Feature Detector Performance for UAV Navigation

J.Z. Sasiadek and M.J. Walker (Canada)


Unmanned Aerial Vehicles; Robot Control; Guidance, Navigation, and Control; Feature Detectors


The reported research was motivated by a desire to achieve accurate navigation using a camera to augment inertial navigation unit data while flying over, or through an urban environment. There are many feature detection algorithms available for this task and a reasonable question to ask is which one(s) might be most applicable to the given problem domain. In the reported research, corner detection is desired and four algorithms for finding such corners are compared and contrasted with respect to their accuracy. The three corner detection algorithms are the Harris, Small Univalue Segment Assimilating Nucleus (SUSAN), and phase congruence corner detectors. The Scale Invariant Feature Transform (SIFT) is not a corner detection algorithm, but rather a feature detector. A set of four corners superimposed on a set of eight urban scenes and at various intensity levels are the test images. It was found that the Harris and phase congruence detectors are the most useful for the problem domain. SUSAN does not find the kind of corners likely to need detection in an urban environment, while SIFT is likely to find multiple detections and miss corners. SIFT performance, in this regard, will need further study.

Important Links:

Go Back