What is a clinical position?

What non-trivial algorithm can I use to derive a smooth heading (azimuth) from x-y position data that includes some noise?

• I have high-precision and fairly high-accuracy GPS position data for a person walking, and I wish to derive the heading at each data point.  My data are discrete time x and y values. The trivial answer is to calculate the vector from each x,y position to the next x,y position, however this doesn't work when the walker stops moving, or when the noise is greater than signal between positions. In other words, if there is little difference between one position and the next, the heading should take into account prior and future heading values. I can't just smooth the position track and interpolate headings because I need the heading data linked to the same discrete time datum as the x,y position it refers to. I think a Kalman filter might help, but I would like to hear of any other possibilities, and/or how best to tune a suitable Kalman filter.

• Answer:

  Can you use Hidden Markov model ? The fundamental assumption of kalman is its linear dynamic. I think you would be better with state space model . You can use probabilistic methods to train the markov models and determine the next probable move/location. If you have a GPS log data, you can quickly develop both and see which one is able to predict better and which allows you to improve the prediction engine by having feedback loop back to the model.