-
Santiago Felipe Luna Romero
Graduate Program on Health Technology, Pontifícia Universidade Católica do Paraná, 80215-901, Curitiba
-
Charles Roberto Stempniak
CEO at AUTOMA VISION Tecnologias de Inteligência Artificial Ltda., Paraná, Curitiba
-
Mauren Abreu de Souza
Graduate Program on Health Technology, Pontifícia Universidade Católica do Paraná, 80215-901, Curitiba
-
Gilberto Reynoso-Meza
Industrial and Systems Engineering Graduate Program, Pontifícia Universidade Católica do Paraná, 80215- 901,Curitiba
Keywords:
Pedestrian detection, Thermal imaging, Unnamed aerial vehicle (UAV), Transfer Learning, Yolo, Computer Vision
Abstract
Smugglers use a variety of tactics to bring goods into a country without paying taxes. These bandits usually operate at night, when the weather and lack of light make it easier for them to move goods. Smugglers' entry points are typically rural areas with a vast distance between them and the city. Thermal cameras and drones are the most effective tools for detecting smugglers because they can identify items that are unaffected by weather, lighting, or body posture, and they can cover enormous amounts of terrain quickly. This research proposes a new thermal model that uses transfer learning from a small hand-labeled database and another model created from a public thermal database to recognize 49 objects with 524,700 thermal images, as well as define and propose a final model of pedestrian recognition in thermal images captured with a drone that has a 94% accuracy.
