One-Class Learning for Human-Robot Interaction

Abstract

A Suitable learning and classification mechanism is a crucial premise for Human-Robot Interaction. To this purpose, several one-class classification methods have been investigated using wavelet features (parameters of Hidden Markov Tree model) in this paper. Only target class patterns are used to train class models. Good discrimination over outlier (never seen non-target) patterns is still kept based on their distances to class model. Face and non-face classification is used as an example and some promising results are reported.