This page contains a Flash digital edition of a book.

resolution; they are affected by different lighting conditions so they are inconsistent. We need to put the data in a form that means it can be shared meaningfully – so if you are barred from the Empire in central London, and you walk over to a casino in Curzon Street, they can also identify you even though the conditions are very different for Playerbook. To do that we have to dial out the inconsistencies due to CCTV – the angle of incidence to the target, lighting, general casino layout and how people flow through and pass the CCTV system – they’re all different in every casino. A reliable capture in one casino couldn’t really be used on the watchlist for another casino, because you’ll never be able to recreate the lighting in that original picture. That’s why we say this is a non-CCTV-based system, it’s working in infra- red so it’s light immune. It works as well in a bright sunny environment as it does in a darker casino floor.

CI: What is the fail rate of the system compared to humans? GJ: The accepted ‘fail rate’ of humans according to

a 2006 study is interesting. If we have 10 images of a person and that person is actually in the room with us, we are incorrect in matching an image to that person 25 per cent of the time. We’re incorrect one in five times when we only have two photographs[Editor’s note – that’s a 20 per cent fail rate with only two images, and 25 per cent with 10 images. People are strange]. The University of Glasgow followed that up with a study where the error rate is one in four when we try to match across different races – so if you have a staff of Western Europeans trying to identify Asian players, we’re wrong one in four times and vice versa.

CI: So how accurate is Playerbook? It surely cannot be 100 per cent accurate?

GJ: We’re not saying that it is. The first proper

comparative study we did that involved any ground truth analysis was at the Sportsman Casino in Marble Arch in 2010. What we found was the accuracy of the Playerbook system was almost identical to humans. We picked a study group of 100 customers at the casino, all of whom smoked – thanks to the smoking ban in the UK, they went in and out of the door more frequently than other customers. We asked the receptionist that every time she saw one of the customers from this watch list she mark it down on a click sheet and we collated the information at the end of the working day. What we discovered was 137 detections on the watch list, and out of those 137 detections we had 136 marked by reception – so nominally, the system was up by one. But then we found the errors both reception and the Playerbook system had made, and they were almost exactly matched. We had 36 detections by the receptionist that were not seen by the sensor, and 37 detections made by Playerbook that were not seen by the receptionist.

CI: So the strength of this could be in combining this with human detection?

GJ: In an ideal world, perhaps. But with budgets declining in many organisations, this is about taking as many human beings out of an equation as possible.

Because the Sportsman trial was for ground truth

analysis, we had a real-time CCTV system recording in reception for the duration, so we could go back to each instance the Playerbook system failed to spot an individual, and work out why. The main reason, it turned out, was people walking into the building behind other people. What that indicated to us was that we needed to move the sensor, so we could separate the faces better as they move past the sensor.

CI: So now you know the optimum configuration for success?

GJ: We redeployed it at the Empire and put it at

the bottom of a staircase, so when anyone came down the stairs, they got nicely separated. We could then pick them off much more accurately. During that trial, because there is no full-time receptionist at the Empire as they have an open-door policy, we used the staff as the basis for the watchlist. In the two week trial, there were nearly 109,000 activations on the staircase leading to the gaming floor. From that, we had a watchlist that was only populated by 58 people. On every single day bar one, when the staircase was roped off because of maintenance work, we were getting positive identifications. We had no false identifications during that trial period.

CI: Presumably there’s the means to integrate this with a player tracking system or backoffice?

GJ: That’s exactly where we’re going with this at the

VIP level. We’re already in technical discussions with one casino’s back office provider because this has great potential in terms of identifying high rollers or consistent customers. It’s also great for people counting, which is incredibly useful in a casino with an open-door policy. NOVEMBER 2011 49

Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48  |  Page 49  |  Page 50  |  Page 51  |  Page 52  |  Page 53  |  Page 54  |  Page 55  |  Page 56  |  Page 57  |  Page 58  |  Page 59  |  Page 60  |  Page 61  |  Page 62  |  Page 63  |  Page 64  |  Page 65  |  Page 66