Real-time targeting and geolocation system based on UAS thermal imagery

Abstract

Nowadays the use of unmanned aerial vehicles is becoming increasingly popular, even in critical situations like emergencies, fire brigades are starting to use drones as a source of information. After a deep research process and analyzing potential user's needs, this project pretends to help incident commanders with a new system able to target and geolocate hotspots or hot areas in real-time, by using UAS thermal imagery. Since this master thesis has been developed at Unblur, a Barcelona-based startup, the designed system has been integrated on its web platform called IRIS, which combines both static and dynamic incident information, so as to ease decision-making and to provide a general overview of the situation. Full system's architecture is composed by two clearly different parts, the geolocation algorithm and the image processing workflow. The geolocation model is based on the inversion of the pinhole camera model, and then doing an approximation by using a DEM. Image processing has been first developed and tested in Python 2.7 with OpenCV library. Once it has been checked results correctness in terms of detection and computing time, it has been programmed a similar procedure in C# by using Accord.NET library. Finally, it has been performed subsystems integration on IRIS and after some flight tests, by means of a DJI Matrice 210 and a Zenmuse XT camera as a payload, it has been concluded that the resulting system is adequate to the proposed application, not only in terms of performance but also in time response.