I'm working on a project that needs to detect two different objects in a video. The idea is that when one object is on top of the other object, the system will trigger a controller to do something.
I'm using colors for detection, but I've no idea how to trigger the controller when two objects overlap, since then the system will not be able to detect the object underneath.
Does anyone have any idea or experience on this kind of stuff in computer vision? Thanks.
Edit:
https://www.youtube.com/watch?v=nvvaegox7PQ
I've attached a video here. The algorithm/method in the video is exactly what I expected. When objects overlap, vision system know there are two objects there.
The general rule for this type of situation is that people won't disappear all of a sudden. So, if there is a group of pixels showing two people with an overlap, you can assume that they both are there, and calculate the groups of pixels that correspond to different persons. You can get a better idea of this kind of approaches from research papers on human tracking in video. I have seen one paper that use Gaussian mixture models to model people, and that works well.
Of course, these systems can be fooled. If they had a door at the place where the persons overlapped and one person slipped out from the back, the system won't detect that.