Increasingly manufacturers are offering products claimed to deliver Artificial Intelligence, Machine Learning and Deep Learning, but what does this actually mean for installers and integrators? Does the deployment of such solutions make sense in today’s market and will the claimed benefits be viable for end users? Whilst these technologies offer great potential to all security disciplines, the management of expectations is essential. However, the hype over what the technologies could deliver in the future is making this a difficult task.
For many installers and integrators, the developments of Artificial Intelligence, Machine Learning and Deep Learning point to the creation of ever more innovative and bespoke solutions. The potential is limitless, and as processing power increases (as it inevitably will), the potential on offer from these technologies can only increase.
Faster computational speed and resources equates to the ability to run more processes and algorithms simultaneously, which in turn equates to more ‘intelligent’ systems. Well, that’s what the experts tell us!
However, smart systems only make sense – and appeal to customers – if they have a convincing use-case. Without a demand from businesses and organisations, the best technologies will not impact on the security and solutions market.
The technologies in question will predominantly be added into products and systems at the manufacturing level. Installers and integrators are highly unlikely to ever have to work with the coding or processors on offer. The challenge is to understand the benefits and sell those to the end user.
Sadly, simply stating that a system uses Artificial Intelligence or Deep Learning isn’t going to persuade an end user to increase their budget as it means little to them. As with most emerging technologies, the real-world benefits are what will excite the mainstream market-place.
The benefits are what the customer will pay for. However, this does not mean that installers and integrators should not bother about gaining an understanding of how the technologies work. Indeed, a level of knowledge is important to ensure that benefits and features are not oversold.
What’s the difference?
Often the terms Artificial Intelligence, Machine Learning and Deep Learning will be used interchangeably. Whilst the technologies are linked, they are very different. What doesn’t help is that the definitions are not very precise or specific in terms of functionality.
Artifical Intelligence (more commonly referred to as AI) is the overarching technology. Both Machine Learning and Deep Learning are part of the AI landscape. AI seems to have risen to prominence in recent years, but the pursuit of AI has been on-going since the 1950s.
AI is simply the delivery of systems and solutions that can apply intelligence to a system. That might sound vague, because it is! Interestingly, researchers have identified a negative impact of the implementation of AI: people no longer think of the processes as intelligent once it is ‘known’ how the result is achieved.
The AI Effect, as it is known, often results in intelligent actions no longer being considered as smart once they are achieved. This has led to some concluding that AI is best defined as ‘things that haven’t been done yet’!
To take a simple example, consider VMD with discriminations for perspective. At one time, the concept of a system detecting motion and being able to discriminate between a small object in the foreground and a large object in the background, despite the two being the same size on the screen, would have been considered intelligent. When such an approach was first being developed, it was considered as Artificial Intelligence.
Today we understand that such a task is relatively simple to achieve via computational analysis. By applying geometric principles, a system can effectively scale objects based upon their location in the viewed scene, using dimensions entered by the engineer, to ascertain their size. This means a bird flying near a camera will not be mistaken for a person scaling a fence in a far view. Because we understand how it is achieved, some argue that the process is not intelligent, but is actually a simple calculation based upon known data.
With this mind, it is easy to see that whilst there isn’t a defined track record for the growth of Artificial Intelligence in security, it has existed for many years.
As such, the clearest definition of AI is where a machine uses data about the environment to maximise its chances of success in a given task. By using reasoning and probability, AI allows a machine, system or solution to participate in the decision making which impacts on its performance.
AI-based systems work because the system has ‘learned’ about its environments and the various actions that take place, either normally or as a part of ‘exceptional’ activity. There are two main types of learning associated with AI: Machine Learning and Deep Learning.
Machine Learning is very common in the IT world and many of us use systems based upon this approach. Machine Learning is used by social media, search engines, on-line services and data management systems. It works by running a lot of data through a lot of algorithms and uses the results to ‘predict’ things in a given environment.
Machine Learning currently works well, but is reliant on knowing how to use the resultant data it has created to make future decisions.
Where Deep Learning becomes more than just Machine Learning is that by using numerous layers of algorithms (which is where the ‘deep’ reference comes from), it can understand an environment and make its own decisions based upon what it has learned.
A good way of understanding the difference between the two is that Machine Learning systems will search through millions of options to quickly find a solution in a given environment, because it has been told what to do based upon specific criteria. Deep Learning systems will use knowledge and experience to understand the environment, and then will decide how to act accordingly.
The sales message for a system using AI or Deep Learning does not lie in the fact that it employs these techniques, but rather how it can add benefits for end users and their businesses. A walk around any expo or conference will reveal many manufacturers with one (or all) of the buzz-phrases emblazoned on their materials. However, that tells you very little.
There are many potential applications for AI in security, and currently the technology seems to crop up more regularly with regard to analytics. AI and Deep Learning enables a system to analyse shapes and objects and classify these quickly and accurately.
The technology can differentiate a cat from a dog from a human on their hands and knees. It can recognise individuals using facial recognition. It can use colour of clothing or vehicles to identify targets. It can analyse behaviours to classify them as normal or exceptional.
Because the system is learning to recognise objects, people, vehicles and behaviours, this means that it can benefit from the advantages of AI. Searching, for example, becomes simpler and more effective. This is because instead of looking for an IVA violation such as a line cross in a specific sector from a single camera, or trying to locate an incident across multiple streams based upon time and date information, the system is searching all captured data for very specific information. For example, this might be for a man in a red jacket, a blue van or a woman pushing a pram.
Where AI, and specifically AI based upon Deep Learning, becomes very beneficial is once the suspect is found. The system can then search for any other associated footage, including footage from previous occasions. As each appearance is found and presented to an operator, the system uses that information to learn more about the target. Using other information, such as height or facial details for example, it can find other footage where the suspect might not be wearing a red jacket.
For example, this might allow the operator to discover how a suspect arrives at the protected site, and this information could then be used to direct further searches. The process simply enables the collection of evidential and forensic data with very little input from the operator.
Searching for suspects or tracking targets might seem like a simple task for a technology that is claimed to offer us a smart and intelligent future, but it offers real benefits and makes a task that can days or even months simple and easy to action. It also eliminates human error that is bound to occur if operators and manually scrolling through hour upon hour of video data.
A conservative estimate is that less than five per cent of captured video is ever used for any meaningful reason. This is due to the required time and resources to extract and manage useful data. However, AI can allow this data to be used for a wide range of purposes, and with Deep Learning a wide range of activities or processes can be assessed and used for management purposes. This adds value to the system for the end user, and makes sense in an age of Big Data!
Another task that AI, and especially AI based upon Deep Learning, can simplify is the configuration of analytics. This can be a slow and labour intensive task when carried out by installers and integrators.
For example, if a simple line-cross analytic is used, it could be configured during normal working hours and tested overnight to assess the level of performance. However, each nuisance alarm activation needs to be ‘tweaked’ out to deliver the right degree of performance. A system installed in summer might act differently in winter when the last few hours of the working day are during periods of darkness.
Benchmark recently spoke to a manufacturer involved in a major project which deployed 1,000 channels of video using IVA. Six months later only 150 channels of analytics were in use. This was because the resources and time required to properly configure them had been underestimated by the consultants involved.
The use of AI in combination with Deep Learning techniques will allow a simpler set-up of IVA and other system functions. This will still take time, as seasonal or irregular exceptions will need to occur.
However, rather than the installer or integrator having to return to site to adjust the system, the system will present the exceptions to an operator who can either flag this as important or dismiss them. As a result, via Deep Learning, the system will learn to ignore nuisance events!
Indeed, Deep Learning need not be limited to video analytics. It can assess a wide range of system activities, link these with behaviours such as access control transactions, recording server back-ups, database management and other system or environmental checks, and can help to ensure the correct operation of a system.
AI, Machine Learning and Deep Learning technologies might whet the appetite with thoughts of advanced and complex video analytics, prediction-based behaviour tracking or wide area situational management, and whilst such features will come in time, it’s still early days.
That said, AI can deliver real benefits simply by enabling advanced analysis of the huge amounts of data that security solutions are already capturing. In short, systems might not become ‘smarter’ over night, but they can manage all the data they capture effectively, and that’s something few systems have been able to do thus far. Maximising the potential of modern systems is a significant step forwards, made possible by these technologies.
AI, Machine Learning and Deep Learning are in their infancy in security, but the technologies promise much. The important point for installers and integrators is that products claiming to support any of these approaches probably won’t deliver the same results. The challenge is to discover what benefits they offer, and to ascertain how those benefits will translate into real value for the customer.
AI will continue to develop, but even today the technology can be deployed for certain tasks, and can add value to applications if sensibly deployed.