accelerometer behaving very strangely

By urischatzberg , 13 July 2016

Hello All,

I am Somatixs CTO and we are based in Israel/New York.
We are automatically detecting smoking gestures from smartwatch accelerometer and gyroscope signals.

I encountered an obstacle and hoped you could help introduce me to someone in your company that can help.

I put 2 LG urbane smartwatches (with Invensense sensors inside) on my wrist and perfomed smoking gestures. I hoped to get fairly similar signals from both devices but didn't.
In some areas the accelerometer signals are very similar while in other the signals are very different.

attached is a plot of the 2 signals. notice that the x axis is in seconds. Also notice thta at some areas the signals converge while in other they divarge.

Inline image 1

All areas with constant acceleration are areas where i did not move my hand.

I am trying to understand this phenomenon. Usually there is a fairly constant bias for each signal. In this case the bias changes very quickly and with strict correlation to the signal.

Any help will be greatly appreciated.

phpbb Topic ID
34776