, right?
Yes, that could work.
Ok i will try it tomorrow - I was just curious to know if there’s a setting i need to be aware of so i wont forget to activate it or something.
But yes, GPS indoors isnt wow
But i think with other softwares like Recap we can match points inbetween photos and that helps to generate a 3d point cloud.
Will keep you posted if my results or problems !
Thank you everyone, appreciate the input !
That reminds me of a project we did some years ago.
We made two pointcloud files.
For the inside of the concrete construction we used 250 50MP panoramic images and we created our own ‘internal GCP grid’ to scale the pointcloud.
On the inside we defined a “0,0,0 x,y,z” location and with a laser measurement tool we calculated the GCP positions in respect to this 0,0,0 location.
Processing back then was done in a commercial application (I switched to WebODM later) but results were very impressive. This gave a nice scaled Pointcloud, but not Georeferenced.
With Cloudcompare I could place the “indoor pointcloud” inside the “outside pointcloud”. So although Im not sure how the results would look like in WebODM I know there are possibilities. Maybe nice to test this again with the WebODM capabilities.
Unfortunately I’m not at liberty to share the inside results.
As far as this is could be a very common use case, resolvable with good but expensive pro instruments or with questionable others at less cost or just in the traditional way with a meter, if exploiting your experience we could sort out a kind of workflow and find an acceptable compromise maybe should be nice for other odm users, me first.
Forgive me if I try to recap in few random thoughts:
as I read from your replies there can be 3 approachs
I’ve done a part bedroom PC with my old Samsung S8, so I would think it should be possible with your more modern phone too.
Georeferencing was an issue though, it did turn out with correct orientation, but I didn’t attempt any measurements, which would no doubt have been incorrect.
3D short-range Lidar is not very expensive … and much more precise than 0.5cm,
I have taken low level photos of my project(30-40 feet), with a tilted camera, and can see under a porch in WebODM, and I can see through the void of the building interior, to see the other sides, but I’ve never had good luck with walking around the building and with WebODM including those photos, to get around the building … I gave up years ago … I should try it again …
Here’s a mock-up I did of my living room wall propose shelving with existing wall art and a little of the couch:
Do I understand this thread correctly, you can use your phone to make “ground view images” like street view and feed that into the process along with drone images which are RTK corrected and the webODM can use the street view images to generate even better cloud point data so you don’t only have point cloud generated from above, but even using the street view for proper facade generation of point cloud?
I did something like this with the interior of my sailboat(for equipment reference), it wasn’t pretty …
Hi @cyberflow,
You can go the photogrammetry route with a phone but even if you get a good quality scan there are often issues with orientation that are quite a pain, like they often render sideways or upside down and potree won’t let you rotate them into the correct orientation (see discussion here). This isn’t as much of an issue if you only need the model and are planning to post-process it on your computer.
For a quick scan from your phone that you can measure, I’d recommend looking at polycam. On the iphone at least, it is quite easy to scan a large space and then measure things on it or output a blueprint. There is even a “room scan” feature (for the iphone at least) that outputs a nice clean floorplan with measurements and all.
I’d note that I have an iphone with lidar though and I haven’t used the android polycam, but I’m pretty sure it works as well and their options for doing stuff with the model are really good.
There are a bunch of other apps like Scaniverse and Sitescan, but I haven’t used them enough to provide an opinion and polycam works really well (and is mostly free).
Polycam also has a pretty good gaussian splat editor that now lets you measure things within the gaussian splat, so that’s definitely worth looking into as well (and often makes super nice looking models and is fund to play with)
One last note: If you do want to make a gaussian that is scale accurate, make sure you measure a few (obvious and straight) things in the scene when you are one site. Polycam now has a feature that let’s you rescale the gaussian to accurate dimensions, but you need to know the length of things in the scan to do that correctly.
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.
