Modeling human geometry is useful in applications from gaming to anthropometric studies for industrial design, fashion, ergonomics, and architecture. Typically, developers of 3-D models of humans have relied on expensive and complex laser scanning or camera systems. Researchers at the University of Alberta (Edmonton, AB, Canada; www.cs.ualberta.ca/~vis/) have developed a method that may make standard turntable-based vision acquisition a simpler, more practical method for recovering realistic models of human geometry (watch the video below).
The main problem in using turntables for human shape acquisition is that the subject often wobbles when rotating on a turntable, resulting in loss of geometry especially on limbs. Neil Birkbeck and his colleagues in the computer science department have taken a vision-based approach using two IEEE-1394b color Grasshopper cameras from Point Grey Research (Richmond, BC, Canada; www.ptgrey.com). Their goal is to acquire static human models in a quick setup with minimum hardware.
The researchers combine silhouette cues with an initial skeletal model to recover the joint motion or wobbling. The recovered motion gives the calibration of each rigid body of the subject, allowing the fusion of image cues such as silhouettes and texture to refine the geometry. Using their two-camera system, the tracked joint coordinates are used to reconstruct the geometry of each body part; these parts are combined to obtain a new geometry. The tracking and refinement are then iterated several times.
Reprinted with permission from the Vision Systems Design article, "Imaging system models rotating humans ", featured in the October 2009 issue.