3D model reconstruction from turntable sequence with multiple-view triangulation

Abstract

This paper presents a new algorithm for 3D shape recovery from an image sequence captured under circular motion. The algorithm recovers the 3D shape by reconstructing a set of 3D rim curves, where a 3D rim curve is defined by the two frontier points arising from two views. The idea consists of estimating the position of each point of the 3D rim curve by using three views. Specifically, two of these views are chosen close to each other in order to guarantee a good image point matching, while the third view is chosen far from these two views in order to compensate for the error introduced in the triangulation scheme by the short baseline of the two close views. Image point matching among all views is performed by a new method which suitably combines epipolar geometry and cross-correlation. The algorithm is illustrated through experiments with synthetic and real data, which show satisfactory and promising results. © 2009 Springer-Verlag.