During this training, you will be able to discover usefull techniques for sensor fusion,
using SCILAB numerical computing software.
Date and place of training
Location: Toulouse (France), dates: contact us
|I - INTRODUCTION|
We will begin by trying to have a global view on the needs and on the available data sources (sensors),
then we will see how to model the dynamic estimation problems in a general way (Bayesian estimation).|
- Applications: inertial centrals, MMI, ...
- Technologies: classical motion / position sensors d(MEMS circuits, GPS, etc.)
- Representation of an object orientation (static model): rotation matrix, Euler angles (roll, tangage, lacet)
- Cinematic relations: Link between the quantities expressed in a fixed referential
and their measure in the inertial referential (measures provided by the sensors)
- Simple (static) estimation methods: Wahba problem, SCILAB solution based on the SVD.
- State representation: generic model for the description of a continuous system with hidden states, and indirect and noisy observations.
|II - KALMAN FILTER|
|In this part, will interest ourself to the Kalman filter and at some of its extensions,
especially for non linear models (EKF filter).
We will put these techniques in practice so as to realize
sensor fusion on an IMU.|
- State representation (linear model): Hypothesis, covariance matrix, observability
- Standard Kalman filter: Matrix approach and intuitive interpretation, complexity analysis
- Computing of the steady-state gain (Riccati equation)
- Extended Kalman Filter (EKF): Local linearization (computing of the Jacobians)