Exploring the most appropriate feature detector and descriptor algorithm for on-board UAV image processing

Abstract

With the development of computer vision technology, many researches about feature detectors and descriptors have been published in the last decades. In order to explore what kind of approaches are appropriate for unmanned aerial vehicle (UAV) onboard video processing, the popular feature detectors and descriptors are analyzed and combined with each other. Three practical videos captured in indoor environments and outdoor environments are used to test the accuracy, runtime and robustness of these combined algorithms. Results validate that the combinations of different feature detectors and descriptors balance well the accuracy and runtime. This will provide references for choosing appropriate onboard video processing algorithms.