How to use the sliding window method in signal processing?

Question

Signal processing

I'm trying to determine the presence and absence of swimming according to the readings of the accelerometer on the swimmer's head. I do this: I run a sliding window (length 1s, offset 0.2s) by values, find the energy of each window (and summarize the energy of each axis), build a graph. On the graph, high peaks indicate strokes, low peaks are all sorts of small head movements in the absence of strokes. That's just a problem: the peaks are shifted. On the video, I see the stroke at the same time, and on the chart a couple of seconds later. Already double-checked everything, I somehow use the method incorrectly.
What should I do to get the position of these peaks to match the stroke moment?

Answer 1

You are integrating a window, so the maximum energy will be shifted in time.
Now your signal is likely to have a sawtooth shape, when it rises sharply, and then slowly decreases, respectively, in any case, the delay between the stroke and the maximum will be 1s (window width) + add the inertia of the sensor itself (I don’t know the truth about this, what is the order delays).
That. the wider the window you make, the more latency you will get.
Maybe it's worth making a window of 0.1 seconds with an offset of 0.02?