Seite - 110 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

III. OBJECT DETECTION An UXO risk for a region of interest is derived from various indicators of combat activities on historical aerial images. These could be destroyed buildings, anti-aircraft artillery positions, trenches or bomb craters; the latter ones are by far the most numerous and simultaneously the most difficult to reliably identify on aerial images, as they can easily be confused with other small round objects such as trees [3]. The development of strategies for automatic detection of such combat indicators is scheduled for the current year. We are planning to adapt state of the art machine learning approaches to the problem; we hope to be able to exploit the fact that typically a time series of registered aerial images is available for the region of interest. As the task at hand is a critical one, a human expert will always be required to validate and refine the results. We will thus, just as for the registration problem, aid the user with an interactive visualization component for parameter exploration. IV. IMPLEMENTATION In order to maximize both the benefit to our industrial project partner and the usage and testing of our methods, we have been developing software tools that blend in to their daily workflow, namely in the form of plug-ins for their preferred GIS software. The first working prototype of the registration component was delivered in February 2017 and tested in both the German and Austrian branch of the Luftbilddatenbank GmbH. Apart from minor bugs and usability issues, the overall feedback was positive and encouraging. REFERENCES [1] M. P. Heinrich, M. Jenkinson, M. Bhushan, T. Matin, F. V. Gleeson, S. M. Brady, and J. A. Schnabel, “Mind: Modality independent neigh- bourhood descriptor for multi-modal deformable registration,”Medical Image Analysis, vol. 16, no. 7, pp. 1423–1435, 2012. [2] D. G. Lowe, “Distinctive image features from scale-invariant key- points,” Int. J. Comput. Vision, vol. 60, no. 2, pp. 91–110, Nov. 2004. [3] S. Merler, C. Furlanello, and G. Jurman, “Machine learning on historic airphotographs for mapping riskofunexplodedbombs,” inProceedings of the13th InternationalConferenceon ImageAnalysisandProcessing, ser. ICIAP’05. Berlin, Heidelberg: Springer-Verlag, 2005, pp. 735– 742. [4] V. Murino, U. Castellani, a. Etrari, and a. Fusiello, “Registration of very time-distant aerial images,” Proceedings. International Conference on Image Processing, vol. 3, pp. 989–992, 2002. [5] S. Nagarajan and T. Schenk, “Feature-based registration of historical aerial images by area minimization,” ISPRSJournalofPhotogrammetry and Remote Sensing, vol. 116, pp. 15–23, 2016. 110