ACTIVE OBJECT RECOGNITION INTEGRATING ATTENTION AND VIEWPOINT CONTROL

We present an active object recognition strategy which combines the us e of an attention mechanism for focusing the search for a 3D object in a 2D image, with a viewpoint control strategy for disambiguating reco vered object features. The attention mechanism consists of a probabili stic search through a hierarchy of predicted feature observations, tak ing objects into a set of regions classified according to the shapes o f their bounding contours. We motivate the use of image regions as a f ocus-feature and compare their uncertainty in inferring objects with t he uncertainty of more commonly used features such as lines or corners . If the features recovered during the attention phase do not provide a unique mapping to the 3D object being searched, the probabilistic fe ature hierarchy can be used to guide the camera to a new viewpoint fro m where the object can be disambiguated. The power of the underlying r epresentation is its ability to unify these object recognition behavio rs within a single framework. We present the approach in detail and ev aluate its performance in the context of a project providing robotic a ids for the disabled. (C) 1997 Academic Press.