Make Your Own Avatar

July 1, 2014 | Vancouver, Canada

Speaker: Ari Shapiro, Evan Suma

This near-automatic pipeline captures a human subject and, in just a few minutes, simulates the person in a virtual scene. The process can be fully managed by the capture subject, who operates a single Microsoft Kinect. No additional assistance is required. The speed and accessibility of this process fundamentally changes the economics of avatar capture and simulation in 3D. Because the avatar-capture cost is near zero, and the technology to perform this capture has been deployed in millions of households worldwide, this technology has the potential to significantly expand the use of realistic-looking avatars.

The short capture time allows frequen, even daily, avatar creation. The avatars are of sufficient resolution to be recognizable to those familiar with the human subject, and they are suitable for use at a medium distance (such as third-person perspectives) and in crowd scenes.

The pipeline consists of three stages: capture and 3D reconstruction, rigging and skinning, animation and simulation. The capture process requires the subject to remain steady for about 15 seconds at four angles with 90-degree offsets from each other. The automatic rigging and skinning uses voxel-based approaches that do not require watertight meshes, and are thus suitable for capture methods that use reconstruction from points. The simulation and animation performs online retargeting of a large variety of behaviors, ranging from locomotion to reaching and gazing.