Hi,
were from Filmkademie Badenw?ttenberg in Germany. Currently there are two Diploma-Projects, which decided to base there "Modelling"-Workflow on Photogrammetry.
Here is a show Video which describes our approach:
http://vimeo.com/97773359As you can see, we've build our FilmSets as Miniatures and Photographed them with a semi-automated Turntable.
For each Prop we shoot between 20 and 60 Turns, with different CameraAngles and Prop-Orientations. eg.: TopView+StandingProp, TopView+LyingProp, SideView+LyingProp, ....
and also a couple of DetailTurntables.
We tried to avoid DepthOfField by choosing the highest possible Aperture, even trough it might till appear in some Images.
Next Step is, to generate Masks for each Image, and the Solve them.
Now, my Questions:
1.) Would you rather solve all the Images at Once or split it all up into Chunks and align and merge them afterwards?I'am a bit afraid of the "Align-Chunks"-Process. Will the Result match the accuracy of a normal "Align-Photos"-Solve?
And will it help, when Photoscan has more different ViewingAngles during the AlignProcess, when solving all Angles in one big Chunk?
2.) In some cases the AllInOne-Photo-Alignment fails, and i'am currently not sure why.Might Spill-Light from the GreenScreen/Ground distract the Algorithms? In the attachment you can see a recently failed Solve
(Align Photos, High, 40 000 Points, Constrained by Masks, 1400 of 1600 Photos "aligned")
Hope someone has some great Tipps
Cheers
Vincent