Automatically detecting changes and anomalies in unmanned aerial vehicle images

Abstract

The use of unmanned aerial vehicles (UAVs) in civil aviation is growing up quickly, enabling new scenarios, especially in environmental monitoring and public surveillance services. So far, Earth observation has been carried out only through satellite images, which are limited in resolution and suffer from important barriers such as cloud occlusion. Microdrone solutions, providing video streaming capabilities, are already available on the marketplace, but they are limited to altitudes of a few hundred feet. In contrast, UAVs equipped with high quality cameras can fly at altitudes of a few thousand feet and can fill the gap between satellite observations and ground sensors. Therefore, new needs for data processing arise, spanning from computer vision algorithms to sensor and mission management. This paper presents a solution for automatically detecting changes in images acquired at different times by patrolling UAVs flying over the same targets (but not necessarily along the same path or at the same altitude). Change detection in multi-temporal images is a prerequisite for land cover inspection, which, in turn, sets up the basis for detecting potentially dangerous or threatening situations.