Project Logs

Collapse

• Run, Forest, Run...

  davidmpride • 04/29/2026 at 22:04 • 0 comments

  Well - the Qwatch pedometer thought I was running - even if taking a gentle stroll. Was great for the ego, but not for accuracy. So I spent time this week looking at the exact data generated by the IMU on the ESP-32S3 board. 
  

  I ended up building this rig to test it over several hours, without having to walk several kilometres!  

  The  IMU gives you acceleration on three axes, because the watch can be at any wrist angle, you do not want to look at just one axis. Instead, you collapse the three axes into a single acceleration magnitude.
  

  That magnitude includes gravity, so when the watch is mostly still it will sit around $1g$. The useful “movement” signal is therefore the deviation from gravity.

  The pedometer then looks for repeated pulses in that motion signal. A candidate step is a local peak where the movement rises above a threshold, then falls again, with sensible timing between steps. 

  Then you add timing constraints so random wrist flicks do not count as steps. This was the part that needed the fine-tuning. 

  In plain English: calculate overall acceleration, subtract the constant pull of gravity, smooth the result, then count only clean peaks that are big enough and spaced like human walking. For Qwatch, the important practical point was that this runs continuously from the IMU even when the display is in “fake sleep”, otherwise the pedometer stops counting during normal wear, this means following the re-working of the step algorithm itself it works sweetly but is fairly battery intensive. 

View project log