3D object reconstruction from images

Abstract

Computers nowadays are still limited to interact with the world and human beings due to the fact that they lack the ability to "see". In this dissertation we will investigate the methodologies incurred in digitizing real life objects and building 3D models. The three dimensional object acquisition can be performed by range sensors, such as laser scanners. However, such devices are quite expensive and consequently are not so popular. The object acquisition by means of a range of image sequences from a monoscopic camera is a low cost method that makes it a much more attractive alternative. The basic volume model will be built by means of the images' silhouettes, however, these will only aid in generating a point cloud representing the object's surface. This will reduce considerably the amount of vertices that must be processed. In order to produce the point cloud, a new intersection algorithm will be devised that involves intersecting the point cloud with each view so that vertices not belonging to the object's surface are removed. The object reconstruction process is concluded by using a triangulation algorithm in order to generate the object's surface. The system will also support plug-ins that will export the object to any desired format increasing significantly the usefulness of the application.