How to identify changes of objects

[robmarkcole]

Object detection models return an object count, but what we actually care about is new objects - we want to ignore background. How could the integration do this? Ideas:

• learn background objects (static) and exclude them from detections?
• assisgn background objects an ID then ignore them, doe this by comparing adjacent images

[MTrab]

would it be possible to look at detection bounding-box?

I thought I could use this for detecting when a car arrives/departs from my garage. It works - somewhat, but when a car is parked in the garage it get detected (thus triggering image_processing.object_detected event every time) as long as it is present in the garage.

[robmarkcole]

This would really need to be implemented by deepstack or a standalone object tracking integration