How to use KALMAN with accelerometer?

[kawabonga]

hi
I found an article that shows how to use kalman filter to models error of accelerometer. they used markov process as stochastic error, then output of this model will be used as input of KALMAN filter.

Now, I don't what to do. I'm not sure, but I think that I need to double integrate the acceleration, then I subtract kalman's filter output from the integration result.

if this is right answer, I don't see the utility of the filter. why I don't double integrate the output of stochastic model ?

link to article : www.tkt.cs.tut.fi/research/nappo_files/Davidson08.pdf

thank you

 

[kawabonga]

I don't understand why we use stochastic model for accelerometer instead of putting its output on kalman filter ?

 

[timthereaper]

The way I understand it, modeling sensors as stochastic processes is standard since there is always some randomization in the measurement. You can't simply take your acceleration measurement and treat it as a data point from a continuous-time model and integrate it twice. It's not going to be correct. Integrating using stochastic calculus might be viable (I've never done it), but implementing a Kalman filter is probably easier.

It seems you're not sure how to use the Kalman filter. The way I understand it, the point of the Kalman filter is to use a statistical method to weigh the sensor data against how good your sensors are (measured noise/bias in a covariance matrix) and the model you're using (linear or nonlinear) to estimate what's going on (the state). You use your predictive model and your covariance matrix to come up with a "guess" or prediction of the state based on the last state estimate. Then, you get the new sensor data and fix your prediction of the state with it.