Knowledge-based recognition of man-made landmarks in a simulated control cycle using a virtual-globe system

Automatic knowledge-based recognition of landmarks in aerial images for UAV navigation is an alternative to GNSS navigation. It provides absolute position estimates thus complementing INS navigation. Relying on knowledge instead of template images or training samples is advantageous because the first may be out-of-date and the latter not representative. The robustness and precision of the method can be assessed using internet-based virtual globe systems such as Google Earth as camera simulator. This provides an almost open-world test-bed, in which the recognition system can be operated in a perception/action loop. The system proposed here has been elaborated for simulated flights over Germany using mainly highway bridges as landmarks. But its behaviour is also investigated over quite different regions. Of course also other landmarks such as churches or mosques can be used as landmarks.