École Nationale Supérieure de
Techniques Avancées

Résumé

In recent years, the practice of high-level sports has undergone a radical change. The pursuit of data and information on both opponents’ performance and strategies, as well as one’s own performance, holds a position of at least equal importance to enhancing individual performance and maintaining a healthy lifestyle. This is particularly true for team sports, such as handball, where it’s possible to record not only one’s own matches but also those of opposing teams. Similarly, due to the media coverage of sports, numerous multimedia channels capture the matches. However, few individuals are equipped to film in a way that enables to extract data efficiently. This report will focus on experimenting with computer vision techniques and artificial intelligence in order to enable individuals to film a handball match using their phones and gather real-time data, all without requiring specialized knowledge. The techniques proposed will come more from machine learning than solely the field of computer vision. The objective is to teach a machine to calculate a homography or a similar operation, and to a lesser extent, to robustly track players, addressing the challenges of partial and complete occlusion.