Classifier Independent Viewpoint Selection for 3-D Object Recognition

Abstract

3-D object recognition has been tackled by passive approaches in the past. This means that based on one image a decision for a certain class and pose must be made or the image must be rejected. This neglects the fact that some other views might exist, which allow for a more reliable classification. This situation especially arises if certain views of or between objects are ambiguous. In this paper we present a classifier independent approach to solve the problem of choosing optimals views (viewpoint selection) for 3-D object recognition. We formally define the selection of additional views as an optimization problem and we show how to use reinforcement learning for continuous viewpoint training and selection without user interaction. The main focus lies on the automatic configuration of the system, the classifier independent approach and the continuous representation of the 3-D space. The experimental results show that this approach is well suited to distinguish and recognize similar looking objects in 3-D by taking a minimum amount of views.