Modular-Neural-Networks-for-Map-Matched-GPS-Posit

  Tactically maneuvering targets are difficult to track since acceleration cannot be observed directly and the accelerations are induced by human control or an autonomous guidance system therefore they are not subject to deterministic models. A common tracking system is the two-state Kalman Filter with a Singer maneuver model where the second order statistics of acceleration is the same as a first order Markov process. The Singer model assumes a uniform probability distribution on the target s acceleration which is independent of the x and y direction. In practice, it is expected that targets have constant forward speed and an acceleration vector normal to the velocity vector, a condition not present in the Singer model. This paper extends the work of Singer by presenting a maneuver model which assumes constant forward speed and a probability distribution on the targets turn-rate

Modular Neural Networks for Map-Matched GPS Positioning.pdf,2303323,2013-08-23