Publications / 2004 Proceedings of the 21st ISARC, Jeju, South Korea

3D Vision Techniques and Applications to Photorealistic Scene Reconstruction

Jun-Sik Kim, Ho-Won Kim, In-So Kweon
Abstract:

We present some novel concepts for the scene geometry recovery by a multi-camera system. A new global appearance measure is introduced for a novel generalized scene recovery methodology, called “the appearance-cloning.”

For the efficient calibration of the multi-camera system, we investigate the projective properties of the concentric circle pattern. Specifically, two algebraic constraints from concentric circles are defined to calibrate the intrinsic parameters of the multi-camera system. Through a variety of scene reconstruction experiments, we demonstrate that the use of concentric circles greatly simplifies the calibration problem for the photo-realistic recovery of dynamic 3D scene.

Keywords: photorealistic 3D reconstruction, volume based approach, camera calibration