A variational technique for 3D reconstruction from multiple views

The possibility of being able to acquire 3D information from 2D images has been attracting more and more interest in recent years. Many researchers have therefore been trying to develop robust as well as efficient methods to reconstruct 3D models from images. In this paper we develop and implement a variational approach for surface reconstruction starting from multiple 2D images taken by a calibrated camera. The approach works directly in 3D Euclidean space based on a level set formulation. The correlation between the 2D images reprojected onto an evolving surface is optimized by driving the motion of the surface using the Euler-Lagrange equation of the correlation criterion. Several successful experiments to evaluate the approach are reported.