Model Deformation Stepping Into Volumetric

[Xittenn]

I'm working on a next generation graphics engine at the moment. I've been thinking on this for many years and one of the most recurring thoughts with regards is on how to produce a more realistic deformation. To better elaborate on what I mean by Volumetric Deformation I link this interesting paper.

 

Current models are deformed through the use of bones and effectors which vertices are bound and weight anchored to. The system is designed where in if a bone is to move a vertice will follow at a rate congruent to the value of the weight associated.

 

There are a lot of problems with the current system where mesh structure tends to deform in a less than desirable fashion. Often the mesh is stretched out in such a way that the structure collapses in on itself where desired bulk would be preferred. There are many examples of this issue in the linked paper. This tends to happen around elbows and other joints or the abdomen in extreme bending situations.

 

There are various ways of correcting these problems. Newer techniques involve making morph targets which supplement the movement of the animation and compensate for otherwise improper deformation. I'm actually quite fond of this method as producing volumetric deformation in a constant fashion can be quite taxing on clock cycles. I am still however looking towards the future and volumetric deformation could reduce the amount of time required to make more realistic and pleasant animation. It could also increase the flexibility of real time interaction with the virtual world!

 

So I'm presently looking at different methods of creating a new animation type. The paper linked is a modification of the present system with essentially a morph target built in which could be represented mathematically and compensate in real time. I however am looking at filling the space between the skin and the bones with volumes, call them muscles, which will create shape and can have attributes of their own. This is something that would require a lot of processing power, in the order of a NVIDIA Tesla.

 

Regardless would love to hear some thoughts from people on how they see this happening.....