Enhancing railway infrastructure monitoring using hybrid VTOL drones: a case study on inspection and surveillance using custom YOLOv12 object detector

Hybrid VTOL drones represent a promising solution for patrolling railway lines for security purposes and conducting macroscopic infrastructure integrity checks. These systems can be an alternative to multi-rotor drones, especially in applications requiring less detailed analysis, such as mapping, macro-object identification, or obstacle detection. To achieve this purpose, a model based on a specifically tailored revision of YOLO12 was trained on a previously gathered dataset, acquired to model different use cases in real-world applications, and tested, laying the basis for an end-to-end embedded framework able to provide railway segmentation and obstacle 3D reconstruction via an end-to-end processing pipeline based on deep learning and photogrammetry. The results demonstrated the feasibility of the proposed methodology under real-case scenarios and laid the basis to be combined with segmentation methods and photogrammetry to provide a complete processing pipeline for the automatic, real-time identification of relevant obstacles on railway tracks.