International Security Journal hears exclusively from Sean Grimm, US Country Manager at Ipsotek about the future of security control rooms.

Millions of CCTV cameras record thousands of hours of footage every single day, yet amazingly most of that data still goes unused.

If businesses are to move beyond traditional analytics from their control room setup and unlock real-time insights, we need to see a shift in mindset.

The evolution from passive video monitoring to intelligent, computer vision requires more than just a technical upgrade.

However, what we’re seeing in the market at present is that while everybody wants AI, very few know how exactly best to deploy or use it.

Challenges with legacy offerings
A lot of this is down to the impact of legacy technology that historically hasn’t been accurate enough to offer detailed alerting for control room operators.

There would typically be a lot of false alarms generated, creating ‘alarm fatigue’. After all, if you’re getting 100 false alarms a day, you’re probably going to turn off the alerting.

On top of this, for a long-time seamless integration of infrastructure wasn’t available, meaning traditional video analytics solutions didn’t easily integrate into a platform that operators were already comfortable using.

This has created a lot of pushback and reticence around adoption.

However, we’re now seeing AI-powered models get better and better, particularly generative AI models for large language learning or natural language detection.

We’re able to get something that is highly attuned and with the integration capabilities that mean it can fit natively into a platform already being used.

The shift from motion detection to scenario logic
Traditional video analytics was based off digitising of the analogue camera.

When that capability became available in the late 1990s and early 2000s, it was quite revolutionary to have cameras that could actually provide insights, but looking back it was of course very rudimentary.

You didn’t have a lot of pixels and clearly there were no AI models. It was all literally just motion detection.

Today, AI-powered models, can quantify just about anything that arises from a pixel change perspective.

We’re now at the point where with scenario based logic and pre-trained AI models, it’s possible to layer 32 AI models on one camera and create a highly customisable alert trigger.

Add to that generative AI where we can ask questions or add a prompt to get contextual insight on events that could otherwise be easily missed.

That’s opening up a huge number of opportunities for driving value.

Large language models
We’re also starting to see far more utilisation of large language models that have been trained on billions of images in security control rooms.

This has significantly improved security capability in areas such as identifying potential risks and reducing false alarms.

Previously, AI models used for activity recognition in security footage would struggle to differentiate between normal operational behaviour and unusual activity. For instance, a delivery associate restocking shelves after hours or a maintenance worker moving equipment through a back corridor might have triggered unnecessary alerts in older systems.

The large language models available today is starting to show value in interpreting these contexts far more accurately, understanding intent and environment to reduce false positives and improve overall situational awareness.

Tackling cultural biases
There is still an element of caution or fear even, about AI in general and this is impacting things from a control room perspective.

There is a concern over AI being able to take over responsibility for generating customised alerts and rendering human roles redundant.

However, that really isn’t the reality from implementations we’ve seen. What AI can do is the mundane tasks and quickly.

It’s not going to make a conscious decision to call the police or call an ambulance when a security alert is triggered.

That responsibility and decision still lies with the operator; what AI is doing is enhancing the overall capability of the control room.

Aside from caution and fear around AI, there’s also a degree of unreasonable or unrealistic expectations about what it can do.

It’s not a magic wand; everything we get from an AI perspective has to have had some form of human training, human intervention or human creation intertwined with it.

The future outlook
We’re going to get to a point where AI gets stronger and ultimately generates a lot less false alarms.

Think back to the mid-2000s when the first cars came onto the market that would automatically switch the engine off when you were stationary at a traffic light or in slow-moving traffic.

When you put your foot back on the pedal to move again it took a second or two to kick in. But clearly it’s no longer like that because the technology evolved to the point that it’s instantaneous.

That’s where things will ultimately head with AI too. It’s going to get to a point where it’s seamless and you don’t even realise that it’s working so effectively in the background. That’s the future.

As algorithms become nimbler, we will learn and train things on larger frameworks and skill sets.

Eventually, this is likely to drive a shift in the market where organisations will take what’s best from a traditional monitoring system and upgrade their approach to adopt AI-powered solutions for smarter decisions and safer environments.

Once businesses start getting wiser to what data can truly do (e.g. in a retail setting – what time of day people most frequently come into the store, how people navigate their way around the store etc) we’re going to see increased adoption of AI analytics in control rooms.

Data scientists will also become more prolific in businesses with increasing demand for generating data led insights from video footage.

The wider ecosystem also has a big role to play in helping to drive AI adoption. Many of these platforms are already doing so thanks to the hardware they’re bringing to market and the speed they can offer.

In turn, customers are starting to ask how they can work smarter and that represents a shift from security personnel who would typically advocate for video analytics, to infrastructure personnel who are telling them they’ve already invested in this architecture.

Ultimately, this is what will increase adoption as things become more business driven at the C-suite than at the security level.

Originally published in International Security Journal