Structure learning for activity recognition in robot assisted intelligent environments

This paper presents a novel structure learning algorithm for the creation of distributed Bayesian networks over static and mobile Vision Sensor Network (VSN) nodes. These compose an assistive, intelligent environment for activity recognition. We provide results demonstrating a higher level of accuracy in the recognition of fine motor tasks when the environment is augmented with a mobile robot and show the ability of our learning algorithm to reduce VSN communication compared to a naive, greedy structure learning technique.