Completed

:speech_balloon:

Identify GPS noise

I’m being asked to confirm a lot of items in the middle of the night, when my phone was on a nightstand. I assume it’s because the GPS fix is not stable. Is there some way (accelerometer?) to tell that this was just noise?

1 vote

Tagged as Suggestion

Suggested 11 July 2019 by user Aneel

Moved into Completed 12 July 2019

  • Sign in to comment. Sign in by email
  • 11 July 2019 Aneel suggested this task

  • avatar

    LocoKit already noise filtering, on multiple levels.

    First is the Kalman filter(s), which is an algorithm used in aerospace for correcting and filtering unreliable location data inputs. LocoKit uses Kalman filters to intelligently remove the majority of location data noise.

    Second is the dynamic smoothing layer, which adds more or less smoothing to the input data depending on various conditions.

    Third, if you are using Arc 3.0, are the CoordinateTrust ML models, which learn how much “Trust Factor” to apply to incoming stationary location data, based on your corrections and confirmations of stationary samples. So each time you confirm or correct a segment of data to “stationary” type, the CoordinateTrust ML models learn from that, and adjust their Trust Factor for incoming data at that location.

    And lastly, there is the “Bogus location data” feature. Whenever you see location data in the app that is wildly wrong (eg more than 100 metres away from the real location), you can mark that segment as bogus location data. This is then fed into Arc’s activity types ML models, helping Arc to auto detect wildly wrong location data in that region.

    Basically Arc is already doing a large amount of mathematical / algorithmic work to clean up the raw data. Arc is leading the pack, in terms of location recording apps, when it comes to dealing with unreliable and inaccurate location data. I’m not sure that any other apps out there even use Kalman filters, let alone applying multiple layers of ML.

    So in conclusion: The best thing to do when you see bad location data in Arc is to tell Arc that it is incorrect. If the data should have been stationary, and is within about 100 metres of the real location, then correct it to stationary type. If the data is more than about 100 metres away from the real location, then mark it as “Bogus location data”.

    I hope that helps to explain!

    Thanks 😄 Matt

    12 July 2019
  • 12 July 2019 Matt Greenfield moved this task into Completed

  • avatar

    I’m sure you’re very proud if the fancy processing ARC does. Unfortunately sometimes the result is just not there. It’s time to face the reality. It’s completely irresponsible to ignore the fact that iPhones often give shitty location data while static indoors. Arc should have an editing tool to merge all the garbage between two timestamps without manually editing out every single junk.

    27 July 2020