A multi-view 3D shape reconstruction system using level sets

Abstract

The 3D reconstruction of a scene from multiple images is a fundamental problem in the field of computer vision. Existing methods can be classified into two strategies: bottom-up or top-down. This paper presents a full system for complete 3D shape reconstruction following the top-down strategy. A rotary table is employed to change a camera’s viewing direction to an object on the table. This offers a cost-effective solution to the multi-view stereo acquisition problem without the need for using several cameras. From the acquired calibrated images of the object, a variational approach is developed for 3D shape reconstruction of the object. The approach works directly in 3D Euclidean space based on a level set formulation. A correlation criterion between the 2D images is optimized by driving the evolution of the surface using the corresponding Euler-Lagrange equation. Several successful experiments to evaluate the proposed system are reported.