Covariance matrix Q about mad-location-manager HOT 7 CLOSED

Hello. Thank you for interesting in our project.
You are definitely right about covariance matrix Q.

We have this :

double velDev = accDev * m_predictCount;
        double posDev = velDev * m_predictCount / 2;
        double covDev = velDev * posDev;

        double posSig = posDev * posDev;
        double velSig = velDev * velDev;

        double Q[] = {
                posSig, 0.0, covDev, 0.0,
                0.0, posSig, 0.0, covDev,
                0.0, 0.0, velSig, 0.0,
                0.0, 0.0, 0.0, velSig
        };

As you said, posSig should be posSigX and posSigY, but we can get speed measurements only from GPS receiver, so for both directions we have one deviation. That's why we use 1 value.

from mad-location-manager.

That's not what I meant. The off-diagonal values of Q are the cross covariance values. As everything is real in this scenario, the matrix must be symmetric (with respect tot he man diagonal). Thus the matrix should be:
double Q[] = {
posSig, 0.0, covDev, 0.0,
0.0, posSig, 0.0, covDev,
covDev, 0.0, velSig, 0.0,
0.0, covDev, 0.0, velSig
};

with two additional covDev values instead of two of the zeros in the lower part of the matrix.

from mad-location-manager.

I see now ))) Will try it today 👍

from mad-location-manager.

Thank you ))

from mad-location-manager.

Hi,
Saw that you uploaded a new version a couple of days ago.
The matrix Q in GPSAccKalmanFilter.java is still non-symmetric.
Wanted to know if there was a problem when one changes it to be symmetric?
Thanks,
Ronit

from mad-location-manager.

Hello.
I didn't upload anything yet. Version is still 0.1.11 :)

from mad-location-manager.

OK. Thanks!

from mad-location-manager.