Incremental PCA: an alternative approach for novelty detection

Abstract

Exploration and inspection of dynamic environments using mobile robots are applications that benefit immensely from novelty detection algorithms. In this paper we propose the use of a new approach for on-line novelty detection based on incremental Principal Component Analysis and compare its performance and functionality with a previously studied technique based on a GWR neural network. We have conducted a series of experiments using visual input from a mobile robot interacting with a controlled laboratory environment that show advantages and disadvantages for each method.