I have an Intel RealSense SR300. I'm running Windows 10 (preferred), and Ubuntu.
My situation is: It's been sitting in a box unopened for years, I just busted it out because I need to build a 3D model of a large object that I cannot easily gain physical access to for measurements (it's the underside of the trim on the roof of an old 3-story building, on a damaged section that is unsafe to walk on; I'm basically just hanging the camera over the side on a pole). The camera works. However:
The original SDK included a 3D model builder (in the form of a coding sample application) that could stitch together multiple frames. This SDK was discontinued (around 2016 I think). I downloaded it, the sample works, but it is insufficient for many reasons (including lack of camera settings, and software-imposed limits on volume size, and also it crashes a lot).
The new SDK2 is great. I downloaded it. It also works. However, it no longer includes a 3D model stitching application. AFAICT the feature isn't even provided through the SDK any more, I think it expects users to rely more on 3rd party solutions (e.g. OpenCV) now.
This is a one-off need, and while I am a programmer, I absolutely do not feel like learning the RealSense API and writing a stitching application right now. Plus, I'm willing to bet it exists, and so that brings me to:
Is there any software out there that:
- Can take either a live or pre-recorded stream from an Intel RealSense SR300 camera and stitch the frames together into a 3D model (doesn't have to be solid)
- Can also integrate texture data from the RGB stream
- Allows setting of all camera options, particularly the following if nothing else:
- Laser power
- Motion/range tradeoff parameter
- Post-processing spatial, temporal, and confidence threshold settings
- Does not place software-imposed limits on model and volume size
- Runs on either Windows 10 (preferred), or Linux (with prerecorded data only; I don't yet have the HW working on my Linux machine).
The output 3D model format does not matter as long as it's something relatively well-known, or at least easily machine-readable and well-documented.
Bonus optional features:
- Ability to exclude certain frames from stitching -- to make it easier to recover from a bad scan or a blip in the tracking.
- Relatively stable.
- Free is always a plus. Definitely not over $100 USD.
I'm not concerned with tight system requirements. The machine I'm running it on has plenty of brains and muscle (2x 4TB SSDs in RAID-1, 64GB physical RAM, 8x2 3.2 GHz i9 cores, Quadro RTX 4000).
Also note: I'm not entirely sure what the full set of data export features are in the out-of-the-box viewer software yet, but if there's no software that's directly compatible, but there is software that can work with arbitrary point sets, it might be possible (albeit not ideal) for me to export recorded data and convert it to some more useful form. That's the limit of coding I'm willing to do myself for this project.