How does your facial recognition system work?
The way our system works is with the existing camera infrastructure. At most casinos, they will have hundreds or thousands of cameras, of which they’ll typically use us for a small percentage of – mostly when people are entering the casino and other main areas. Our solution is based on AI; but more technically, there are neural networks that are running and what we do is analyse every frame of every video as it’s happening in real time. We’re not looking at everyone, essentially what we have is a very limited watchlist – which starts at zero. Then the casino builds the watchlist themselves, in which people are placed in different groups. You might have one of self-excluders, known felons, advantage players, aggressive panhandlers, prostitutes – or even VIPs on the other side of the list, it’s not all bad guys. What we’re doing in real time is looking at the video to say, ‘is there a face in the frame?’ If so, we’re looking for distinguishing features – what we call landmark features – such as the eyes and nose, and turning that into a digital vector of the face.
Keep in mind, it isn’t the same as going through immigration at an airport where you will be looking directly at a camera; we’re dealing with a situation in which most of the cameras are high up and far away. The person could be entering with a group of people, obscured from the camera’s view; so, as we analyse that frame, we compare that image to all the images we have saved in the watchlist – which all happens in less than a second. Then, if we find a match, we can alert security – or hospitality, if it’s a VIP – that this person has just entered the casino; we're now armed with who it is and where in the casino they are.
What added value does this bring to the casino?
What’s interesting is that, in some cases, like for an advantage player, I always think security would have escorted them out of the building – but they don’t. It’s not illegal to be a card counter, so they’ll often let them play games, slots and as soon as they’re at a blackjack table, the dealer knows what to watch out for. However, the actual value is the real-time notification that this person is in the casino. When you go to a video surveillance room in one of these larger casinos, you’ll literally hear an audible alarm – at which point everybody springs into action. The value of that is when you remember, historically, these places would have 1,000-2,000 people on the watchlist. How could a security guard possibly remember the faces of all those people? It’s virtually impossible, right? And so that’s why when casinos deploy our technology, and many on the Vegas strip have used it, it’s becoming more typical. We’re providing real-time alerts and the number of detections is going up 300-400% because you’re automating the whole process. The VIP example in London’s Les Ambassadors Club is going to become typical because your high rollers represent 0.5-1% of all the players but they contribute 60-80% of the revenue. You need to spot them quickly, so casinos like that are getting very smart about identifying them and making sure they get the white glove treatment.
Are there limits to the amount of people the system can begin to recognise?
There are. This is why we take a lot of care when we sign up a casino, it’s not a plug and play; we need to understand where the cameras are. Sometimes the cameras are very, very high, so if you’re dealing with one that isn’t of great quality with a high turnover of people, and all you’re seeing is the top of somebody’s head – well, our solution isn’t going to work very well. This is where we say we need to reposition the cameras. Or, in some cases, we need a higher quality of camera, meaning that even if you’re far away you can get a crisp image when you zoom in. But for the most part, casinos have cameras down at a good level; we don’t need to tell them to have higher resolution cameras – they already know that. Our ability gets better over time because we’re fine-tuning it, figuring out the balance of false positives and false negatives because you can over-tune it; so, this is where you work with the casino to find the right balance.
The other important part here is that not all solutions are the same. What I mean by that is we train our algorithms to work expressly for these kinds of conditions, real-world conditions where the person isn’t looking directly at the camera, they’re far away; we might only have a profile, they could be wearing a mask or sunglasses – because in many cases people know they’re not allowed inside – though our solution works well even under those circumstances. Your ability to detect those kinds of people under those kinds of conditions is only as good as your underlying AI algorithm. And, when you train your algorithms to work under those kinds of environments, they’re naturally going to perform better. You see facial recognition is a whole slew of technologies, especially coming from places like airports, where you look at a sensor which can recognise you from your passport photo at a 1/1 ratio. What we’re trying to do is make that on a grander scale and in a busier environment.
Are you aiming for expansion in other markets?
Absolutely. We’re obviously in Europe, and we’re expanding in Asia-Pacific – while Macau has its own set of challenges there, as well. I think even within the US we’re looking to expand, because most of our footprint is in the larger Vegas casinos. However, right now we’re spreading to Native Indian casinos. I was absolutely shocked because one of our customers is out of Tulsa, so I went and visited them a couple of weeks back and was surprised at how large those casinos are. They’re every bit as large, in terms of square feet, as a Vegas casino – and they have the same amenities. They are the ones who are much more bullish about talking about facial recognition, because they know they have the law behind them, and at the heart of everything they are doing is the customer experience.
That didn’t always used to be the case. It used to be more of a security play, but what we’re finding now is there are two groups within a casino and they each have very different budget levels. First, there’s your security team, then there are the people that are supposed to be looking after the whales. Those people looking after the whales have much bigger budgets because they’re generating 60-80% of the revenues, so they want to identify those guys much earlier on in the process.
With setting up this system in the casinos and reading all of the data in real time, how stringent are you on data protection?
It’s a great point, especially with GDPR. You might think a lot of the data is traversing from the casino to our central servers, or to our cloud servers – that’s not the case. Everything is done locally; traditionally in a Vegas casino, all the servers are sitting on the premises and they are running our algorithms themselves, so we’re not getting that data on our servers. Now, naturally any data that goes from the camera to servers still needs to be encrypted; and it is, both in transit and at rest, but it isn’t going anywhere on our servers.
The other thing is privacy, which is a big term. It encompasses not only how the data is handled but where it's managed. It even has to do with how long someone is on a watchlist for, right? So, if somebody is a criminal, are they a criminal forever? For that reason, we have very, very granular controls that allow casinos to set that. The other thing we can do, which is specifically relevant in Europe, is we have the ability to blur people not on the watchlist on video playback – that’s primarily for GDPR purposes. If you’re an innocent person, that means when the people are playing the footage back using our solution, they can blur out all the good actors and only see the people that are on the watchlist.
The casino still has the raw video footage from all the cameras, which they need to keep for compliance purposes – so they are taking the right due diligence – but the retention of the footage is up to them. Facial recognition is nuanced, and when used ethically it can help solve a lot of problems like catching bad actors, while simultaneously really improving the whole player experience. This is the reason most of our customers are deploying our system.