Page 46 - SyI Quarterly - Q3 and Q4 Edition 2023

Our Membership

Effective artificial intelligence implementation in the

operational security environment (Part 1 of 2)

Martin Lynch MSyI

Our systems operate as standalone without access to the IOT, save the agreed tunnel connection to the
supplier. The systems do have the ability for key members of the team to log in via an encrypted two-stage
verification and authentication protocol for out-of-hours monitoring if needed. The benefits of this type of
access speak for themselves. With CCTV we can monitor alarm triggers, review footage in real time, and
review the historical. Overall, a great time saver and a boon when the alarm goes off for a non-urgent item
but not much different from your more common non-AI-enabled CCTV. The system runs approx. 200+
cameras with 30% being PTZ. Each of these has features such as virtual trip wire alarms, loitering
detection, gait detection, and mask recognition, these you may be familiar with, so I will not labour. What
will be new to most of you is the machine learning and AI monitoring of the system. In part 2, I will discuss
the ingress system.

Let’s define AI

I have no doubt that some confusion exists as to what AI is. Firstly, as to how intelligent it is I would ask
you to read an article by David Strudwick MSyI from The Security Institute AI+NEST Special Interest Group
whose writings I would agree with. Artificial Intelligence is a placeholder for a great many things such as
Computer vision, machine learning, and neural networks. In terms of intelligence none of these nor large
language models, chat GPT for example are any more intelligent than a programmable dishwasher. The
extant systems in CCTV for example use large data sets to train the algorithms that look for patterns or
anomalies. Algorithms can be described as recipes, akin to making a pizza, ingredients go together, and
they are processed in an oven resulting in a cooked and tasty margarita. They are simple tools, however,
these tools implemented with good understanding and
pedagogy are powerful allies for the security industry.

What is AI in CCTV?

In simple terms, AI in CCTV works by sending data to a recorder and processing it via an AI layer to make
sense of the video feed. AI utilises software programs that analyse the images from cameras to recognise
humans, vehicles, or objects that display out-of-pattern behaviour. Rule-based AI cameras are manually
set up with axioms and reference images such as humans in different poses, angles, or movements. The
AI layer is developed from large data sets, for which you may not be able to readily identify the source,
indeed this may not exactly fit your environment so work will have to be done. The equipment may well
be set up for you as part of the package, however, it will need to be localized, in my version this is called
teaching by example and took a great deal of time. I will discuss this along with considerations and
limitations in part 2.

Part 2. Will be available in the Quarter 1 2024 Edition of the Institute Quarterly Magazine

So, the board wants you to investigate AI or a part thereof, or you want to do the same as your position
demands, you get the picture. In preparation, I would direct you to review the mail and sales brochures
you may have received. I am not going to be critical of the articles, they explain what is available and what
the equipment can do. Let’s look at what these do not say.

46