MULTIPLE-MODEL BEARINGS-ONLY TARGET MOTION

         ANALYSIS FOR INFRARED SENSORS

                  Stefan Larsson
 
       Matematisk Statistik, Lunds Universitet


When trying to estimate the motion (the kinetic parameters
as position, velocity, etc.) of an aircraft you normally
use a radar, sometimes combined with a laser, as sensors.
However, both of these kinds of sensors are active and can
be detected by the aircraft. This is of course a huge
disadvantage and it is therefore desirable to look into the
possibilty of using the measurements from a passive sensor
- e.g. an infrared camera - to do target motion analysis.
An infrared camera gives very accurate measurements of the
bearing and elevation angles to the target, but it doesn't
provide range measurements. This makes the problem difficult
due to lack of observability.
 
In this thesis different estimation algorithms using some
kind of Extended Kalman Filter is investigated. First the
easier problem of estimating the motion of a non-maneuvering
target is investigated. It turns out that choice of
coordinates is an important factor for the behaviour of the
Extended Kalman Filter. Then different models for manuevering
targets is tested. The best results is got when using several
models in parallel, an approach which is called Interacting
Multiple Models. The resulting estimation algorithm can track
a wildly maneuvering target, provided the target doesn't do
any maneuvers in the very beginning.
 
The thesis can be of interest for anyone involved in
parametrized nonlinear estimation.