Abandoned Object In Crowded Scene - Who Can Provide Accurate Alarm?

If someone tells you they can accurately detect bag left behind with low false alarm rate in a busy scene, laugh at them and trust someone else. Bombers usually use trash cans anyway.

Fredik, good question. The key parameter I have seen used to reduce false alarms when analyzing abandoned object in crowded scenes is dwell / delay time.

For example, take a train station or an airport. If I put my bag or suitcase down, am I dropping off a bomb or just resting? That's one reason why the delay time can be set very long. Otherwise, you will be responding to nuisance alerts all day long (until you shut it off).

I have not heard anyone recently make any claims about improvements in this area. I would be curious to hear what others have seen.

I’m pretty familiar with the concept and I would say the issue is more environmental and operational than technology. 

I will agree in a static environment almost any analytic manufacturer can do object left behind and object removed.

In a crowded environment the issues revolve around timing as John said and also how many objects the analytic can identify and track in a single scene.  Some are limited to less than 30 while others claim 300 or more.  How busy is the scene if the package is under a bench and hundreds are waiting to catch a train after a ball game?

Now let’s talk the hard part, operations:

Lets say someone wants to explode a device at the end of a marathon race.  There are hundreds of people milling about and various chairs, backpacks, vendor carts and temporary trash cans around.  

Your security team gets an alert that something looking like a McDonalds large bag is “left behind”.  

What do you do?  

Evacuate the area, call in a bomb squad?  Frighten the community and stop the event?  Review video to see what appears to be a 10 year old leaving it?  Maybe that’s it and you caused a huge unappreciated commotion.

Or, disregard because it appears innocent and when it goes off, share more liability than if you hadn’t been alerted?

What’s the timing?  1 minute too short?  10 minutes too long to track and too long to prevent?

Manage the expectations and operational requirements.

Fredrik,

I think this is an opportunity for you to display your expertise and earn the customers trust by explaining to them why they can't get what they want.

What they are asking for simply does not exist at this time.

It's tough to tell customers this because there are lots of unscrupulous folk who will sell your customer a half-assed solution that they will never be happy with - and will never work as the customer wants it to.

There are numerous strings here on IPVM that discuss the technical and operational weaknesses of object left behind analytics.

Manufacturer President: "Customer Is Now Very Angry"

Boston Bombings - Outrageous Manufacturer Claims

City Video Surveillance Guide

From the City Surveillance Guide:  Key Findings #9: 

Analytics: LPR/ALPR analytics are the most common types used. 'Video' analytics were uncommonly used, e.g., loitering, face recognition, object left behind, or direction of travel analytics. The key barrier to wider adoption cited was inadequate performance given the cost.