The general area of the 3D-SAM team will be to advance methodologies and algorithms for modeling and analyzing static and dynamic 3D shapes. Specifically, we will develop tools and methods for :

Shape analysis of surfaces: a large number of applications involve the statistical analysis and modeling of 3D objects that can be represented as surfaces in 3D. The difficulty in analyzing shapes of such objects comes from noisy and missing data, variability in pose and articulation, arbitrary mesh parametrizations during data collection, and shape variability within and across shape classes. These issues require metrics, representations and statistics that should have certain invariances and robustness to the aforementioned variabilities.

Modeling dynamics of shape variation in Videos: many human-machine interaction environments are characterized by temporally-evolving shapes of 3D objects. Examples include gesture and action recognition using full body scans, emotion classification using facial data and so on. It is critical to identify precise mathematical representations of underlying shapes, and then impose efficient dynamical models on representation spaces that capture the essential variability in shape evolution.

In the following we will describe some applications with high societal and economic impact such as biometric and security, health and retail sentiment analysis :

• Applications to Human-Machine Interaction using 3D Sensing: we will target some specific application scenarios such as 3D human activities recognition to implement and evaluate the proposed techniques. An important application for future human-machine interaction is to develop estimation of human-state using 3D sensing.
• Applications to 3D Face Dynamic Sequence Analysis: we will address the facial geometry analysis for recognition and facial expression recognition from 3D sequences (3D plus time, or 4D).
• Applications to 3D Retrieval in Large Datasets: the use of 3D-model databases is growing both in number and size. To browse and search in such large and high-quality datasets, the 3D retrieval methods developed so far must be adapted and extended both in accuracy and speed.