Thursday, 18 February 2010
The Great Garage Experiment: Part 2
Does anyone remember Part 1 of the Great Garage Experiment from a few weeks ago? No we hadn't forgotten about this gripping VFX project, but the day-job work has been hampering our efforts somewhat, so it has taken Rich slightly longer than usual to complete his experiment.
This time the objective was to matchmove the 3D model of the garage (which we prepared last time, if you recall) into the real world - and to do it convincingly. Alas we never did manage to get any decent footage of a real road (weather and moaning kids did not permit) so...erm...we used footage of our very own table tennis table. Yeah, yeah, I know you don't see too many real garages located on ping-pong tables, but the proof of concept is the same - combining real world and CG so that you can't tell the difference.
The process for aligning the garage on the table was to photograph an HDRI globe (big shiny silver ball – purchased from local garden centre - no expense spared) so that we could capture the environmental lighting of the table and then use image-based lighting for the render. This was so that the lighting and colouring of the render matched the table. The next stage was to put some tracking markers (in this case ping-pong balls) and shoot the video that we wanted to use with a camera on a dolly.
Next we loaded the footage from the camera into SynthEyes and we tracked the ping pong balls so that we obtained a virtual camera which matched the real one. The output of Syntheyes was then loaded into Softimage so that we had a virtual camera in Softimage that matched movement of the camera in the real world. We then loaded in the garage model and positioned it appropriately on the virtual table. We created two render passes, one image based lighting (IBL) and one ambient occlusion (AO). This completed the rendering portion of the experiment.
We then moved to Adobe After Effects CS4 for the compositing stage and we loaded the original camera footage and our two render passes. This gave us three layers. At the bottom we had the original camera footage, overlaid on that was the IBL pass which contained the colour and shadows, and finally on top of all of that was the AO which contained the overall brightness and darkness of parts of the model. After adjusting the various levels of each layer and adding a noise overlay, it was ready for final output:
Personally, I think the result is pretty darn realistic, don't you? Now what would be really cool is if the little toy car on the left there (which is real) would drive itself up into the garage and park itself. The resident visual effects artiste tells me that he is working on it. Stay tuned...
This time the objective was to matchmove the 3D model of the garage (which we prepared last time, if you recall) into the real world - and to do it convincingly. Alas we never did manage to get any decent footage of a real road (weather and moaning kids did not permit) so...erm...we used footage of our very own table tennis table. Yeah, yeah, I know you don't see too many real garages located on ping-pong tables, but the proof of concept is the same - combining real world and CG so that you can't tell the difference.
The process for aligning the garage on the table was to photograph an HDRI globe (big shiny silver ball – purchased from local garden centre - no expense spared) so that we could capture the environmental lighting of the table and then use image-based lighting for the render. This was so that the lighting and colouring of the render matched the table. The next stage was to put some tracking markers (in this case ping-pong balls) and shoot the video that we wanted to use with a camera on a dolly.
Next we loaded the footage from the camera into SynthEyes and we tracked the ping pong balls so that we obtained a virtual camera which matched the real one. The output of Syntheyes was then loaded into Softimage so that we had a virtual camera in Softimage that matched movement of the camera in the real world. We then loaded in the garage model and positioned it appropriately on the virtual table. We created two render passes, one image based lighting (IBL) and one ambient occlusion (AO). This completed the rendering portion of the experiment.
We then moved to Adobe After Effects CS4 for the compositing stage and we loaded the original camera footage and our two render passes. This gave us three layers. At the bottom we had the original camera footage, overlaid on that was the IBL pass which contained the colour and shadows, and finally on top of all of that was the AO which contained the overall brightness and darkness of parts of the model. After adjusting the various levels of each layer and adding a noise overlay, it was ready for final output:
Personally, I think the result is pretty darn realistic, don't you? Now what would be really cool is if the little toy car on the left there (which is real) would drive itself up into the garage and park itself. The resident visual effects artiste tells me that he is working on it. Stay tuned...
Labels: matchmoving, visual effects
Friday, 22 January 2010
The Great Garage Experiment: Part 1
Resources: Autodesk ImageModeler, Softimage
The purpose of this exercise was to create a photorealistic 3D model of a real world object. Any object would suffice – in this case a toy garage provided a worthy subject for our modelling experiment.
The first step was to take a series of still photographic reference shots from lots of different angles so as to give full coverage of the object.
The purpose of this exercise was to create a photorealistic 3D model of a real world object. Any object would suffice – in this case a toy garage provided a worthy subject for our modelling experiment.
The first step was to take a series of still photographic reference shots from lots of different angles so as to give full coverage of the object.
These images were then loaded into Autodesk’s ImageModeler and reference points were created in each photograph. Each reference point represented a feature of the object, for example the corner of a window or wall. The more points you put in, the better. If you tell ImageModeler enough of the points in enough of the photographs then the software can determine the shape of the object and how it is aligned in each picture, and consequently it has enough data to build a model of the object. ImageModeler then performs some really clever calculations in order to extract the textures for the objects.
When this process was complete we exported the object and all of the textures into the Softimage modelling software. We then further refined the texture maps and edited the textures in Photoshop. The result was a model that exactly matched the object in the photograph. Finally the image was rendered in Softimage in order to produce the final output.
When this process was complete we exported the object and all of the textures into the Softimage modelling software. We then further refined the texture maps and edited the textures in Photoshop. The result was a model that exactly matched the object in the photograph. Finally the image was rendered in Softimage in order to produce the final output.
The second part of the experiment will involve putting the object into the real world, in our case, match-moving the garage onto a real road. Stay tuned for the results in a couple of weeks time....
Labels: matchmoving
Subscribe to Posts [Atom]
Freekstorm on Facebook
Links
- DVXUser
- Effects Corner
- Visual FX Blog
- Lux Visual Effects Blog
- VFX Talk
- Visual Effects Society
- CG Society
- The Smalls
- Indy Mogul
- Filmmaker Blog
- All About Indie Filmmaking
- Living the Romantic Comedy
- Truly Free Film
Archives
- December 2008
- January 2009
- April 2009
- May 2009
- June 2009
- July 2009
- August 2009
- September 2009
- October 2009
- November 2009
- December 2009
- January 2010
- February 2010
- March 2010
RSS Subscribe