Tag Archives: machine vision

733–744 M. Dlouhy, J. Lev and M. Kroulik
Technical and software solutions for autonomous unmanned aerial vehicle (UAV) navigation in case of unavailable GPS signal
Abstract |
Full text PDF (577 kB)

Technical and software solutions for autonomous unmanned aerial vehicle (UAV) navigation in case of unavailable GPS signal

M. Dlouhy¹, J. Lev² and M. Kroulik¹*

¹Czech University of Life Sciences Prague, Faculty of Engineering, Department of Agricultural Machines, Kamýcká 129, CZ165 21 Prague 6, Czech Republic
²Czech University of Life Sciences Prague, Faculty of Engineering, Department of Physics, Kamýcká 129, CZ165 21 Prague 6, Czech Republic
*Correspondence: kroulik@tf.czu.cz

Abstract:

 The article presents autonomous navigation for Unmanned Aerial Vehicles (UAV) without GPS support flying in extremely low altitudes (1.5 m – 2.5 m). Solution via visual navigation as an alternative to missing GPS position was proposed. MSER (Maximally stable extremal regions) was used as a navigation algorithm for detection of navigations objects. While GPS is useful for waypoints specification there are scenarios where GPS has unreliable signal (orchards) or is not available at all (indoor machinery halls or greenhouses). For that reason existing installed camera which is already needed for the task of inspection was used. The navigation algorithm was tested in two scenarios. The first experiment was done with dashed line marked on the floor of the hall. 8-loop testing track was created approximately 10 meters long so it was possible to fly it several times. Then outdoor experiments were performed on the university campus and park roads.
One of the discoveries was that MSER algorithm, proposed for finding correspondences between images, is possible to run in real-time. High reliability of the navigation algorithm was found during the indoor testing. The incorrect detection of the dashed line was found only in 1% of cases and those failures did not cause failure of navigation.
Although outdoor road recognition is difficult in general due to various surfaces and smoothness, MSER was able to find suitable candidates. When the UAV was fed with the parameter of road width it could verify that information with estimated distance and camera pose to accept or reject the detected pattern. The road was successfully recognized in 40% cases. Similar to the indoor algorithm in the case of navigation failure navigation along the absolute trajectory (line) was used.

Key words:

, , ,