iHunter: Empowering Autonomous UAVs for Robust, Real-Time Aerial Terget Tracking and Trajectory Prediction

Overview

Autonomous aerial systems are rapidly transforming surveillance, security, and search–rescue operations—but achieving reliable, real-time detection and tracking of agile aerial targets remains challenging. In this talk, I introduce iHunter, an integrated framework that empowers UAVs to robustly detect, track, and forecast the trajectories of moving targets by addressing two critical challenges: resilient object detection and precise trajectory prediction.

To mitigate intermittent misdetections by primary vision-based systems, we developed our SMART TRACK framework, which fuses multi-modal perception with high-frequency state estimation. By leveraging Kalman filter as feedback to rapidly re-detect targets, SMART TRACK attains sub-meter tracking accuracy. For forecasting agile trajectories, we propose two approaches: an online selection of simple kinematic predictors and a GRU-based neural network for rapid velocity prediction—both validated through simulation and field experiments. Complementing these advances is our FLIGHTGEN framework, which automatically generates large-scale synthetic UAV trajectories using realistic simulations. Remarkably, the GRU model trained solely on this synthetic dataset provided accurate predictions in real experiments with unseen target trajectories, underscoring the effectiveness of our sim2real approach. Together, these innovations dramatically enhance tracking stability and predictive performance, offering a comprehensive solution for aerial surveillance and dynamic target interception.