The human bodies papers covered a variety of topics, from reconstructing heads from skull data, to building character skins from metrics, to capturing skin deformation using video cameras.
Reanimating the Dead: Reconstruction of Expressive Faces from Skull Data
If you've ever seen a modern forensic TV show, they are pretty much guaranteed to show you somebody building a face by putting little spacers and clay on a human skull.
That process may be sped up considerably if the researchers who wrote this paper have their way. Although the technique is far from refined, these folks are creating a method to create expressive faces from skull data in minutes or hours instead of the weeks it usually takes when doing it by hand.
The process involves scanning the skull (which can be done through a variety of methods, including CT scans and range scanning), marking landmark points on the skull with a mouse and telling the computer what those points correspond to. The computer then takes over and uses data from a well known database to add height to the markers and then creates the face by warping an existing model head onto the landscape points.
This method has some significant advantages over the existing manual method, including:
- Quicker initial construction (minutes/hours instead of days/weeks)
- The ability to generate multiple heads from the same model data (using the physical method, you need to start from scratch to make a second model)
- The resulting face is expressive, meaning you can change the expression to match the requirements.
It's clear that these folks know their stuff and there are some good applications for this project moving forward.
Building Efficient, Accurate Character Skins from Examples
The researchers from the University of Wisconsin who gave this talk are working on a method to create fast interactive deformable skins. In order to keep everything fast, one of their design constraints was a requirement to use only linear blend skinning (because this is handled in hardware by most modern graphics card). Unfortunately, existing LBS methods have weaknesses that often make them collapse or self-intersect if they aren't carefully tweaked.
This paper details a method for automatically creating additional joints in the model to keep the skinning from creating artifacts.
One proposed use of this method is to allow animators to use more complex skinning algorithms in modeling programs but output the data using a system that can use this method to create a more "boiled down" version of the geometry that will still deform correctly.
This technique shows promise for the gaming industry especially.
Free-viewpoint Video of Human Actors
In the first of three papers that discussed taking human forms from captured data, this group of researchers proposed using video shot from eight synchronized cameras to capture data from free motion. Thus, instead of requiring a fixed set of actions to be motion captured, it would capture a full range of motions and their effect on the skin and store this for use later. When the model is then animated, the data is used to determine the level and type of deformation. Although promising, it is still a bit heavy, requiring 5 stat-of-the-art CPUs each with a GPU to handle scanning one frame of reference video per second.
Continuous Capture of Skin Deformation
This paper dealt with the automatic acquisition of human skin geometry, similar to the previous paper. Also similar to the previous paper, this paper does so with requiring a specifically choreographed set of moves. However, this is where the similarities stop. The method presented here by the folks from MIT used a combination of video and MoCap data to create an animatable 3D geometry. The system can scale down to a single camera for acquisition, although most examples were done with three operating on the same subject. The algorithm does a beautiful job of capturing geometry and has some other interesting attributes besides as regard being able to describe the human form by a set of parameters.
Space of Human Body Shapes: Reconstruction & Parameterization from range scans.
This method was unique from the other two because it does not involve taking new measurements. Instead, the system relies on the large human-form data sets now available commercially to serve as the basis for a global model of what man and woman should like like at particular ages.
Using this data, they have been able to come up with highly realistic deformable models that provide a high level of accuracy while at the same time providing extreme control over the models. As an example, they showed sliders for height, and a number of width (weight) measurements.