Naar realtime objecttracking met onbemande luchtvaartuigen

This dissertation contributes to the research on UAV-based object tracking systems by integrating innovative strategies across hardware, software, and data generation domains. As the application of drones expands into surveillance and environmental monitoring, the demand for precise and reliable real-time tracking capabilities intensifies. This research makes significant strides in four key areas: evaluating drone operating systems, employing synthetic data to bypass the limitations of scarce real-world scenarios, refining visual tracking algorithms through temporal information exploitation, and enhancing time predictability with hardware acceleration. Firstly, the study conducts a comprehensive evaluation of two real-time operating systems (RTOSes) tailored for drones, identifying the systems that bolster reliability and performance. This foundational work sets the stage for further technological enhancements in drone operations. Following this, the research addresses the challenges of training robust object tracking models through the creation and utilization of a synthetic dataset. This innovative dataset, developed using advanced simulation tools, encompasses a variety of complex scenarios that enhance the algorithm's adaptability and performance in unpredictable environments. Building upon this foundation, the study introduces improvements in Siamese visual tracking algorithms by employing a dynamic templates approach, increasing tracking accuracy by making better use of temporal information. This focus on algorithmic precision forms a core component of the software enhancements proposed. The research then shifts its attention to hardware, specifically exploring the benefits of hardware acceleration techniques such as FPGA implementations. This investigation reveals how dedicated hardware can drastically improve the predictability and efficiency of tracking operations, offering a more consistent and reliable performance. By reducing the variability in tracking latency, these hardware solutions provide a stable foundation for real-time applications. Through the integration of these research areas, the dissertation demonstrates how combining hardware acceleration, advanced algorithms, and synthetic data generation advances UAV object tracking technology. This comprehensive approach improves the reliability and sophistication of tracking systems while establishing practical guidelines for future research, emphasizing the importance of addressing real-time object tracking challenges through multiple technical disciplines.