Sébastien Piérard

Welcome on my website !

I am a research engineer specialized in the field of motion analysis, and more specifically in the interpretation of human movements. This is a challenge with many potential applications. For example, my multi-disciplinary project GAIMS introduces objective measures in the longitudinal follow-up of patients with multiple sclerosis, with a clinical gait analysis based on non-intrusive sensors. My areas of interest are threefold: the data acquisition, the decision making (or predictions) and the evaluation of the entire process.

Data acquisition. The choice of the most appropriate sensors for the intended application is crucial. In my research, I put emphasis on non-intrusive measures. This means that the sensors (color cameras, 3D cameras, range laser scanners, etc.) can be placed around the observed scene, but not inside. In particular, we avoid equipping the observed persons with any sensor or marker.

Decision making. Unlike other other fields in computer science, the relationship between the inputs (the measured data) and the outputs (the decisions or the predictions) is not always well characterized. For example, if someone is asked to decide whether a binary silhouette is human or not, he or she will be able to do so, but he or she will not be able to explain how it is done, so that the human expertise cannot be easily converted into algorithms. That's why I am also interested in machine learning techniques.

Evaluation. A carefully designed evaluation methodology is the cornerstone to help with the various design choices related to the data acquisition and to the decision making. This covers the optimization, the comparison, and the selection of the most appropriate decision strategies. From my experience, small details of the evaluation methodology, often overlooked in the literature, can have a considerable impact on the drawn conclusions. Indeed, the problem to be solved remains insufficiently characterized until there is a clear understanding of how the proposed solutions should be evaluated.