Abstract
This master thesis studies the synergies between 5G-supported drones and cloud systems with a main emphasis on helping in human detection for search and rescue operations during natural disasters. The foundational technological components, such as the Jetson Nano, 5G Module, 360 cameras, and GPS, make up an advanced ecosystem that supports innovative disaster response applications. Significantly, much importance is attached to the concept of 5G technology, which implies an increased data transmission rate, decreased ratio of lost packets, and enhanced communication. This enhances human detection systems’ speed and efficiency since it allows the collected data from drones to be transferred in real time and ensures a reliable link with cloud systems. Within the implementation phase, the YOLOv5 series employs state-of-the-art computational platforms such as OpenCV, CUDA, cuDNN, and PyTorch, reflecting a commitment to leveraging technologies for precise and reliable human detection. Therefore, from the set of models, the evaluations’ critical approaches reflect the levels of performance that depend on the datasets applied for testing. The research concludes with a strategic exploration of confidence score optimization, aiming for a delicate balance between a high F1 score and elevated confidence thresholds to achieve optimal human detection. In this thesis work, we design and implement an integrated 5G-drone-cloud system that is optimized for human detection. We evaluate our solution's performance using a public dataset. We compare the performance of different machine learning models and system architectures, which helps explore the design space for human detection in critical search and rescue missions during disasters.