Improving Sensor Scheduling in UAV Multi-Target Tracking with Sensor Performance Models

In recent years, unmanned aerial vehicles (UAVs) have gained significant interest for multi-target tracking, particularly in surveillance, search and rescue, and environmental monitoring. This thesis focuses on tracking multiple ground targets using a UAV equipped with a steerable, variable focal length visual sensor, addressing challenges in perception, sensor scheduling, and control. A novel sensor performance estimation method is introduced, linked to defined perception levels that quantify target perception quality. A modular system architecture for UAVs, sensor performance models (SPMs), and sensor scheduling based on a Partially Observable Markov Decision Process (POMDP) are proposed. Real-world imagery is used to train and evaluate these models. Extensive Monte Carlo simulations assess scheduling and control algorithm performance across different scenarios. This thesis contributes to UAV-based multi-target tracking by developing and evaluating novel methodologies, offering insights for future advancements in the field.