Feature Analysis for Maritime Object Classification in Single Images

Technical report


Michael Teutsch


Technical report IES-2010-15. In: Jürgen Beyerer, Marco Huber (eds.), Proceedings of the 2010 Joint Workshop of Fraunhofer IOSB and Institute for Anthropomatics, Vision and Fusion Laboratory, Karlsruher Schriften zur Anthropomatik, vol. 7, KIT Scientific Publishing, 2010.





In order to implement a holistic image exploitation process for detecting, tracking and classifying maritime objects, it is necessary to understand the prospects and profits of the information coded directly in a single image. In this work, a big variety of features is extracted from given detection results. These features are analyzed automatically for their separability and benefit for classification tasks. This allows the innovation of a generic system, which is able to work independent of important properties such as perspective, object appearance, image quality and environmental condition. In two application examples, the scope and limits of this work are demonstrated and further ideas are outlined for improving the current results.