starseeker (Nov 03 2019 at 23:53): starseeker (Nov 03 2019 at 23:53):This sounds nifty - anyone played with it? https://blog.kitware.com/3d-reconstruction-from-smartphone-videos/
Shubham Rathore (Nov 04 2019 at 07:08):This sounds nifty - anyone played with it? https://blog.kitware.com/3d-reconstruction-from-smartphone-videos/
Ah! looks great. Had done the same with the TSDF pipeline some time back, but didn't use the disparity derived depth. Thanks to the TOF sensors in new gen smart phones. :)
Sean (Nov 04 2019 at 21:28):No, but looks very similar to https://alicevision.org which I've been keeping an eye on. It's photogrammetry, similar principles (i.e., typically low-count images instead of video).
Sean (Nov 04 2019 at 21:32):There's definitely some other similar projects using depth sensor devices -- that was big 5 or so years ago when Google had a small tablet (project Tangerine or something I think?).
Sean (Nov 04 2019 at 21:40):Following the research, biggest challenge has been point registration and localization. For example, scanning around something 360, the start points and end points end up misaligned. Doesn't matter much at the resolution they're showing in the kitware demo, but it comes into play with bigger models (e.g., a car or room etc). Impressive work regardless, looks fun to play with.
Last updated: Oct 09 2024 at 00:44 UTC