Edouard Villain’s scientific contributions

This communication will address these questions in the practical case of man machine interaction. A consumer-grade camera is used to perform eye and face tracking with the intent of using this information to drive the computer’s mouse in the most intuitive way possible. In this scenario, we will address the main issues of building AI-based vision systems: the choice of network topology, acquisition of the learning dataset, pre-processing and labelling of the data, learning and evaluation of the model. While doing this we will comment on the traps that paved our way and the strategies we used to solve them. Finally, we compare this approach to the more traditional one with the goal of providing insights on the pros and cons of the ubiquitous usage of machine learning when building a computer vision system.