3 D Shape From Multi-viewpoint Images Using Deformable Mesh Model

This paper presents a two-stage method for 3D shape reconstruction from multi-viewpoint images. In the first stage, a rough 3D shape of an object is estimated by using volumetric intersection from geometric information of the object, i.e., silhouettes. In the following stage, the estimated shape is refined by using “deformable mesh model”. This model changes its shape so as to satisfy the following constraints: 1) “photo-consistency” constraint, 2) silhouette constraint, 3) smoothness constraint. The volumetric intersection estimates rough but stable shape using geometric information, and the deformable model refines the shape using photometric information. Hence, the method has two advantages of stability and accuracy. Some experimental results are also shown for demonstrating these advantages.