How to segment data (clustering) of motions in video?

Question

Clustering

There is a video analysis algorithm using the OpenCV library.
It looks for movement in the frame and records their amplitude.
The algorithm produces data . It is

necessary to segment the data into large sections, for example, as follows

. At the moment, we use the
dobigeon.perso.enseeiht.fr/app_AR_seg.html
algorithm, but it is sharpened more for sound waves, and large fragmentation - finds from 9-20 segments (green lines where now cuts indicated), and in this example, 3 large ones are needed.
How to analyze this raw data, what algorithms are there ...
Here is the data https://yadi.sk/d/OYup97BxaG26G
Perhaps you can just smooth the graph ... how to build a smooth average, and slip it into our segmentation algorithm. But what is the best way to implement it.

Answer 1

It depends on what you want to get, from what material, and of course you need to look at how you consider the "amplitude of motion". For cinema it will be one thing, for sports it will be another, and clips or home videos will be quite a third.
In general, one usually starts by extracting individual frames (not single frames) and extracts more related data as the slices are decomposed.