Autonomous UAV Safe Landing Detection
Developed a fail-safe landing detection system for Beyond Visual Line of Sight (BVLOS) drones as part of the BEAST project at NHL Stenden Professorship in Computer Vision & Data Science. Implemented semantic segmentation using U-Net and U-Net++ architectures to identify safe vs. unsafe landing zones from aerial imagery. Optimized the model through tiling, hyperparameter tuning, data augmentation, and evaluation with IoU, F1-score, precision, and recall metrics. Results demonstrate that segmentation can reliably detect ground obstacles, providing a foundation for EASA-compliant autonomous emergency landing systems.