/
Registration: 2602194625922
The registered work includes a methodology for the generation of synthetic flight trajectory datasets based on kinematic modeling and flight envelope limits extracted from bibliographic and technical sources. Trajectories are generated through mathematical models defining stochastic acceleration patterns constrained by realistic flight limits for different aerial targets, including drones, UAVs, birds, aircraft, and missiles. This approach enables the creation of potential realistic long-range trajectories while providing flexibility and efficiency for large-scale dataset generation.
The registration also covers the procedures for processing simulated trajectories and transforming spatial coordinates into camera angle representations corresponding to tracking system actuators (yaw and pitch). These angles are equivalent to encoder measurements of real tracking systems and are used to derive velocity and acceleration profiles. The methodology includes data analysis, normalization, and feature extraction processes, allowing trajectories to be characterized through kinematic descriptors. These features can be further adapted or expanded through mathematical transformations to optimize their suitability for model training and inference.
Finally, the registered work includes the classification model that uses the extracted kinematic features to distinguish quadcopter drones from other aerial objects based on their long-range trajectories. The predictive model, trained primarily with simulated data and validated with real-world observations, is fed by dynamic motion parameters derived from camera angles. Its design enables integration into existing aerial surveillance systems and allows performance scaling through the incorporation of new training datasets. The register models correspond to two operational scenarios defined by the approximate distance between the drone and the detection system. These scenarios are characterized by ranges of approximately 1000 meters and 400 meters, respectively.
Authors: J. Sanz ([email protected]), L.Sanchez ([email protected])
This work cannot be made available to AI systems.
