Illumination-invariant image matching for autonomous UAV localisation based on optical sensing

Abstract

This paper presents an UAV (Unmanned Aerial Vehicle) localisation algorithm for its autonomous navigation based on matching between on-board UAV image sequences to a pre-installed reference satellite image. As the UAV images and the reference image are not necessarily taken under the same illumination condition, illumination-invariant image matching is essential. Based on the investigation of illumination-invariant property of Phase Correlation (PC) via mathematical derivation and experiments, we propose a PC based fast and robust illumination-invariant localisation algorithm for UAV navigation. The algorithm accurately determines the current UAV position as well as the next UAV position even the illumination condition of UAV on-board images is different from the reference satellite image. A Dirac delta function based registration quality assessment together with a risk alarming criterion is introduced to enable the UAV to perform self-correction in case the UAV deviates from the planned route. UAV navigation experiments using simulated terrain shading images and remote sensing images have demonstrated a robust high performance of the proposed PC based localisation algorithm under very different illumination conditions resulted from solar motion. The superiority of the algorithm, in comparison with two other widely used image matching algorithms, MI (Mutual Information) and NCC (Normalised Correlation Coefficient), is significant for its high matching accuracy and fast processing speed.