Phd position on Topological and Semantic Characterization of Moving Point Clouds for Technical Gesture Learning

Objectifs de la thèse A gesture is generally represented as a labeled point cloud evolving over time. Topological data analysis [4] is a well-established approach for studying this type of data and can provide descriptors that are independent of the notion of distance between points, thereby addressing issues related in particular to morphology. However, their added value compared to classical descriptors (geometric, kinematic, or dynamic) has yet to be demonstrated, as has their expressiveness for human learning. The objective of this thesis is to explore the relevance of an analysis based on algebraic topology in the context of gesture learning in virtual environments. To this end, the thesis work will be structured around four milestones: (1) selecting a set of gestures and identifying, for each of them, the observation criteria used by experts; (2) building a dataset, labeled by an expert, comprising gestures performed using motion capture tools (marker-based in a first approach). This includes becoming familiar with the tool, data acquisition, and data cleaning using the functionalities of the proprietary motion capture software Qualisys Track Manager; (3) implementing a configurable analysis pipeline, in which it is possible to select a set of classical and topological descriptors, taking as input a gesture represented as a labeled point cloud evolving over time, and producing an analysis report, the format of which remains to be defined; (4) conducting a study on the relevance of the algebraic topology–based approach compared to an approach based on classical descriptors. Required skills • Master’s degree (or equivalent) in computer science, signal processing, or a related field; • Experience or expertise related to gestures (dance, martial arts, weight training, etc.) is a plus; • Strong interest in research and computer graphics (rendering, modeling, analysis); • Appreciation for fundamental disciplines (mathematics, algebraic topology, etc.); • Programming skills, in Python or C++. Familiarity with 3D modeling software or game engines is an asset. Other Information • Location: XLIM Laboratory, Futuroscope site, Poitiers; Fra,ce; • Start date: October 2026, flexible