Building 3D object representation using SURF local features

Abstract

The paper discusses an approach to create 3D representation of physical objects. The aim is creating a visual representation of an object, which allows for robust recognition, irrespectively of the distance and the direction of observation. The approach uses a set of rotational views of an object, which are transformed into a set of keypoints using the SURF visual feature detector. The key points are then collected to build a 3D model of the object. Such representation allows both for recognizing the objects based on local characteristics, and distinguishing different global geometry transformations that are needed to recognize the object in its 3D environment.