Good Loitering Analytics?

Steve, I am not sure if that is any loitering analytics can be 'goo'. The general issue is that no video analytics can differentiate between an object that a user has simply temporarily set down (as they are helping a friend or family member, resting, reading something, talking on the phone, etc.) or someone has purposefully left an item. Because of that, every loitering analytic we have seen comes with a time delay (i.e., wait 30 seconds, 2 minutes, etc.) before triggering. Make the delay short and you'll quickly get flooded with false alerts. Make it long and those decrease, but you might miss the (rare) real incident.

That said, we have never formally tested loitering analytics because most industry people seem to have little confidence in any of them. I am sure something out there must be better than average, I just do not have enough confidence that any will work very well.

Steve, does that make sense? How good do you need it to be?

Here is a little something to help you determine what is good. Find out how the analytic classifies an object as human or vehicle if you are looking for one of those. How much of a human has to be seen to be considered human (think wheelchair) and how many objects it can track at any one time (think crowd or group and some can only track 20 or less) and how it manages items that disappear and reappear (think poles or walls in scene) and if all of those are met when working with your desired scene and camera placement, you will have a good analytic!