Automatic Sports Video Summarization with Identity-Aware Highlight Selection

Today, Marcos Rodrigo successfully defended his PhD thesis titled: "Automatic Sports Video Summarization with Identity-Aware Highlight Selection".

During his presentation, Marcos addressed the challenge of the explosion of audiovisual content in sports, where far more footage is generated than any viewer can consume. His research proposes a unified framework that not only summarizes the most important moments but also personalizes them according to the athlete.

These are the fundamental pillars he shared during his defense:

Intelligent Highlight Detection: He presented an evolution from classical motion-based methods (optical flow) for complex sports like martial-arts tricking, to SportCLIP, an AI-based system that understands natural language. This allows for searching highlights using simple text, without the need to retrain the model for each sport.

Identification in Real-World Conditions: A key point was the study of face recognition in sports broadcasts, where distance, occlusions, and the speed of movements often cause systems to fail. His work analyzes how to optimize CNN and Transformer architectures, so the system is robust even in long-range shots.

Personalized Video Summarization (PVS): The final result is a modular system capable of generating "identity-aware" highlight reels. The system selects the event (such as a three-pointer, a kick, or a goal) and automatically assigns it to the corresponding athlete, enabling the creation of on-demand summaries for every user.

This thesis contributes not only new models but also fundamental datasets such as MATDAT, Olympic Highlights, and UPM-GTI-Face, which serve as a foundation for the future of personalization in the sports industry.

Congratulations, Marcos Rodrigo, on this excellent work and a brilliant presentation!