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)