Modeling and Information Processing

Estimation

Objectives

To define and characterize estimators from non-linear and implicit observations (signals or images) in an ambiguous context, taking into account possible data association problems.
To design efficient algorithms to compute these estimators, using an accurate knowledge of the system into which the processing must be integrated (sensor modeling, dynamic model, etc.).

Contribution

The aerospace and defense fields provide numerous estimation problems that can be ordered according to the nature of the observation relationship between the data available (signals or images from various sensors) and the underlying variable that we are trying to estimate. Starting from the simple linear-Gaussian context, we consider non-linear contexts, hybrid models that incorporate different types of nuisance parameters, implicit equations and finally contexts in which the observation model cannot be properly defined without addressing a data association problem.

The proposed approaches mainly fall within the general field of statistical estimation and are based on related recent tools: Markovian modeling, particle techniques, robust estimators, etc. The emphasis is on the external constraints which modify the estimation process: sensor model, background context, requirements associated with the integration into an onboard system. Our contributions concern the modelization of the problem, the design of algorithms and the performance evaluation. This last point is the focus of a large part of our work. It is crucial in operational applications, both for delimiting the application field of concurrent techniques and for choosing the relevant variables to supervise the embedded processing modules.

Applications

Support for the autonomy of systems

Navigation/Guidance
See-and-Avoid

Fire control
Surveillance
Physical measurements

Projects

ADO, data association
DOI, detecting objects of interest
DOP, detecting single objects
EMI, estimating image motions
SAM, structure and movement
SR, super-resolution
TAP, advanced particle techniques