Recognition and Localization Techniques

Abstract

Having obtained a feature-level description of the scene, the next step is matching these features with those computed from the internal representation of the object models. The purpose of the matching process is two-fold—first, to label the features in the scene as an instance of an object model and, second, to determine the position and orientation of the object in the scene with respect to a reference coordinate frame. The process of labeling scene features as having arisen from a particular object model is referred to as recognition, whereas determining the position and orientation of an object in the scene is referred to as localization. In fact, most matching techniques treat recognition and localization as two inseparable issues of the same overall problem, to the extent that the term recognition is often used to denote both recognition and localization.