3D Reconstruction Of Curved Surfaces from Single View

Abstract

This project work aims to find 3D model of curved object with surfaces having circular cross section from a single 2D image with minimum human intervention. Problem of generating 3D model is of great interest in many fields like medical science, painting modeling, virtual reality and heritage documentation. Single View Reconstruction appears to be shifting away from Multi view Reconstruction where only single image is available. Though a single view of a scene does not have enough information about the depth, the problem of 3D Reconstruction can be solved to some extent by imposing constraints or through user interaction. 2D images contain cues about real object. However, their interpretation is ambiguous because depth information is lost when 3D object is projected on 2D image. Multiple images from different viewpoints resolve these ambiguities. However, sometimes it is not possible to obtain such images, and only single photographs or paintings exist or even when working with multiple images, parts of the scene appear in only one image due to occlusions. Problem of generating 3D surface can be expressed as optimization problem. From Single view Silhouette is extracted. It generates best 3D surface which projects exactly to that silhouette.