[Bayes++] predict in EKF

[Sunil C. <sjc...@tp...>]

Dear All,

 

I had a query regarding using Bayes++ Bayesian Filter Classes. I am trying
to use Bayes++ classes to implement a Nearly Coordinated Turn (NCT) Model
Kalman filter. As this dynamics are non-linear in nature, I need to use the
Extended Kalman Filter (EKF). I am using, Covariance_scheme from the Bayes++
classes. 

 

In this for the predict the code is as follows:

 

Bayes_base::Float Covariance_scheme::predict (Linrz_predict_model& f)

/* Standard Linrz prediction */

{

                // Extended Kalman state predict is f(x) directly

x = f.f(x);                             

                

// Predict state covariance

                noalias(X) = prod_SPD(f.Fx,X, tempX);

                noalias(X) += prod_SPD(f.G, f.q, tempX);

 

                return 1;

}

 

In this I can implement f(x) in my predict class, so as to give the
linearized update to state.  However, the state matrix Fx also requires
re-computation as the jacobian of f(x) at x(k|k), and this needs to be
updated too before call to predict. However, I am unable to figure how this
is catered for in Bayes++, because f(x) is a const function, I cannot update
Fx in f(x).

 

Could you kindly clarify the same?

 

Thank you for your time.

 

Regards,

 

Sunil Chomal