Identificação de vértices em imagens multivista de formas 3D.

Abstract

The objective of this dissertation is the development of a method for identification of 3D vertices for multi-view images of scene composed by one or more objects. This identification allows the definition of polyhedral object models, which can be used for 3D recognition. The developed algorithms operate by data-flow, thus enabling the whole modelling process to be perfomed in real-time. The method works with three real images of a scene, taken from three different points of view, also known as multi-view images: top, side and front views. The method recognizes not only the external contour of the object, but also its internal edges.