MATCHMOVING POV-Ray-rendered objects to a video.
(An .avi movie file, compressed using xvid-- 'mpeg4 part 2' compression. 1280 X
720 resolution, about 15-seconds long. The file quality isn't the best; it had
to be heavily compressed in order to upload here. Let me know if there's a
problem viewing/downloading it.)
See my earlier post here for some additional info...
For this test, I first made an HD video with my Canon SX130 camera, simply
walking forward in my driveway. It's purposely shaky, in order to give my
matchmoving code a real workout. The camera's 1280X720 resolution set the rez of
the CG object renders in POV-Ray as well. For the CG model, I used my handy
"dog-like police car" (which I've been improving)-- rendered as .png images with
Output_Alpha=on, to get a nice transparent background.
The reflections in the police car are from a large no_image sphere, mapped with
an image of ground and sky. (The image doesn't really *match* the surroundings
in the scene; but it adds a lot of realism anyway.) It's interesting that the
pristine and sharp CG renders of the police car needed a bit of Photoshop
'post-blurring' to better match the quality of the video and blend in better.
The final animation is a composite of three image files: the background video,
the police car, and the matted tree on the left. (I hand-matted 154 frames for
that, to hide part of the police car behind the big tree.) The compositing
process is a simple-and-fast 1:1 pixel re-rendering (in POV-Ray) of the images,
as layered textures on a box.
The shadow under the police car is fake, just an alpha-channel image_map on a
horizontal box surface; it was rendered along with the police car model. But
that shadow really helps to 'sell' the overall effect-- and it also helps hide
some 'glitches' in the spatial-z matchmoving (which I did by eye, essentially.)
The result here isn't perfect, but I hope to improve the overall method.
Post a reply to this message
Download 'matchmove_test_1.avi.dat' (4061 KB)