The engineers need to research into machine learning image comparisons. There are many open source solutions to start from.

The basic idea is you use your previous set of images to compute a model, and then that model can be used very swiftly to calculate just how similar (or different) the current frame is from that baseline model. If its X% different, a notification should be triggered. This X% is what the user would consider "sensitivity".

Here is an article about how to use TensorFlow in python to do this, but ML is available in most popular languages, including C# (ML.NET), Java, etc.
https://towardsdatascience.com/image-similarity-detection-in-action-with-tensorflow-2-0-b8d9a78b2509