The "Digital Signals & Images, Robotics" (DSIR) department is focused on theoretical tools and practical solutions in the area of signal processing, in particular audio, video, 3D-scenes, motion signals and medical images, and more generally speaking, systems of –n-D sensors delivering potentially composite and heterogeneous multi-channel information. These are aimed at being used either for localization, servoing, navigation or for compression, communication or classification. In a number of cases, real-time needs, resource limitations, particular architectures or scalability requirements impose specific constraints.
The scientific framework on which is based the department's activity relies on backgrounds such as probabilistic and kinematic modelling, statistical estimation, information theory, non-linear control, model-based tracking and sparse representations.
In this framework, four scientific axes are addressed:
- The definition of adequate features and representations for the analysis of a large variety of signals;
- The formalisation and design of versatile models, algorithms and control laws able to account for observed data and to deliver appropriate feed-back in a given context;
- The conception of efficient compression and decoding algorithms adapted to a number of architectural and applicative constraints;
- The handling of large scale multi-dimensional signals and the integration of multi-modal (i.e. heterogeneous) data.
Applications domains cover sectors such as medical imaging, robotics & augmented reality, telecommunications & multimedia and to a lesser extent, aeronautics, oceanography and security.