More Robust Features for Adaptive Visual Navigation of UAVs in Mixed Environments

In this paper, we present an autonomous visual navigation system that determines the location of the unmanned aerial vehicle (UAV) in GPS-denied environment by detecting semantic features (roads centrelines, intersections, outlines of forest and river) in aerial imagery and matching them to a pre-built dataset. This work is centred around testing the capability of a road centreline modelling and matching algorithm to localise accurately. Alongside, dynamic feature modelling and minimalistic description to optimise data association are proposed. We test three novel datasets with satellite imagery covering the same rural area with significant seasonal and lighting variation.