Thesis of Nicolas Jacquelin

Top-level athletes movements extraction from video streams


The improvement of top-level athletes performance requires a detailed understanding of the athlete's movement. However, this movement remains complex to characterize because it is multi-scale, at the level of the same individual and his skeleton, but also of several individuals (team) and their environment (ball, accessory, field). Video streams remain the most efficient method of collecting this information because they are non-intrusive. This thesis aims to extract quantitative (geo-spatial motion, time series) and structured (names of individuals, body parts, type of motion) information from video streams.

Advisor: Romain Vuillemot
Coadvisor: Stefan Duffner