Nonrigid Stereo Reconstruction Using Linear Programming

摘要

Nonrigid reconstruction plays an important role in many applications like image-based modeling, human-computer interaction. In this paper, we propose an effective approach for 3D non-rigid structure reconstruction using linear programming from an image pair taken with a stereo rig. In contrast to previous approaches, the proposed method neither involves smoothness constraints nor need prior knowledge between consecutive frames, which enables us to recover shapes of surfaces with smooth, sharp and other complex deformations from a single image pair. Specifically, we model the surface as a triangulated mesh and formulate the reconstruction problem as a Linear Programming (LP) problem using L_∞ that can be effectively solved. The LP problem consists of data constraints which are 3D-to-2D keypoint correspondences and shape constraints which preserve original lengths of mesh edges. The robustness and accuracy of our approach are evaluated quantitatively on synthetic data and on real data.