Mohamed Younes successfully defended his PhD at 10 a.m. on May 24, 2024.
Title: “Learning and Simulation of Sport Strategies (Boxing) for Virtual Reality Training”
This thesis investigates the extraction and simulation of fighter interactions, mainly for boxing, by utilizing deep learning techniques: human motion estimation from videos, reinforcement learning-based imitation learning, and physics-based character simulation. In the context of sport analysis from videos, a benchmark protocol is proposed where various contemporary 2D human pose extraction methods are evaluated for their precision in deriving positional information from RGB video recordings of boxers during complex movements and unfavorable filming circumstances. In a second part, the thesis focuses on replicating realistic fighter interactions given motion and interaction data through an innovative methodology for imitating interactions and motions among multiple physically simulated characters derived from unorganized motion capture data. Initially, this technique was demonstrated for simulating light shadow boxing between two fighters without significant physical contact. Subsequently, it was expanded to accommodate additional interaction data featuring boxing with actual physical contact and other combat activities, along with handling user instructions and interaction restrictions.
Mohamed also took part in writing the site during his thesis, thank you very much!
We wish you the best for your future!
