A Computationally Intelligent Framework for UAV Forced Landings

D. Fitzgerald, R. Walker, and D. Campbell (Australia)


Uninhabited airborne vehicles (UAV), UAV forced landing, UAV safety, computationally intelligent framework, machine vision, fuzzy systems, radial basis probabilistic neural networks, classification


A computationally intelligent framework has been developed for the forced landing problem for uninhabited airborne vehicles (UAVs). This framework locates landing areas within an image that are of the appropriate geometrical size and free of obstacles. The surface-type of the candidate landing areas are classified into categories such as grass, trees and water. The classification results are combined with other information such as, the spatial relationships between candidate areas, the presence of moving objects (for example cars and people) and the objects in surrounding the areas, to nominate candidate UAV forced landing sites. A discussion is presented that shows that a type-2 fuzzy based approach is expected to be useful in resolving data-set uncertainties allowing a reliable UAV forced landing site recommendation to be made. Examples of data-set uncertainties include the surface type classification and the models of motion of various objects. Results are presented showing the successful location of appropriate candidate UAV landing sites. A success rate of 90% has been achieved using a neural network classification approach and based on the testing of 500 images. These results are based on actual flight imagery collected from a Cessna 172 flight over Brisbane, Australia.

Important Links:

Go Back