Image database of low-altitude UAV flights with flight condition-logged for photogrammetry, remote sensing, and computer vision

Abstract

The growth in the number of aerial images available is stimulating research and development of computational tools capable of extracting information from these image databases. However, developing a new computer vision (CV) software is complicated because many factors influence the extraction of information from aerial images, such as lighting, flight altitude, and optical sensors. The CV has been incorporated in most modern machines such as autonomous vehicles and industrial robots. The aim is to produce a high-quality image database of low-altitude Unmanned Aerial Vehicle (UAV) flights with flight condition-logged for photogrammetry, remote sensing, and CV. This work resulted in a collection of aerial images in the visible and thermal spectrum, and this set of images was captured in different schedules of the day, altitudes of flight, times of the year. The cameras are synchronised with the UAVs autopilot, and they were spatially and spectrally characterised in the laboratory. This research makes available low altitude aerial images of a region in Brazil to all community, with the precise flight and capture information, as well as additional features such as ground truth and georeferenced mosaic. Examples of the use of the database are shown for mosaic generation and development of CV algorithms for autonomous navigation of UAVs [1,2]. Furthermore, this database will serve as a benchmark for the development of the CV algorithms suited for autonomous navigation by images.