Classification and Novelty Detection of Omni-view Images Taken from a Mobile Robot

Abstract

We propose to use novelty as one of intrinsic motivations for learning in developmental robotics. Our approach classifies omni-view images taken from a mobile robot, finds outliers, and detects novelty. We use linear discriminant analysis for classification due to its optimality within linear computation. Experimental results demonstrate that although there are many misclassifications, there’s a possibility of making a new class composed of novel data.