<title>Active object recognition using appearance-based representations derived from solid geometric models</title>

Abstract

Abstract We present new test results for our active object recognition algorithms. The algorithms are used to classifyand estimate the pose of objects in different stable rest positions and automatically re-position the camera if theclass or pose of an object is ambiguous in a given image. Multiple object views are now used in determiningboth the final object class and pose estimate; previously, multiple views were used for classification only. Afeature space trajectory (FST) in eigenspace is used to represent 3-D distorted views of an object. FSTs areconstructed using images rendered from solid models. We discuss lighting and material settings for photorealismin the rendering process. The FSTs are analyzed to determine the camera positions that best resolve ambiguities.Real objects are recognized from intensity images using the FST representation derived from rendered imagery. Key Words: active object recognition, active vision, Bayesian estimation, pose estimation 1 Introduction Active computer vision systems change sensor parameters such as position, orientation, zoom, focus, aperture, andvergence in response to visual stimuli or task parameters [1-3] . In our work, we assume that we have the abilityto move the camera relative to the object and take additional images to reduce ambiguity in scene interpretation.This involves estimation of a rigid object's class and pose from one view and using this information, and an objectrepresentation, to determine where to look next.