Kalman Filter Implementation

To the best of my knowledge a Kalman filter relies on regular updates of various instrumentation readings during movement to do a best estimate. If you are merely trying to project the trajectory of a ball based on parameters at the time of the balls release, and no updates during the flight of the ball, then I don't think a Kalman filter is of any benefit.

TC

Hi There,

I have done some further reading, and now have some idea of what is required. And yes it is easy to implement it in a programming language, now that I know the controlling equations.

My question now is, that I have created a simple setup, with the ball just travelling at a constant velocity in the 3 directions. I am for the moment ignoring air resistance, gravity etc.

What I need now to do is calculate / state the initial values in the Q,P and R matrices. What do I do for this, pick some values that a reasonable? OR is there a better way? For Instance, I don't know the measured error for the initial point(s), only that say the error isn't going to metres wrong, more like a few cm.

You need to know some physics.

People often complain that Kalman filters have a large start up cost associated with them. What they are saying is that the it takes time for the Q, P and R matrices take a number of iterations to converge to somewhat constant values. The number of iterations is proportional to how poor your intitial estimates are.

You may also wish to look at a alpha-beta-gamma filter, which is like a Kalman filter but with constant gains. This works well if you have an accurate model of the dynamics.

-paul

Paul Oh Boondog Automation