Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Learning of utility functions for the behaviour analysis in maritime surveillance tasks

: Anneken, Mathias; Markgraf, Sebastian; Robert, Sebastian; Beyerer, Jürgen

Fulltext urn:nbn:de:0011-n-5824671 (981 KByte PDF)
MD5 Fingerprint: 30a5acc80573987dbe60c9b60a00c01a
(CC) by
Created on: 17.7.2020

Michael, J.:
Modellierung 2020 Short, Workshop and Tools & Demo Papers. Online resource : Companion Proceedings of Modellierung 2020. Short, Workshop and Tools & Demo Papers co-located with Modellierung 2020 Vienna, Austria, February 19-21, 2020
Wien: CEUR, 2020 (CEUR Workshop Proceedings 2542)
ISSN: 1613-0073
Tagung Modellierung <2020, Wien>
Conference Paper, Electronic Publication
Fraunhofer IOSB ()
maximum entropy inverse reinforcement learning; maritime domain; agents; decision support; spatiotemporal data

For detecting suspicious activities in the maritime domain such as illegal, unreported and unregulated fishing, it is crucial to counter the increasing amount of available information regarding vessels at sea with sophisticated algorithms and user interfaces. Deep learning and other data driven approaches create good results, but for an operator to be able to intervene suspicious activities, the supporting algorithms must be explainable and transparent. One possibility is to use simulations based on utility functions and behaviour models derived from the field of human behaviour modelling like game theory. Here, an expensive and time-consuming way is to model these utility functions by experts. As this is not always feasible or the behaviour patterns might not be easily expressed by experts, this work follows a different approach by utilizing inverse reinforcement learning in order to estimate the utility functions for different ship types. For this study, data based on the automatic identification system (AIS) is used for comparing the behaviour of cargo vessels and fishing boats.