This could feel like a repeated question, but please bear with me. In my work we want to compare different methods of subtracting gravity from an accelerometer data to decide on the better way and maximize the accuracy of linear acceleration result we get as much as possible. We have so far seen the following methods listed in different places.
- Using exponential filter because unlike the linear acceleration the gravity stays constant between consecutive samples.
- Rotating gravity norm (taken at a stationary situation) to the device coordinate system by using rotation vector and subtracting from the total acceleration.
What other techniques are there or how can we enhance these already existing ones to improve our results. A related question would be how can one measure if a gravitational subtraction implementation is good enough?