For robots to function successfully in human environments, fast and accurate recognition and verification of 
objects are essential. Often a single view of an object is not sufficient to accurately classify objects 
especially if they are occluded or appear in cluttered scenes. Actively exploring objects by changing viewpoints 
can increase the accuracy of classifying an object.

This talk will presents an efficient feature based active vision system for the verification of objects that appear 
in cluttered scenes or are occluded. This system is designed using an selector-observer framework where the selector 
represents the automatic view selection. A new method for automatically selecting the `next best viewpoint' using 
vocabulary trees is presented and a Bayesian `observer' updates the belief hypothesis for object verification and 
provides feedback. The vocabulary tree also speeds up the matching process. New images are only captured at the 'next best viewpoint'
when the belief hypothesis of an object is below some pre-defined threshold. This system is shown to be more accurate 
than randomly selecting the next viewpoint. It is also faster because only necessary views are captured and processed.