• Hi there @Gordon, I was wondering what the release plan / schedule is for the 2.9 firmware and the Oxford Pedometer. Steps are basically broken, even after tweaking the ActivePedom settings its not very accurate. I use a sensivity level of 90.

  • Hi,

    Right now 2v09 firmware is available - it was released last week. I don't have a specific timeframe for the Oxford pedometer though.

    I was really hoping the Oxford pedometer would be the solution to all our step counting problems, but I'm not convinced it is. The issue was that it was trained based on recorded data of people walking - not of standing still, typing or doing other tasks.

    So the oxford pedometer code is arguably just as inaccurate with the contributed pedometer data (which contains a mix of activities) as the original Bangle.js one, plus it randomly records loads of steps even when the watch is sitting un-worn on the desk (it seems this can be fixed by stopping the automatic threshold getting too low though) and has quite a large extra resource usage.

    I think the underlying issue with the pedometers is actually that they need to be able to detect a sequence of repetitive steps, rather than just an 'instantaneous' step that we do right now (eg https://blog.st.com/pedometer-mems-step-­count/).

    So, I have been working on making a new step counter, using the contributed data to test on - but I don't know when that'll be ready. If others are interested in developing something or looking at what I've done I'm happy to upload it so they can have a play.

    No embedded development experience needed - right now it's just a single C file that you run on your PC.

    If there's any interest I'd be happy to have some kind of competition/prize for the most accurate step counter that's efficient enough to run on Bangle.js?

  • Actually scratch that - I just had another go - using just a bandpass filter limited to acceptable step speeds I'd googled online - and it's doing a much better job with the step counts.

    RAM usage is now ~60 bytes vs the 16 before (or 1400 for the oxford counter). I've just committed it to master, so if you try one of the latest builds it should work. All you need is to use a normal pedometer - not the active one.

    The filter does a pretty good job of detecting repetitive steps all by itself now.

    The Oxford step count folks deserve a lot of credit for this though. While I didn't use any of their code, I did use the filter designer website they had found, as well as the idea of brute-forcing through pre-recorded data to get the best values.

    Only thing to note is we still don't have a huge amount of data that includes washing dishes, typing, etc - so probably the coefficients could be better if we had a bunch of that too.

    1 Attachment

    • Screenshot at 2021-04-29 11-44-12.png
  • having data for typing, staying at desk is easy for me :) I can generate bunch of them.

    i'm upgrading to v2.09 and use another app than activepedom is enought to test the new algorithm?

  • i'm upgrading to v2.09 and use another app than activepedom is enought to test the new algorithm?

    Yes, absolutely! Just the normal HRM widget is fine: https://banglejs.com/apps/#widpedom

    Carried away by my step counting success, I also applied the same ideas to the heart rate calculations. The form of data reported is a bit different so some apps may not work (eg the heart rate app needs tweaking) but the heart rate widget https://banglejs.com/apps/#widhrm should be a good way to test and will hopefully provide faster, more accurate data as well.

  • Great news ! I'll grab the latest build of 2.09. Is there a release note for whats in 2.09, looks like I missed the announcement. I'll give this a go over the weekend. I too sit at a desk all day typing, but do walk the dog. I might try wearing my Amizfit Bip on the same arm and compare. If its within 2% of a Fitbit or Amizfit Bip it will be good enough as far as I am concerned.

  • Its bad news, I'm afraid - I did 842 steps over a 3 hour period sat on the sofa with chromebook on my lap developing a new slot in app for kitchen combo. Definitely using 2.09.7 and widpedom (not activepedom). I noticed I can do 5 steps by twisting my wrist.

  • I'm at 4600 steps, I did wear the bangle all night, but it's still only 8am, AFAIK I'm not sleepwalking, only had a breakfast and did some washing up.
    Noticed the same 5-7 steps increase as HughB when twisitng my arm. Also when I twist my arm, and stop, and the counter goes up 2-3-4 steps while not moving at all (the watch is as stationary as I can hold it).
    Installed 2v09.7 and the pedometer widget after Gordon's #5 post.

  • Is there a release note for whats in 2.09

    Yep - http://forum.espruino.com/conversations/­362761

    The IDE should now prompt you to update, but the BangleApps loader doesn't (yet). I'd like to make sure nothing is too broken first :)

    Its bad news, I'm afraid

    Ok, thanks. Looks like I'll have to tweak the minimum threshold for it.

    If you're able to upload some acceleration logs (like last time) with some typing at a computer or other tasks that would be really handy (just let me know how many steps are expected for each one!) - because right now I try and get something usable but then I calibrate the values based on the pre-recorded data, and that doesn't include much in the way of 'normal' tasks.

  • Understood. I will have to remind myself what to do and download the app again.
    I guess the idea is to train the model to ignore small arm movements when stood or sat still.

  • Post a reply
    • Bold
    • Italics
    • Link
    • Image
    • List
    • Quote
    • code
    • Preview

Wondering what the release plan is for Firmware 2.9 and the Oxford Pedometer

Posted by Avatar for HughB @HughB