Navegando por Assunto "Unreal engine"

Agora exibindo 1 - 1 de 1

Acesso aberto (Open Access)
Simulação de missões de VANTs em ambientes 3D fictícios e gêmeos digitais com georreferenciamento direto de pixels
(Universidade Federal do Pará, 2023-08-14) CONDE, Lucas dos Santos; KLAUTAU JÚNIOR, Aldebaro Barreto da Rocha; http://lattes.cnpq.br/1596629769697284
The world is rapidly entering a reality where Artificial Intelligence is becoming increasingly present in various systems, from the domestic environment to sectors such as industry, urban mobility, and agriculture. In this context, with the advancement of computational power, sim ulators for developing and testing autonomous systems have garnered significant interest from large companies and the scientific community due to the visual and physical fidelity they offer. These simulators are often regarded as "digital twins" of real scenarios and systems and have brought significant advantages in terms of cost and time, saving physical and human resources during the conception and improvement of algorithms. Among these systems are Unmanned Aerial Vehicles (UAVs), which have proven to be of great utility in contexts such as urban and rural mobility and monitoring. They are applied, for example, in detecting defects in photovoltaic panels, identifying weeds in crops, extending the mobile network and in search and rescue missions. Therefore, this work presents the conception of a methodology that integrates realistic mission simulation with UAVs, using the AirSim simulator in conjunction with the Unreal Engine graphics engine and computer vision capabilities. The objective is to perform object detection (employing the YOLO AI model) associated with their georeferenced location and generate geolocated image files that are compatible with commercial software for aerial image processing. The results were evaluated using the WebODM software for scene recon struction from geolocated image files (generating orthophotos). And to evaluate the direct pixel georeferencing algorithm, the ability of the drone to return to the position of the detected person (or object) after the algorithm provided the GPS position was tested, with errors smaller than 5 meters in relation to the real position (in UTM coordinates) of the element in the 3D environment