DRONE BASED NEAR REAL-TIME HUMAN DETECTION WITH GEOGRAPHIC LOCALIZATION
Pieter-Jan Baeck, Nicolas Lewyckyj, Bart Beusen, W. Horsten, Klaas Pauly · The international archives of the photogrammetry, remote sensing and spatial information sciences/International archives of the photogrammetry, remote sensing and spatial information sciences · 2019
Abstract. Detection of humans, e.g. for search and rescue operations has been enabled by the availability of compact, easy to use cameras and drones. On the other hand, aerial photogrammetry techniques for inspection applications allow for precise geographic localization and the generation of an overview orthomosaic and 3D terrain model. The proposed solution is based on nadir drone imagery and combines both deep learning and photogrammetric algorithms to detect people and position them with geographical coordinates on an overview orthomosaic and 3D terrain map. The drone image processing chain is fully automated and near real-time and therefore allows search and rescue teams to operate more efficiently in difficult to reach areas.