Hybrid rigid and non-rigid image-based modeling of articulated figures

Aaron Hertzmann*$\dag $
Trevor Darrell*

(*) Interval Research Corporation
($\dag $) Media Research Lab, Dept. Computer Science, New York University,


We describe a new method for rendering new poses of articulatd figures. The long-term goal of this work is to take images or video of a person in different poses, and then render images of that person in new poses. The human body is modeled as a rigid kinematic chain. A morphing technique models non-rigid changes in shape and texture of each kinematic link. This system achieves much more generality than either pure rigid or pure non-rigid models could achieve alone.
We demonstrate results on planar motion of a single arm. From only two images of the arm, the algorithm can generate a wide range of believable poses. We believe that the new approach is powerful enough to model 3D motion of a complex, articulated figure.

Interval Research Corporation Technical Report 1998-045.

This report is available in HTML at or in PostScript form at

Result images and sequences are on this page.

©1998 Interval Research Corp.

Trevor Darrell