How to calculate the level of human activity according to the readings of the accelerometer on the body?

Question

Accelerometer

We have a three-axis accelerometer recording 100 readings per second. You need to understand whether a person is moving or not.
Now I first find the derivative of the signal as (F(t) - F(t-1))/0.01. Then I break the entire signal into segments of 1s.
Then I subtract the average from the values ​​of the derivative, square them and sum them up. I believe that I get the signal energy for each segment in this way. Logically, the greater the energy of the signal, the higher the activity.
I do this for each axis of the accelerometer and add the resulting values. And then I go through the moving average filter to smooth out the values.
And the problem is that I can’t always determine from the received values ​​whether there is activity or not. Sometimes when there is no activity, the energy is high, and when there is high activity, it is low. Where did I go wrong and how to do it right?
Used this article as a guide www5.informatik.uni-erlangen.de/Forschung/Publikat...