I am implementing kalman filter using accelerometer and GPS sensor. I wanted to know how to obtain the value of Q which is the process noise model for kalman filter.
Nikolay Krobka
Please verify the following kalman filter results in attached file lat_long_combine.bmp. It based on a constant velocity model. I am tuning the Kalman filter using Page 17 of the following bachelor thesis https://pdfs.semanticscholar.org/0b93/eb84ff2f48ea8d9e2770e4b45d30609095b1.pdf.
From the figure you show, I think your filter is performing well and this might be the best you could get from a constant velocity model. However, there is offset in the results which is apparently due to accelerometer bias. Try to have static data (vehicle is stationary ) at the beginning of the experiment and log the accelerometer output measurement. average this stationary data so you get the sensor bias then subtract this average from the whole trajectory data.
Thank you Malek Karaim I forgot to mention in my figure that red line represents kalman filter estimates and the blue line represents the GPS measurements.
Was your analysis is based on that ?
Presently I am using the process noise model in the figure above
in my constant velocity Kalman filter implementation.
acceleration variance =0.0625 which is based on the following thesis
https://pdfs.semanticscholar.org/0b93/eb84ff2f48ea8d9e2770e4b45d30609095b1.pdf.
I will try the suggestion you have provide to reduce the average of stationary acceleration data from the whole trajectory data.
Article Multi-site modeling and prediction of annual and monthly pre...
- Badges
- Science topic
- Similar topics
- Methodology
- Sensors