Distinguishing a Drone from Birds Based on Trajectory Movement and Deep Learning

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Computer Vision

Background:

• Unmanned aerial vehicles (UAVs) and birds share low-altitude airspace, necessitating reliable identification for safety and security.
• Traditional appearance-based recognition fails in long-range scenarios where targets are small and lack detail.

Purpose of the Study:

• To develop a robust method for distinguishing between drones and birds in long-range aerial surveillance.
• To leverage motion trajectory analysis as a complementary cue to appearance-based detection.

Main Methods:

• A model-driven simulation pipeline generated synthetic data of multicopter, fixed-wing UAV, and bird trajectories.
• A bidirectional long short-term memory (LSTM) network was trained on time-series image-plane coordinates and apparent size to classify motion patterns.
• The model analyzed trajectory characteristics like smoothness, turning behavior, and velocity fluctuations.

Main Results:

• The LSTM network successfully classified drone-like and bird-like motion patterns on synthetic data.
• Motion-trajectory cues alone enabled reliable separation of drones from birds, even with scarce visual details.
• The method demonstrated generalization capabilities for real-world surveillance systems.

Conclusions:

• Motion-trajectory analysis is an effective method for early distinguishing drones from birds in challenging long-range scenarios.
• The proposed simulation and classification pipeline provides a reproducible testbed for trajectory-based recognizers.
• This approach offers a complementary and robust signal for integrated surveillance solutions, independent of appearance-based limitations.