I would like to share with the community a presentation I made, demonstrating the use of the Wolfram Language image processing and machine learning capabilities on the reading of an analogical field instrument.
I’m just posting a small part of the presentation, but you can find the whole notebook, plus some pre-processed files (~80 mb), at the end of the post.
Starting with the film footage of a pressure gauge, the presentation guides the audience through the dial value acquisition process (image processing → machine learning → plotting the results).
The following film is used for the demonstration (adapted from: https://www.youtube.com/watch?v=Kl2pa7pmES4):
After some ImageAlign, ImageTrim and masking (details in the notebook), we end up with a stable image:
I then proceed to the extraction of the dial (removing the background):
and profit to show a nice 3D time evolution image of the dial position:
At this stage, a set of training images is prepared (similar to what is presented here, but based on the manipulation of the extracted dial of the real film footage), and a prediction function is trained:
A nice output panel is created at the end, showing the process steps’ results:
I invite you to download the presentation, where you will find many more details (including the entire code and other surprises 
I hope you enjoy,
Pedro
Presentation notebook plus pre-processed files (~80 mb)
Just the presentation notebook (~0.6 mb)
