There are many misconceptions about machine learning and artificial intelligence (AI). In all the hype around emerging technologies, what is often lost is a clear understanding of the practical steps needed to make it a reality; in the case of AI, it is the amount and quality of data needed to train smart systems before they are ready to be deployed for real-time decision making.
While this can lead to confusion about the capabilities of AI, it can also result in lost opportunities. Many people overestimate the ability of machine learning or the ease of producing data suitable for training systems, others, on the other hand, are sitting on data resources that could be exploited with relative ease for business intelligence purposes.
Physical security systems, for example, create large volumes of data which is currently often under-managed and under-utilised. Analogue and IP-enabled CCTV systems, for example, are infamous for filling tapes and hard drives which have historically been reused after a short period of time to keep costs manageable.
But this data can be put to other uses beyond preventing or investigating incidents of crime or shrinkage. The growth in adoption of the Internet of Things (IoT) and networked devices, combined with cloud data centres and machine learning technologies creates enormous potential for changing the way we think about physical security.
Two things are driving this change. First and foremost, video data that was previously only kept for limited amounts of time can now be stored almost indefinitely: the cost of long-term data storage drops dramatically when it’s moved to the cloud, enabling collection and retention of information that would have previously been destroyed relatively quickly. Secondly, cloud computing enables organisations to enrich this data with information from other sources and deploy real-time analytics and AI to derive business intelligence.
From security to business intelligence
Typically, physical security systems are deployed to reduce incidents of petty theft, robbery or aggression, and the digitisation of data offers many ways to improve on these core functions. One current application for AI, for example, is teaching security systems to identify suspicious behaviours. Platforms exist which can tell the difference between an individual loitering suspiciously by an ATM machine, and someone who is innocently digging in their bag for a lost card. The more data they can gather about a specific place, the better these systems become at identifying unusual activity.
The same platforms can also help to pre-empt dangerous situations by learning to identify aggressive behaviours against staff. Many surveillance systems in retail spaces are unmonitored and used primarily for forensic investigation after an incident has occurred. A smart (IP) system capable of identifying tension and aggressive behaviours can trigger automated alerts for security teams, or play a pre-recorded warning through the in-store PA; so, a situation can be dealt with as it happens in real time, potentially lessening its impact and pre-empting criminal behaviour.
Beyond using AI to enhance security, however, the same network of devices is also well placed to deliver valuable information about other business activities. Retailers can enhance the customer shopping experience through automated alerts to lengthy queues and putting more staff on check-outs. And combining camera data with marketing platforms, for example, could give store managers a better understanding of how effective in-store adverts are. Is a particular display, banner or video attracting attention, how are people responding to it?
Making intelligence a reality
Historically, this kind of intelligence would have been costly, slow to gather and analyse. Today, these observations can be turned into real-time decision making. Visual analytics can distinguish between men and women, or old and young with a relatively high degree of accuracy. Combing the data from security cameras with digital display advertising, for example, means adverts can be updated with real-time targeting depending on who is in store at that moment. This can be achieved by using systems and devices that most retail spaces have already deployed, delivering new value and return on investment from existing kit.
The potential for these kinds of systems is still only just being fully understood, and early adopters are continually finding new ways to draw on security data to optimise and deliver business value across their organisation. Airports are combining visual analytics with access control to ensure employees really are who their card says they are, office parks are integrating security cameras with building management systems to identify cars as they enter a complex. The potential applications are limitless, and to really cut through the hype around AI and start feeling its benefits, you have to start using it.