Recent Research Projects in
Computer Animation
from the
GRAVE (GRaphics, Animation, and Visualization Environment)


Reconstructing 3D model of person from video

Description: Video sequences of a person moving are analyzed and used to construct the three-dimensional motion of a synthetic version of the person. We are currently looking at using a silhouette from a single-camera. The synthetic camera and person is used to synthesize a silhouette and this is compared to a subject silhouette extracted from live video. The pose space of the synthetic model is searched in order to find an optimum fit between the synthetic silhouette and the live-action silhouette. Determining good fitness functions and search strategies are the subject of the research.

These techniques have been applied in medical gait analysis, inserting synthetic characters into live action sequences, tracking people in a lobby environment, and surveillance.

Click here for more information.


Facial Animation

 image of person's head Description: The representation and display of realistic facial animation is being researched in order to support video teleconferencing applications in which limited bandwidth prevents the transmission of live video. Image processing techniques (at Texas Instruments) are used to identify the positions of key feature points of the face. These are transmitted to a remote site and a synthetic face is reconstructed using local computational resources in real-time. Our task is to determine how to best display the face in terms of both speed and realism.

Click here for more information.


Secondary Motion for Human Figure Animation

Description: Human motion will be analyzed to identify characteristics of secondary motion in order to generate more realistic looking human motion automatically without detailed physically based models.

Student: Meg Geroch

Support: None

Work performed at: ACCAD

Status: Ongoing


Human Figure Generation for Animation

Description: Development of a system to easily design and generate synthetic humans. The emphasis is on simple geometries and effective visuals for VRML applications.

Student: Matt Lewis

Support: None

Work performed at: ACCAD

Status: Initiated Summer, 1997

Click here for more information.


Automatic Generation of Muscles and Articulation Skeleton of Arbitrary Structures

Description: Automatic techniques are being developed which will generate a articulation skeleton for an arbitrary objects and supply a musculature for the character generated.

Student: Lawson Wade

Support: None

Work performed at: ACCAD

Status: Ongoing

Click here for more information.


Anatomical Representation of the Human Figure

imae of skin over arm imae of skin over arm Description: Research into a more anatomically based representatio of the human figure is being conducted in conjunction with ACCAD. Human anatomy is used to model the skeleton and its associated articulations in greater detail than ever before. Anatomically based muscles are incorporated into the model based on their importance in producing realistic form. These are used to define an implicit function and B-spline control points are found at an isosurface. A B-spline surface can then be fitted to the form implied by the skeleton and musculature.

Student: Ferdi Scheepers

Work performed at: ACCAD

Status: Ph.D. in '96

For more information:


Implicit Surface

 image of person's head Description: Low level techniques have been investigated to speed up the display of distance-based primitives used to generate implicit surfaces. In previous work, the ray tracing of cone-spheres, spherically-capped cylinders, and rounded polygons was sped up by incorporating the same type of coherence considerations used by Blinn in his '82 TOG article, often credited for introducing 'metaballs' to the graphics community.

Student: Karanasher Singh

Papers:

more images

Status: Ph.D. Dissertation completed December, 1995


 image of person's head In more recent work, an adaptive octree polygonalizer has been developed which takes advantage of the primitives used to generate the implicit surface in controlling the amount of subdivision. This is combined with more complex intra-cell surface elements.

Recent work performed at: AutoDesSys, Inc.

Status: ongoing


Implicit Surfaces: Virtual Space Teleconferencing

Description: Implicit surface primitives embed parts of the human figure in order to make collision detection and response more efficient. The implicit surface primitives deform under collision with each other which in turn deform the embedded polyhedral object.

Implicit surface definitions are also used to form the fillet between two independently digitized parts of the human figure, such as the upper and lower arm. Polygonal sections outside the area of overlap are retained in their original definition while implicit surfaces are computed in the area of overlap.

Student: Karan Singh

Work performed at: ATR Labs, Japan

Status: Ph.D. Dissertation completed December, 1995

Papers:


Shape Change

 images showing H turning into chess piece Description: 3D morphing is a difficult problem. One general approach to the problem is to preprocess the origianl object defintions so that their edge-vertex topologoy match. The objects can then be blended on a vertex-by-vertex basis. One approach used here employs a recursive division of the surfaces, adding vertices and splitting faces as it progresses to maintain similar topologies. Another approach used was to map each object onto the surface of a sphere and then intersect edges. The resulting new vertices and edge segments were then added to the original object definitions to produce objects with matching topologies.

Papers:


Computed Human-Like Motion

 frame of animation Description: A genetic algorithm was used to search an obstacle-filled environment for possible paths to one or more goals (e.g. tip of end effector at a certain location and wrist at a certain location). Several optimization functions were tested as possible ones to use to produce the most human-like motion: shortest end-effector path, smallest maximum torque, etc. The animation shown here demonstrates the path found in order to satisfy positional goals (represented by spheres) for the tip of the limb and for the last joint of the limb.

Student: Dave Miller

Status: Dissertation completed 1994

Paper: Dave Miller and Richard Parent, ``An Articulated Limb Motion Planner for Optimized Movement,'' The Journal of Visualization and Animation, John Wiley & Sons, New York, Vol. 5, No. 2, pp. 89-123, April-June 1994. (also OSU CIS Technical Report, OSU-CISRC-2/93-TR8, February 1993.)