Photofly with Kaidan turntable
is meant to be used for static objects or scenes where you walk around taking lots of pictures at different angles. And then you load them into the Photofly app and you get back a 3d model from the "cloud". But this guy has success with the traditional object movie approach by using a turntable with extra markings to help the program generate correct views.
These sort of 3d from photo of objects have been around for ever but the appeal of Photofly is that it can be totally automatic and with some subjects it can get very accurate, finely detailed results (it can find dense point correspondences, and fine silhouette detection I guess). Some of the core technology comes from Acute3d I think.
People are using Kinects for some scanning applications too. Kinects have a limited depth and detail resolution compared with Photofly type approaches but it is an active (structured light) system and has a built in dimensional accuracy and does progressive scans with realtime feedback in some implementations
.. good for building quick, dimensionally accurate scans of rooms,offices etc
Useful I think for realCG transition movies between realworld panos in a VT.
I have played with this app thinking how to shoot synchronized panoramas and depth map panoramas and it sort of works. You can mount a Kinect onto a pan head and pan it around slowly and it builds up in realtime a 3d model of the room ( 3d shell) - if the walls are not too far away. And you can capture the successive images (visible, depth map and IR) and stitch them if you wanted.
It would be a useful thing maybe if you wanted to do dimensional measurements of objects in a panorama.
I have tried using Photofly with a rotating widely separated stereo pair for pano depth map capture and it fails (it reports cannot find stereo positions or something). I am sure it would start to work if you did multiple stereo (or offaxis single) pano sequences in the one
But generally Photofly does not work great with extensive conventional exterior urban scenes. With single buildings, with natural non-shiny, textures, with some manual input, it can work ok.
It does work well often with bodies, head, clothing -- though it is not good on hair. It works great with bark, cliff surfaces, some aerial sequences.
Depth maps and panoramas are big in robot vision. Depth maps are used a lot in stereo photography, 3d film conversion, content making for autostereo displays and lenticular prints.
You can use depth maps for focus control (eg. Photoshops Lens Blur filter), relighting effects
and various 3d rendering applications -- fog effects etc
-- so it would be nice to be able to make accurate depth maps easily.
Post processing DOF adjustments of panoramas for instance
One thing I have been thinking about is using depth maps for control of stereo parallax and alignment with stereo panoramas. Panoramas show everything so the depth range in the average stereo pano is great - so you have to keep the parallax minimal for many parts of the scene. Would be better to have variable parallax depending on where the user is looking. You could adjust the depth map beforehand or control its density and contrast in realtime
This is quite a bit of research work on stereo panoramas with depth maps
I have been shooting 3d heads with a 11 camera CanonTX1 SDM rig.
Here is one of the heads on Youtube as a flyaround
I can use a single photo from the camera sequence plus a depth map from the model to generate multiple close views for an autostereo display or for depth control for regular stereo views (using Gimpel3d -- great free app for 2d/3d conversion). I think there are intriguing possibilities with historic 360 panoramas converting them into stereo with apps like Gimpel.
- Wow, Peter, lots of food for thought here. Thanks for the fascinating post. I feel sad not to be younger and with a greater budget so I could follow up some of these possibilities.
Sent from my iPad
On Aug 5, 2011, at 7:46 AM, "panovrx" <panovrx@...> wrote:
> is meant to be used for static objects or scenes where you walk around taking lots of pictures at different angles. And then you load them into the Photofly app and you get back a 3d model from the "cloud". But this guy has success with the traditional object movie approach by using a turntable with extra markings to help the program generate correct views.
Lots of good stuff deleted.