Graphical methodologies for early environmental disasters analysis from aerial vehicles

This thesis presents the research conducted during a PON-funded PhD under the initiative "Research topics in systems for urban environment safety, environmental monitoring, and prevention of critical or risk events (Action IV.5 - GREEN)". The work focuses on applying innovative image-based methodologies to address natural disaster risks, particularly through aerial imagery analysis. Three key application areas are explored: landslide damage assessment, early wildfire detection, and search and rescue operations for missing persons. To tackle these challenges, the research develops automated and scalable systems integrating image processing, georeferencing, and human-machine interfaces. The goal is to support timely and informed decision-making in emergency scenarios through rapid data acquisition and real-time analysis. Different methodologies were evaluated, including both classical image processing and modern deep learning techniques. Artificial Intelligence models, particularly tailored for small-object detection, showed significant advantages in terms of accuracy, speed, and adaptability. A dedicated wildfire detection model was designed for UAV deployment, addressing issues such as smoke and fire identification, real-time processing, and precise localization. A communication architecture was also developed to enable efficient data exchange between aerial platforms (both drones and manned vehicles) and ground stations. This facilitates real-time situational awareness and operational coordination. Furthermore, a digital environment combining image processing, operator interfaces, and augmented reality was created to enhance user experience and field responsiveness. The research highlights the superiority of AI-based approaches over traditional techniques for complex and dynamic environments. Results from simulations and case studies validate the effectiveness of the proposed solutions. In conclusion, the thesis delivers a comprehensive methodology for disaster monitoring and response, combining detection, segmentation, communication, and visualization. Future developments will focus on increasing model robustness, integrating multi-sensor data, and expanding the system's applicability to broader environmental monitoring tasks.