Zero out baseline in accelerometer RMS data
I am trying to analyze accelerometer data (minute by minute logged RMS data) from the MetamotionR product. This device was worn as a pedometer. I am hoping to classify the data (i.e. sedentary, walking, running).
However, over time it looks as though the accelerometer RMS data has a migrating baseline. For example, when the pedometer is at rest, sometimes the RMS values are stable at ~0.6, but the next day at rest RMS values will stabilize at ~0.4 (note total RMS range might be 0.4 to 0.9). Likely this baseline is also migrating while the person is in motion as well, but it is harder to tell. I also don't understand why the RMS reading isn't at 0 while the device is stationary.
Do you have recommendations on how to zero out a baseline for all data for this device to better classify the data into unique activities?