In Internal Testing
What it does:
Determines if a highly similar event has been seen before by survail by comparing similarity of the image. Unlike our vehicle or person Re-Identification models, this model also compares the surrounding area to see if the background has changed.
Groups similar events and allows you to suppress alerts that are highly similar to past events.
Why it is Needed, the Parked Car Problem
A parked car does not suddenly stop becoming a vehicle, so a computer vision based system that is looking for a vehicle will find a parked one just as easily as a moving one. There's still a use case to make sure that you continue to record a parked car: someone might vandalize it. It just isn't very useful to get an alert each time that the software creates a new clip or each time that an object (like another car) passes between the camera and the parked vehicle (causing it to disappear and then come right back).
Why it is Needed, the Man-at-a-Desk Problem
In the same way, repetitive events are also common in office environments where someone may be seated at a desk for hours at a time. You probably still want to record this footage to calculate if this person is productive and determine that nothing is happening that would endanger their safety, but you may not want to get alerts about this person frequently.