Output details

A generative framework for fast urban labeling using spatial and temporal context

First published in the proceedings of Robotics-Science-Systems (the premier international robotics conference) in 2008 and was consequently invited for journal publication in Autonomous Robots. This paper is amongst the first in robotics combining vision and 3D laser data in a relational learning framework for large-scale-outdoor semantic mapping. The contribution here is the classification of the surroundings of the robot taking into account contextual prior knowledge, and in the unusual way it frames the problem probabilistically using machinery which became the underpinning of the FAB-MAP algorithm a best-in-class scene interpretation algorithm which was licensed to Google (http://www.ox.ac.uk/media/science_blog/100412.html).