Detección automática de meteoros en redes de cámaras comerciales de seguridad

Abstract

The recovery of meteorites is one of the most valuable resources to provide clues to understand the origin, formation and composition of our Solar System. Thousands of them hit the Earth’s atmosphere every day, but the vast majority disintegrate fastly during their path and do not reach the ground. In contact with the air molecules and due to their high kinetic energies, they leave luminous trails in the sky, called meteors, from which it is possible to calculate their trajectory, both before entering, in their orbit, and later, when they impact. Having appropriate instruments to detect them at its entry and being able to analyze all the data obtained with them is necessary so, in case of impact, we can know the location and origin of the meteorite. In this work, the architecture of the Fireball Alert and Exploration Terrestial Observation Network (FAETON) meteor network has been implemented, which uses commercial video surveillance cameras for detecting bright meteors. In addition, a software for processing the image sequences, analyzing them and confirming the presence of meteors has been developed. Furthermore, different techniques currently used in Computer Vision have been introduced, with the intention of enlarging the state-of-the-art in the tracing and characterization of its trajectory in images with geometric distortion. A total of 2824 possible detections have been analyzed using Neural Networks, reaching a precision of 88.0 % in correctly classified meteors and a 4.6 % of false positives.