Human Intention Recognition in Collaborative Environments using RGB-D Camera

Jelena Maric, Luka Petrovic, Ivan Markovic

Abstract: The increasing presence of robots and other autonomous systems in everyday environments, such as households, workplaces, schools and hospitals, requires their safe cohabitation and collaboration with humans. Accurate and reliable human intention recognition that fosters safe and efficient human-robot collaboration is thus a crucial component of these autonomous systems, especially if attainable from relatively cheap and widely available sensors such as RGB-D cameras. In this paper, we propose a hand-crafted model of human intention when reaching for one of multiple objects present on the table in front of a person. By coupling our hand-crafted model with a method for human skeleton tracking from RGB-D images, we devise a computationally efficient human intention recognition method suitable for collaborative pick-and-place scenarios. We experimentally verify our method in real-world scenarios of a person reaching for one of two, three and four objects placed on a table. We complement the paper with an open-source implementation of our human intention recognition method.