Again @jremington thanks for your time & assistance.
As i originally stated, i have read some papers that propose possible solutions but not exactly what i'm hoping to achieve.
I have thought for example that ,(after reading the paper `A More Reliable Step Counter using Built-in Accelerometer in Smartphone & following the algorithm there,) if someone does not stand still and thus(does not get static values) , and if someone is not walking or doing similar activities that produce similar spikes on the serial plotter then he must be exercising . Although i know the principles, code-wise i cannot find a way to make that happen.
Where abs(a) = sqrt(x*x + y*y + z*z)
*Where abs(a) is average acceleration value; X,Y and Z are three axis values. *
The final stage is a key to complete this proposed model, providing the accurate steps. We use a new reliable technique based method of peak. In frequency data of the average acceleration signals, the method finds the peaks points, which is greater than the other points. Then the method checks the number of the occurrence point between two peaks. In which, the number of occurrence points must be greater than two. If the peak point satisfies this condition, it is identified as a one-step. All step points are picked according to the principle rule of the proposed method. The step counting is subsequently taken over one-step to another until reaching the stopping criterion.