Publikation

Visual pose estimation for autonomous inspection of fish pens

Alexander Duda, Jakob Schwendner, Annette Stahl, Per Rundtop

In: MTS/IEEE Conference 2015. OCEANS MTS/IEEE Conference (OCEANS) May 18-21 Genova Seiten 1-6 ISBN 10.1109/OCEANS-Genova.2015.7271392 IEEE 5/2015.

Abstrakt

One of the main issues for using autonomous underwater vehicles in fish pen inspections is the pose estimation of the vehicle with respect to the fishnets. In this paper, a novel Xjunction detector (Fast-Cross) is proposed, which is able to detect fishnet knots and their topology in camera images. The detected knots are used for estimating the pose of a camera relative to fishnets. This avoids the tracking of image features and therefore does not suffer in situations with repeated scene structures. The algorithm is evaluated on test images taken at a fish pen in Norway, and on images in a controlled underwater test facility. The results show that the proposed algorithm provides robust features and stable pose estimation in the tested environments.

Projekte

Weitere Links

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence