Abstract
This paper introduces an extremely computationally inexpensive method for estimating monocular, feature-based, heading-only visual odometry - a visual compass. The method is shown to reduce the odometric uncertainty of an uncalibrated humanoid robot by 73%, while remaining robust to the presence of independently moving objects. High efficiency is achieved by exploiting the planar motion assumption in both the feature extraction process and in the pose estimation problem. On the relatively low powered Intel Atom processor this visual compass takes only 6.5ms per camera frame and was used effectively to assist localisation in the UNSW Standard Platform League entry in RoboCup 2012. © 2014 Springer-Verlag Berlin Heidelberg.
Cite
CITATION STYLE
Anderson, P., & Hengst, B. (2014). Fast monocular visual compass for a computationally limited robot. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 8371 LNAI, pp. 244–255). Springer Verlag. https://doi.org/10.1007/978-3-662-44468-9_22
Register to see more suggestions
Mendeley helps you to discover research relevant for your work.
