When CGI is overlaid onto an existing piece of video, and the spatial geometry of the CGI matches that of the video - it's most noticeable when it's hand held. What software is used to determine where the video camera is in the 3D world? Never mind the actual generation of the CGI or the compositing, just how does one work CGI into the 3D world of the video? I can't believe it's done by eye frame by frame...
What software does this?
Miscellaneous Forums/General Discussion/What software does this? Matchmoving software, this. Back-issue 86 of Digit group tests a bunch - Boujou 3; RealViz Matchmover Pro; Science D-Visions 3D Equalizer; PFMatch & Boujou Bullet.
It's not done by eye, but it's not done by software either. It involves detailed measurements made during the live camera shoot. Then you duplicate all of the measurements in your 3D software.
Yep and you have to know your camera FOV distortion or whatever, you know how some lens actually morph the picture very slightly towards the edges. Less so if you zoom in, or on professional cameras.
I think all three of you could be right! I had a feeling you could do it with basic video as I watched a Child Of Our Time special yesterday, with clips from five years ago overlaid with big 3D year numbers that matched the movements of the camera.
Just looked up the Matchmoving software, looks amazing. And amazingly expensive.
Just looked up the Matchmoving software, looks amazing. And amazingly expensive.
Take the HP commercial for example, the guy which is singing and "taking pictures" by holding a white frame in front of his head. (It is shown here in the Netherlands...).
They use the high contrast of the frame to determin the perspective and size of the frame contents. Maya can do this (just like the software that has been mentioned).
At the moment the guy holds the frame in front of his head and moves it away, a "snapshot" of the guy is rendered inside the frame and it looks like it has become a photo.
A simple concept, but amazingly well done.
For movies like Lord of the Rings, "motion track cameras" (I dont know the exact name) are used. They "save" the position of the camera, and when you pan the camera in X,Y,Z, the coordinates are saved. When connected to equipment like a crane, the crane then later can playback the exact movement that has been recorded earlier.
Even better, the X,Y,Z data can be important by a 3D application (3dmax, Maya, ...) which exactly matches the viewport perspective, FOV etc.
In the end the 3D is placed on top of the movie as a layer. During modelling the movie is used as a background. I think the real hard thing is to match the light of the 3D models with the actual movie footage.
They use the high contrast of the frame to determin the perspective and size of the frame contents. Maya can do this (just like the software that has been mentioned).
At the moment the guy holds the frame in front of his head and moves it away, a "snapshot" of the guy is rendered inside the frame and it looks like it has become a photo.
A simple concept, but amazingly well done.
For movies like Lord of the Rings, "motion track cameras" (I dont know the exact name) are used. They "save" the position of the camera, and when you pan the camera in X,Y,Z, the coordinates are saved. When connected to equipment like a crane, the crane then later can playback the exact movement that has been recorded earlier.
Even better, the X,Y,Z data can be important by a 3D application (3dmax, Maya, ...) which exactly matches the viewport perspective, FOV etc.
In the end the 3D is placed on top of the movie as a layer. During modelling the movie is used as a background. I think the real hard thing is to match the light of the 3D models with the actual movie footage.
I do quite a bit of motion tracking for work as a VFX artist. measuring all the camera data manually is not as important as it used to be, but any extra measurements help out when it comes time to composite.
and for anyone that wants to try it out, here is an EXCELLENT tracker that is free, and integrates tightly into 3dsMax
http://www.digilab.uni-hannover.de/docs/manual.html
and for anyone that wants to try it out, here is an EXCELLENT tracker that is free, and integrates tightly into 3dsMax
http://www.digilab.uni-hannover.de/docs/manual.html
its done using software such as boujou or pftrack, you run the video through it and it calculates where it thinks objects are in the scene based on how the images move, it can then make a guess at where the camera would have to be, it's very clever but incredibly slow going.
Pongo, thanks for that link! Have downloaded it and the examples and am looking forward to having a good rummage...
"I think the real hard thing is to match the light of the 3D models with the actual movie footage. "
I think they use 2 spheres on the end of sticks during filming - one is mirrored and the other neutral grey. Don't ask me how this translates to lighting information, best to ask Google.
Here is a test I did a few years ago. This particular example was never finalized, so there is still some slipping on the feet. Most tracks seem to work 80-90% straight away, and that last 10% takes the real work.
http://www.andimages.com/proofs/pongo/gfx/dino.mpg
For lighting, I photograph a mirrored ball in the same environment. I usually take 12 exposures and create an HDRI image I can use to recreate the lighting at any level. I then unwrap the ball into a texture that can be placed on a skydome for accurate reflections and lighting using global illumination.
http://www.andimages.com/proofs/pongo/gfx/dino.mpg
For lighting, I photograph a mirrored ball in the same environment. I usually take 12 exposures and create an HDRI image I can use to recreate the lighting at any level. I then unwrap the ball into a texture that can be placed on a skydome for accurate reflections and lighting using global illumination.