Getting into 3d scanning
polycounter lvl 11
I've been looking into 3d scanning lately but I'm relatively new to it.
If anyone here has any experience with 3d scanning, what would you recommend and what have you found from your experience? If you have any thoughts on setups or any advice, please share.
Ideally I would like to keep things as inexpensive as possible while still keeping the data reliable enough that it can be used (even if there needs to be some cleanup).
In general I haven't had good experience with photogrammetry. I'm really not a fan of Autodesk one, 123D Catch, where you upload the pictures to the cloud. That seems more geared to hobbyist 3d printing and sharing, and the results weren't very good. Things might have changed but I'm still apprehensive about it. It gave similar quality to using the old kinect to build a mesh (also lacking definition and accuracy).
Recently I've tried doing a budget version of an Agisoft PhotoScan setup. I made a light tent and used a rotating christmas tree stand to create a slowly rotating pedestal. I was hoping that I could get away with using one Canon t2i for the setup, at least to test things out, but even when the data was good the photogrammetry wasn't giving great results.
It was also incredibly time consuming. Each image needed to be masked for best results, and while I'm sure this could be automated with a good setup, I had not prepared for it. The processing of the data also takes a really long time if you're trying to get decent quality, and you aren't even sure if your scan will come out.
I looked into it and it seems that the people who use photogrammetry have carefully calibrated and expensive setups. It also seems that there's a general awareness that unless you have a good setup you're not likely going to get great results (even the really expensive setups require a decent amount of cleanup and image processing to get finer details), but they can be instant and include color data, unlike other scanners.
Ten24 has some good looking results using structured light scanning for a high-quality mesh along with a smaller scale Agisoft setup for the color data. http://www.ten24.info/?p=815
Structured light scanners like the Artec ones are generally very, very expensive. Hand held 3d scanners cost as much as a small car. Oddly the actual technology behind the scanners isn't necessarily expensive or difficult to implement. They work by projecting fringe patterns and taking stereoscopic images of them.
I looked around and it seems there are several open source efforts for structured light scanner setups. This is one of them http://www.3dunderworld.org/ using 2 Canon EOS cameras and a projector. It provides color data as well. I think this will be the next thing I look into, I already have one Canon camera, I could easily rent another for $35 or so and a projector would be good to get anyway. Any reason not to go that route? Does anyone have feedback on structured light scanning?
If anyone here has any experience with 3d scanning, what would you recommend and what have you found from your experience? If you have any thoughts on setups or any advice, please share.
Ideally I would like to keep things as inexpensive as possible while still keeping the data reliable enough that it can be used (even if there needs to be some cleanup).
In general I haven't had good experience with photogrammetry. I'm really not a fan of Autodesk one, 123D Catch, where you upload the pictures to the cloud. That seems more geared to hobbyist 3d printing and sharing, and the results weren't very good. Things might have changed but I'm still apprehensive about it. It gave similar quality to using the old kinect to build a mesh (also lacking definition and accuracy).
Recently I've tried doing a budget version of an Agisoft PhotoScan setup. I made a light tent and used a rotating christmas tree stand to create a slowly rotating pedestal. I was hoping that I could get away with using one Canon t2i for the setup, at least to test things out, but even when the data was good the photogrammetry wasn't giving great results.
It was also incredibly time consuming. Each image needed to be masked for best results, and while I'm sure this could be automated with a good setup, I had not prepared for it. The processing of the data also takes a really long time if you're trying to get decent quality, and you aren't even sure if your scan will come out.
I looked into it and it seems that the people who use photogrammetry have carefully calibrated and expensive setups. It also seems that there's a general awareness that unless you have a good setup you're not likely going to get great results (even the really expensive setups require a decent amount of cleanup and image processing to get finer details), but they can be instant and include color data, unlike other scanners.
Ten24 has some good looking results using structured light scanning for a high-quality mesh along with a smaller scale Agisoft setup for the color data. http://www.ten24.info/?p=815
Structured light scanners like the Artec ones are generally very, very expensive. Hand held 3d scanners cost as much as a small car. Oddly the actual technology behind the scanners isn't necessarily expensive or difficult to implement. They work by projecting fringe patterns and taking stereoscopic images of them.
I looked around and it seems there are several open source efforts for structured light scanner setups. This is one of them http://www.3dunderworld.org/ using 2 Canon EOS cameras and a projector. It provides color data as well. I think this will be the next thing I look into, I already have one Canon camera, I could easily rent another for $35 or so and a projector would be good to get anyway. Any reason not to go that route? Does anyone have feedback on structured light scanning?
Replies
Photogrammetry is used in place of laser scanning by professionals in architecture, archeology...
With a correct calibration, it's very reliable. I think it was like 1cm deviation for 10 meters ? It's cheap, takes less place, have better colors, but ofc less precise than laser.
But it highly depends of your pictures quality. You say you used a rotating pedestal. With photogrammetry, you should turn around your model, and move into space. Lighting must not change over the model. Nothing should move but you.
If you stand still and rotate the camera around, it won't work either.
Materials with high reflections won't work well...
Most of the time, you shouldn't have to mask the pictures. Also the cloud generation is pretty quick i think, so you have a preview. Yeah, you'd better have a cutting edge processor for the mesh generation. Still cheaper than a 30k $ scanner.
I have no info about structured light scanning, except it has to be done in a studio i guess. So it won't beat photogrammetry for exterior models. So in the end, and as usual, it depends of what you want to do with it.
I understand that the lighting isn't supposed to change (that's what the light tent was for, although it wasn't perfect by any means). If I wanted to do a better interior setup I'd need many, many more cameras, which is too steep for my own personal work, nevermind a test.
Yeah all the drawbacks for photogrammetry that you mentioned are factors to why I'd like to investigate other things. Even if the PhotoScan results came out great, processing takes a lot of time and the materials have a major effect on the potential quality.
Ideally masking wouldn't be an issue, and it's probably not absolutely necessary if you have a good setup, but I wanted to make sure that I'd get a good test capture so I went through with it. Although, in practice I wouldn't want to wait for a mesh to process just to find out that I should have masked the images. Generating the mesh (if you're using some of the higher settings) can take 7hours+ of processing time (and that's with a relatively frugal set of pictures). The details aren't always very good. I think it really depends on the subject. There's only so much you can do when you're tracking unique points of color based on arbitrary light information.
For my needs keeping things inside would be fine. I still intend to capture some exterior things for textures and references and such, and for that I would use something like PhotoScan, but my main goal is to work towards getting some decent quality, full-body scans, ideally good enough to get deformations from them (so not-just in the ballpark of the shape). Structured light scanning requires a dark room, which would be fine.
Right now I'm apprehensive because I haven't heard much. Most studios I've read about seem to do PhotoScan captures. I can see maybe part of the reason for that being that structured light scanning takes an amount of time for the capture. It's not an instant, action-capture like you could get with a multi-camera photogrammetry setup. It also doesn't do the entire 360 of the model all at once, so there would need to be some stitching the meshes together (but that's still doable). Currently I don't think getting color info would be an issue, but historically it might have been and still might be for certain scanners.
The 3dunderworld seems promising given it only requires two Canon EOS cameras and a projector. Much less steep than a $30k scanner or 60 Canon cameras.