Good Job IPVM Team. John have them add fog machines, strobes and digital media panels for background on the next session to jazz it up a bit and keep us entertained. Could make it into a "Laser" Tag like joint with all that open space, built a few 'obstacles' to hide behind. A opening or two for Brain to use for Access Writes Ups and showcase door hardware, a weakness for most of this crowd.
Wonder how this will work in something like a ATM environment to help with 'Harvesting' Detection.
Harvesting tends to be a bigger issue in non-vestibule/controlled access locations so I am curious how well it performs outdoors in various elements to detect 'loitering' at the ATM itself. To properly combat harvesting in a real time situation you need integration between the ATM back end and the VMS/PSIM. In my case this is a proprietary OS and not something like NDC+ or 91X etc. You need to first detect multiple ATM transactions from the same object so the systems needs to talk to each other.
What is ATM harvesting? I haven't heard that term used before, and a quick google search doesn't show any definitive results.
Overall I have found that ATM loitering detection to detect active fraud becomes difficult because you have some people that take FOREVER to do a couple of basic things, even though there is no fraud involved.
Harvesting is the when a person comes to an ATM to use multiple skimmed accounts that they have made cards and have PINS for. The purpose of this is to drain the accounts for as much as possible in a single transaction. Often times the perpetrators will do over a dozen transactions/withdraws before they are done at that location and move on to another. To prevent Harvesting you need a analytic to monitor the object and you need real-time ATM data to know there are multiple transactions within a given time frame. The goal is to detect the potential harvesting before to many accounts are drained by shutting the ATM down. Yes there are ATM analytics that can detect fraudulent transactions but with combined real time video analytics the combo of the two can help combat the fraud.
As I mentioned they tend to avoid vestibules where access is controlled even with a door w/o access control so outside locations such as drive up ATMs are often the prime target.
It sounds like a better approach would be to determine if the person at the ATM changed between transactions, instead of just detecting for general loitering. We could do this with our face rec component, or maybe even easier with a variant of our Target Tracker re-identification implementation.
This video shows a search to find other instances of a vehicle, but this is just an example of a common use case. We could take ATM data and compare similarity of the user doing the transaction vs. the previous transaction, for example. This can also be done in the cloud if you don't want on-prem hardware, send a short (~10 sec) clip of the person with each new transaction, or something like that. You can also do this across multiple views, so in theory if you have a bunch of ATMs you could do similarity comparisons across all recent transactions in a region to see if the person is bouncing around. Of course, if they are going to ATMs from different unrelated banks, it wouldn't work unless they all pooled together in some kind of anti-fraud coalition.
Just to be clear, this is intended for indoor use:
This sensor is designed for indoor locations such as ATM rooms and other closed indoor vestibules within banks, pharmacies, retail stores, prisons, health care facilities and other environments where privacy and highly accurate detection is needed.
I did test it outside just monitoring a corner of the building and it accurately detected my presence and my dwell time in the area without dropping presence within ~13', similar to indoor performance. The presence detector did get a presence event on some leaves blowing around right under it though.