Luca Francesco Rossi

In recent years, state-of-the-art algorithms have demonstrated exceptional capabilities in the recognition and detection of elements within images and videos. These technologies have often achieved performances comparable to, if not surpassing, human capabilities. Despite these advancements, automating the unambiguous identification of entities in heterogeneous environments remains a significant challenge. This research proposal aims to address this gap by investigating novel methods and algorithms to achieve a more fine-grained identification of entities, particularly in situations where ambiguity prevails. While current algorithms excel in controlled settings, the dynamic nature of real-world scenarios introduces numerous challenges. Unpredictable factors, such as shadows, partial occlusions, reflections, and adverse lighting or weather conditions, significantly impact the robustness of automated identification systems. Real-time processing imposes stringent constraints, necessitating algorithms to operate at a 60 Hz rate for timely decision-making. Technical complexities, including camera calibration, precise pose estimation, and unfixed focal distances while zooming, are crucial for accurate spatial mapping, ensuring the precise localization of identified entities.

---

This Ph.D. proposal arises as part of an executive path built on an industrial collaboration with netventure Group, a company specializing in live broadcast graphics, augmented reality, and virtual graphics for international sporting events, live shows, and entertainment productions worldwide. Given this nature, practical applications will be primarily tailored for dynamic sporting environments and computer vision applications related to the intricacies of athletic activities.