Hello people!
Just thought to share this. Maybe someone has already tried it, maybe not. At least I haven’t seen it done this way before.
All this because I don’t have a scanner and I’m not a Super Modeler Individual From Outer Space 🙂
No, I did not track hundreds of markers on the head and try to clean up a resulting mesh etc. Yes, ofcourse, I have tried that too, but it is too time consuming and complicated. My approach is different and more straight forward.
First of all, let’s make it clear we have different shots for different purposes. So I’m not talking about a shot in some scene. I’m talking about a short, separate clip of video taken from our talent for the purpose of modeling his head.
To model the face as accurately as possible, I make my talent turn his head side to side so that I have him facing the camera in the beginning and then he turns side to side. This is because I want a 180 degree, or so, rotation of the head.
Ok, You can also make him turn from one side to another and make sure you know the exact frame in the middle where the face is absolutely pointing straight forward. Whatever best serves your needs.
I did not have any trackmarkers on the face but You may want to have them because it can take a while longer without them. It depends on whether or not you need the face clean or not and about the level of visible detail on the skin. RED gives you face texture material as you work on. If you have the format available.
Anyway, all this stuff I do to achieve the following situation:
I want to model the head, so I want my geometry to be in the origin of my 3d scene and I need a rotating camera running around it projecting the footage back onto it.
So I track the head but use the 3d track as if I was solving for a moving camera. This means I get a stationary 3d world with markers and a moving camera.
Then I go to my 3d package (in my case 3dsmax) and bring in my moving camera.
Prior to this I had modeled a simple piece of geometry, only including the most important features and edgeloops of the face area only and I place it to one side of the face and roughly make it fit the areas of the real face as I want. Then I make the face geometry symmetrical sideways so that I get the other side as well. I don’t mind about asymmetry too much at this point.
Then I use camera mapping to project the tracked footage on the face geometry (using Camera Map Animate plugin (free)). When I hit visible edges, I can see myy fotage on the rough face geometry along with my edges.
Now I have a setup where I have a rotating camera projecting the tracked video footage on the geometry and I can scrub back and forth through mt timeline and it is showing me exactly if my edgeloops are sliding or not as I work on to refine the geometry step at a time.
When the face works, then go for the rest of the head.
It is up to you what framecounts, resolutions and formats you wish to use. I used 1024 as width and jpg sequence which makes it relatively light to scrub with, yet reasonable accurate in the viewport (and remember to customize how your 3d package shows your texture detail to get enough detail).
Ok.
So, seeing the edges along with the projected footage of the turning head made my life really much easier because I could model the head and scrub through the timeline as if I asked "my talent to turn his head" to some other position I needed.
And, most importantly, I got reasonable results in a much more acceptable time than I thought before trying this out. Thenagain, maybe it’s just me.
