Determinação de Riscos de Sobrevoo por Visão Computacional

In the 20th century, aviation has proven itself as one of the most importanttechno-social revolutions of modern human history. Now,Unmanned Aerial Systems (UAS) are on the verge of leading thenext techno-social revolution, remodelling several areas and movinga global market that does not stop growing. The establishmentof UAV Traffic Management (UTM) systems is necessary to makethis revolution happen, and one of the main obstacles is the risksinvolved in UAS flight, especially in urban environments. A methodologyto estimate risks on the overflown environment was carriedout by the usage of convolutional neural networks on the imageryof such environments, allowing the definition of safer routes andtheir management on the fly. Development and experimentationprocesses were carried out, with promising results including a convolutionalneural network for pixel-wise domain that was capableof estimate risks from satellite imagery and return overflight risksheat maps.