A GOOGLE-earth based test bed for structural image-based UAV navigation

In this contribution a test bed is presented for the assessment of structural recognition methods for the localization of landmark objects in aerial images. Instead of really flying a system which might jeopardize the utilized platform - or using only a limited test image data set which is of poor relevance - the recognition system is included into a control loop with the camera being simulated by a publicly available geo system such as GOOGLE earth. The structural knowledge is represented as declarative production system. An approximately correct any-time parser is used for controlling the search. A typical example run is discussed. Such test bed can help evaluating and improving the performance of such systems for the task of autonomous visual UAV navigation.