search

psas / filters

Sensor fusion filters
7 stars 4 forks source link

Filter scrubs wrong data #1

Open philiparola opened 9 years ago

philiparola commented 9 years ago

Thinking back to when @jameysharp wrote/demonstrated the code, it appears that the Kalman filter filters out the wrong data variance. We noticed that the line largely remains flat, with peaks that (might) coincide with the peaks on the unfiltered data. Perhaps we accidentally reversed the filter data at some point in code?

jameysharp commented 9 years ago

More context, in case anyone looking at this wasn't at the meeting last night: I coded up an example Kalman filter from the Wikipedia article on the topic in front of a live audience, and it went as such demos go: the filter didn't work.

@philiparola, can you find a test that would prove or disprove your hypothesis? I'm afraid I don't quite understand what you have in mind, but I'm delighted that you're thinking about it!

I've found the bug and will present the fix at our next study session, on the second Tuesday in March (two weeks away). Meanwhile I think discovering the fix yourself is a fine exercise and I encourage interested folks to try it.

For a hint: I thought for a minute that we might have found an error in the Wikipedia example, but I was wrong about that.