Goal:
Capture of realistic whole-body animated models of clothed people.
Background:
Previous whole-body capture systems have focused on accurate measurement of surface shape. Active 3D
sensing technologies such as laser-scanning are expensive due to the requirement for rapid
acquisition without movement. Such systems result in high-accuracy shape with low-quality colour
resulting in accurate but unrealistic human models.
New Approach:
Model-based reconstruction from multi-view colour images. A set of four colour images of a person
are captured from the front, back and sides. The images are used to morph a 3D generic humanoid
model to have the shape and appearance of a specific person. Models are generated in a standard VRML Humanoid Animation format which can be animated
in any VRML browser using pre-defined animations.
Breakthroughs:
Low-cost whole-body capture of clothed people from digital colour images.
Automatic reconstruction of animated models.
Collaboration:
This technology was pioneered by the 3D Vision group at the
Univesity of Surrey. Industrial collaboration under the EU i3 POPULATE project lead by 3D Scanners and REM
Infografica has lead to the commercial development of this technology. An AvatarBooth based on
this technology has been developed by AvatarMe and launched
at SIGGRAPH'99.
The process of capturing models from images is shown in this overview page.
This is a gallery of results obtained from the process. The results show the original images, a number of static views of the resulting model, and links to the resulting VRML file and a short video for those of you without VRML browsers. Click on a picture to see the results for that dataset.
 Dan |
 James |
 Andy |
 Tazos |
 All four of the above avatars on a virtual catwalk VRML97 - Scene total ~2Mb |
|||
© Copyright 1999 University of Surrey - A.Hilton@ee.surrey.ac.uk
Last modified 6/8/99