Toward Enhanced Tracking of Small Airborne Objects Using Transformer-Based Particle Filter

The widespread proliferation of small civilian drones has raised growing concerns about their potential misuse and the difficulty of real-time airspace monitoring. These airborne objects are often minute in scale and exhibit irregular motion patterns, making them difficult to detect and track consistently. To address this challenge, we propose a tracking algorithm that integrates a deep learning-based time-series prediction model into a particle filter framework to improve robustness and real-time performance. The tracker receives measurements from an external object detector and employs a transformer-based prediction model during both the prediction and update steps to mitigate the inefficiencies of conventional sampling. Additionally, the Hungarian algorithm is applied to perform optimal matching in multitarget tracking scenarios. Experimental results confirm that the proposed algorithm outperforms conventional methods, particularly in low-visibility and sparse detection environments, and is capable of reliably tracking small airborne objects over mid- to long-range distances (e.g., 100–300 m).