3D model building

Abstract

One of the key issues in the field of 3D object recognition is the provision of an automatic and complete modeling system for 3D objects. Since a single range finder can only provide the 2 1/2 D range image from a single viewpoint, the complete 3D description of the object is not attainable directly. However, several 2 1/2 D range maps can be extracted from different viewpoints and integrated together to form the complete 3D model. This is an important aspect of object recognition since recognition using complete 3D models is computationally more efficient than using partial models. The ability to build 3D models efficiently is useful in a variety of applications such as the establishment of a large library of manufacturing parts, or in automated reverse engineering whereby original parts are scanned in order to make redesigns in a CAD system. The focus of this paper is to illustrate an application of using point signatures proposed by Chua and Jarvis (see Int'l J. Computer Vision, vol.25, no.1, p.63-85, 1997 and vol.17, no.1, p.77-99, 1996). In addition to its application in recognizing partially represented objects, it can be used to integrate multiple overlapping range images into an object-centered 3D representation, without the prior knowledge of the viewpoints from which the range images are extracted. The novelty of the approach is that the interframe relationships between different range images need not be known a priori. This paper describes an aspect of model building by integrating different range views autonomously.