Kalman filter
Ashok from India  [1 posts]
14 years
    Hi, I've heard that the Kalamn filter is a very good filter and it predicts the current position of a moving object. I know RR is very fast in Image Processing. But it would be very good if we were to use the Kalman filter as it would further increase the speed of Processing.
     I've searched the newer version for the Kalman Filter but was unable to find it. I would like to know if the Kalman Filter module is available in RR, if not I request you to include that too which would make RR even more comprehensive.... Sorry for keeping the post long...
Anonymous 14 years

Sounds like a good idea. We've added it to the list. Unfortunately that list is very VERY long and thus we don't expect to see that filter added for many months .. but it WILL be added at some point.

Thanks for the suggestion.
from United States  [214 posts] 14 years
Hi Ashok,

I too was looking at the Kalman filter for predicting object position but then I realized that it might be overkill for most of my needs.  I don't know your application, and maybe you've already tried this, but I have found that I can
simply use a speed measure to predict object position.  In other words, as you track object position, also track the change in object position divided by the sampling interval.  Then use this estimate of speed plus the current position to predict the next object position.  If the object motion is significantly non-uniform, you can also throw in acceleration by tracking the change in speed as well.  Since position sensors like IR and sonar tend to have a little noise from sample to sample, I usually use a running average over 3-5 samples (sampling at 10 times per second), otherwise the speed and distance estimates can take sudden random jumps and throw things off.

Using this technique, my robot can successfully "catch" a rolling ball over a wide range speeds.

Rolfiboy from Denmark  [1 posts] 14 years

I will just let you know that i am looking forward too se a kalman filter module for RR. I am working on targeting system for a robot, that would lock on a target(one target out of multiple targets) and hold the lock even is the target and robot moves. I think the kalman filter could be used to make the prediction more precise at higher speeds.
Anonymous 14 years
Thanks for the note. This will help to increase that module's priority.

pranavtrehun007 from United Kingdom  [33 posts] 13 years
My vote also goes for Kalman Filter. Blob tracking would be more stable and useful with it.

What i am expecting is that we give it a point like COG X and COG Y. and it outputs another set of Kalman_COG_X and Kalman_COG_Y. The new points will be more stable and immune to noise etc.
Later we can have use fusion of sensor data etc. I am a newbie when it comes to kalman filters.

Anonymous 13 years
Not quite a Kalman filter but the new variable filter module should help in this regard. Taking the mean of the last X numbers should help to smooth out the signal and provide a more stable number.


Note that the variable is averaged and placed back into the same variable. If you want to preserve the original value use the Set_Variable to set another variable with the value BEFORE the filter variable module is used.


This forum thread has been closed due to inactivity (more than 4 months) or number of replies (more than 50 messages). Please start a New Post and enter a new forum thread with the appropriate title.

 New Post   Forum Index