The Signals, Images and Systems (SIS) team of the I3S Laboratory conducts research in the fields of signal and image processing, multimedia data compression, communication networks, robotics and autonomous systems, and optimal experimental design. The team is composed of 30 permanent members, and approximately as many non-permanent members at a given time, including PhD students postdoctoral researchers, interns and visitors. The team is composed of UNS, CNRS and INRIA personnels.
- Biological and Biomedical Signal and Image Processing - This axis addresses biological and biomedical (‘bio’) problems by means of signal and image processing tools. Multidisciplinary in nature, this axis is characterized by a tight collaboration with actors of a variety of fields, from methodology (applied mathematics) through instrumentation (optical physics) to applications (biology, cardiology). The originality of the approach undertaken here lies in the fruitful constant interplay between theory and application: the bio problems to be tackled define the needs driving the development of signal and image processing methods specifically tailored to the applications under study, but novel methods often also find application beyond the scope of the axis. Signal processing tools are mostly developed to answer clinical and physiological demands, while image processing methods are devoted to biology. Biological problems are the main focus of the Morpheme joint team with both INRIA and IBV, including various applications from neuron development modeling to cell growing characterization. Regardless of the application domains, researchers in this axis share common interests in the areas of signal and image analysis and modeling, imaging systems and machine learning. More information about our biomedical signal processing activities can be found in the Signal group's webpage. Our biological image processing activities are carried out in the Morpheme group.
- Wireless Communications Systems and Networks - The research activities on wireless communication systems and networks span both theoretical and applied aspects. On the theoretical side, our research works started ten years ago with the objective to apply tensor decompositions to nonlinear system modeling and identification. This pioneering work was pursued, during the period, with the double purpose to reduce the parametric complexity of Volterra models and to develop efficient parameter estimation algorithms for structured tensor models. A second important application of tensors has concerned the design of MIMO systems (first point-to-point and then cooperative), including tensor coding and resource allocation, with semi-blind receivers. These fundamental works have led to the development of several new tensor models. On the applicative side, our research has focused on cognitive radio, with more recent developments in the field of massive MIMO systems. Substantial emphasis has been laid on problems at the network and transport layer on the one hand and at the application layer on the other hand. The former relates to our investigation of various networking issues (TCP and SDN) in data centers, which form the core of cloud solutions and our activity in network measurement applied both to network neutrality and the study of the control plane of the Internet (BGP). The latter covers the Quality of Experience (QoE) guided design of current and future wireless access networks and also the distribution of content within ephemeral (infrastructure-less) networks. More information about our works focused on tensors can be found in the Signal group's webpage. Our networking activities are carried out in the Signet group
- Multimedia Coding - The main objectives of this research activity are to develop coding solutions for modern multimedia objects: images (2D or 3D), UHD/4K video, massive 3D objects, animations, etc. The theoretical tools developed in this research activity include wavelets, geometry coding, sampling optimization, parametric statistical classification and bio-inspired image processing based on neuroscience mathematics (dynamic filtering, spiking neuron models. . . ). The members of this axis have several national and international collaborations (Portugal, Italy, Tunisia, Brazil) and participate actively to the National GDR ISIS. In addition, the project is involved in several industrial contracts with partners such as Orange Labs, CNES, Alcatel Alénia Space, IFP Energies Nouvelles, 4G-SGME, ETSI... More information about our data coding activities can be found on the Mediacoding group's website.
- Autonomous systems - This axis focuses on the development of the fundamental science and technology underlying the perception and control of mobile robots. The first topic is non-linear control of aerial vehicles, aimed at delivering global or at least semi-global stability and robustness with respect to dynamically changing environment conditions. Solutions focus on the underlying non-Euclidean nature of the state representation of a flying vehicle, typically the special Euclidean group SE(3) for pose control or the special orthogonal group SO(3) for attitude control. They should be tailored to work naturally with sensor systems that can be effectively mounted on the vehicle and deal with the high noise levels of such sensors. The second topic is real-time sensor perception and particularly real-time visual localisation and mapping (visual SLAM). Robust non-linear estimation algorithms offer the potential to provide the location and orientation of autonomous systems within complex and varying 3D environments. Providing the dynamic reconstruction of the environment allows autonomous systems to interact directly with the environment. This group focuses on developing robust, efficient and accurate large-scale 3D world models with low-cost consumer sensors (such as stereo cameras and RGB-D sensors). In particular, the group focuses on applications such as the multi-contact control of humanoid robots as well as real-time augmented reality application.
- Observation and Modeling - The core of the activities of this axis is the investigation of statistical tools and methodologies for the design of optimal experiments (DOE) in prediction and estimation problems. We study both non-parametric and non-linear parametric models. In the two cases, a central interest is on capturing the dependency of the optimal designs on the unknown distribution of the data, and on constructing efficient algorithms for their approximation. For instance, this resulted in the past in the definition of design criteria adapted to a particular model non-linearity, without resorting to simplifying normal asymptotic arguments. More recently, in the context of our participation to the GdR MASCOT-NUM, we focused on the design of experiments for random fields. Both model-free space filling designs — related to sphere packing and sphere covering problems — and designs for Gaussian processes are studied. Applications of our results to computer experiments, as well as modeling of environmental phenomena, are ongoing in the framework of various national and international collaborative projects. More information about these activities can be found on the webpage of the Design group.