I would expect the initial motion to be more consistently in the same direction opposite the flight path of the object (rather than the very inconsistent evidence I showed above) if the bumps were caused by sudden decelerations the object made before starting to rotate
also:
• sudden deceleration should be visible against the cloud background, but it's not
• sudden deceleration should at first be compensated by the fine tracking and not cause a bump
Here I posted the source code for the current implementation of my algorithm for extracting the geometric properties of the blob. In that repo I've included the latest output (graphs/images/tables/logs) of the notebook. I have a separate repo here where I pushed just the code to help track the history of the changes. It also includes additional utilities that the notebook downloads when it starts.
I ported the code to Python to integrate better with the existing data science infrastructure. The notebook has links at the top to open it in Google Colab or Kaggle, both of which allow you to run the code online for free. You can also download the notebook and run it with e.g VS Code, but that requires some setup and is less tested at the moment. It uses HuggingFace to store the input/output data, downloading the video from here and uploading results here.
In this implementation I'm using gamma curve fitting instead of histogram matching to convert the white hot frames to black hot.
The same notebook also runs Sitrec as a console application (support for which I helped add to Sitrec recently) to get the calculated human horizon angles, which are used for further processing in other notebooks.
