How AI is impacting

Physical Access Control

Security systems integrators and end users are experiencing a new, exciting and some
what scary transition. Over the last ten-plus years traditional analog security systems have
been disrupted by IP-based products. Most recently, the camera market has been
transformed by the used of standardization and open source approaches to the
compatibility of systems. This paper will examine the application of the next wave of
disruption to hit the physical security industry – artificial intelligence and how it will impact
the access control system market which has yet to fully adopt AI and IP-based solutions.

By examining the adoption of AI-enabled devices on the video surveillance market we can
observe parallels as to how the advances of this technology may follow a similar path in the
world of access control. We will examine the reasons for this and study the application of
integrated systems and how AI technologies will be instrumental in building traction in the
access control market. Lets begin by reviewing some basic access control system
concepts then we will re-define these concepts by relating them to the field of artificial
intelligence. Finally, we will look at the main types of access control system installs found in
the wild and observe some barriers and impacts of AI on access control and integrated
system design.

CLASSIC KEY SYSTEM

The most common type of access control is with mechanical lock and key system. Though
these devices are not controllable in an automated system, they have established some of
the concepts about the control of access and hierarchies that are carried through to
electronic systems. The various levels of key security include a single key, master key
system, grandmaster key system, and great-grandmaster key system. A single key
unlocks one type of lock by one type of key. A facility may have multiple doors in a building
keyed the same (using locks with identical keys) for convenience. A master key system
however, keys each door individually, but has a master key that can open all doors within
the master key system. This provides two levels of access. This key system is commonly
used in commercial buildings such as office buildings with several tenants. Each tenant
possesses a key that will only open the door to individual office space. While the building
management personnel can open any door within the master key system with the master
key. A grandmaster key is one level higher than a master key. In this system, a
grandmaster key can open all doors within multiple master key systems. This type of
system can be used in a building that has multiple floors or multiple divisions within an
organization. For example, a two-story building has a master key system for each floor.
The grandmaster key would open any door on either master key system, unless a specific
door is not a part of either master key system.
The hierarchy concept can be extended as far as needed, resulting in great-grandmaster
key systems and beyond. For example, a great-grandmaster key system will open any
door within each grandmaster key system and master key system, unless the specific door
is not part of the master key or grandmaster key systems. Sometimes a common
entrance or door must be accessible by people with keys under different master key
systems. This is accomplished by cross keying. Cross-keying is a keying method for
master key systems that allows a lock cylinder to be opened with two or more different
keys that would not normally operate together.

Neural Network Basics

Now let us review one of the more well know data processing techniques drawn from AI –
the artificial neural network.

To construct a computer system capable of “human-like thought”, researchers have used

the only working model they have available – the human brain. However, the human brain
as a whole is far too complex to model. Rather, the individual cells that make up the human
brain are studied. At the most basic level, the human brain is composed primarily of neuron
cells. They are the basic building blocks of the human brain. Neural networks attempt to
simulate the behavior of these cells. Biological neural networks are analog. Simulating
analog neural networks on a digital computer can present some challenges. Since a
biological neuron is analog it is only capable of processing information as a voltage signal. If
the voltage is within a certain range the neuron will activate. When a neuron activates, a
new analog signal is transmitted from the active neuron to other neurons on the network.
The signal connections between two or more neurons which represents the data inputs
and outputs is called the synapses. A neuron makes a decision by being active or non-
active. The decisions being made are extremely low-level decisions. It requires a large
number of decisions to be made by many neurons just to read this sentence! Higher-level
decisions are the result of the collective input and output of many neurons. When a large
number of neurons are connected they form the core neural network architecture which
consists of an input layer, a middle or hidden layer, and finally the output layer.

Figure 2. Neural Network applied to real world traffic signal processing

Problems Not Suited For a Neural Network

Solutions that are easily written as a flowchart are examples of problems where a neural
network would not be the proper choice. Another criterion to consider is whether the
requirements and logic of the solution is likely to change. One of the primary features of
neural networks is the ability to learn. If the algorithm used to solve your problem is an
unchanging business rule, there is no reason to use a neural network. In fact, it would be
detrimental to the system if the neural network attempts to find a better solution, and
begins to diverge from the desired process and produces unexpected results. Finally, neural
networks are often not suitable for problems in which you must know exactly how the
solution was derived. A neural network can be very useful for solving problems for which it
was trained, but the neural network cannot explain its reasoning. The neural network
knows something because it was typically trained to know it. The neural network cannot
explain how it followed a series of steps to derive the answer.

Problems Well Suited For a Neural Network

Although there are many problems for which neural networks are not well suited, there are
also many problems for which a neural network solution is quite useful. In addition, neural
networks can often solve problems with fewer lines of code than a traditional algorithm. It
is important to understand which problems call for a neural network approach. Neural
networks are particularly useful for solving problems that cannot be expressed as a series
of steps, such as recognizing patterns, classification, series prediction, and data mining.

Classification
Classification is the process of classifying input into groups. When applied to an access
control system, access groups such as maintenance supervisor(s), operations managers
can be managed and mapped to doors in the same access group as well as personnel that
need cross access to areas on a facility. Often, the neural network is trained by presenting
it with a sample group of data and database references as to which group each data
element belongs. This allows the neural network to learn the characteristics that may
indicate group membership. A well constructed neural network should be able to process
groups that has fuzzy data or data lacking clearly defined boundaries.

Prediction

Prediction is another common application for neural networks. Given a time-based series of
input data, a neural network will predict future values. The accuracy of the guess will be
dependent upon many factors, such as the quantity and relevancy of the input data.
Pattern Recognition

Pattern recognition is one of the most common uses for neural networks. Pattern
recognition is a form of classification. Pattern recognition is simply the ability to recognize a
pattern. The pattern must be recognized even when it is distorted. Currently, there are
many cameras on the market that implement this neural network function.

Optimization

Another common use for neural networks is optimizations. Optimization can be applied to
many different problems for which an optimal solution is sought. The neural network may
not always find the optimal solution; rather, it seeks to find an acceptable solution. This
neural network approach can be applied to integrated systems design; where an access
control system is integrated with a building automation system, PSIM, or some other
internet-of-things implementation. Integrated solutions tend to have a large number of
sensor and device nodes on a given network. Brute force methods of calculation do not
work nearly as well as a neural network approach. Neural network performance can also
be enhanced with the application of another AI algorithm known as genetic algorithms.

Training Neural Networks

The individual neurons that make up a neural network are interconnected through their
synapses. These connections allow the neurons to signal each other as information is
processed. Not all connections are equal. Each connection is assigned a connection weight.
If there is no connection between two neurons, then their connection weight is zero. The
weights are what determine the output of the neural network; therefore, it can be said that
the connection weights form the memory of the neural network.

Training is the process by which these connection weights are assigned. Most training
algorithms begin by assigning random numbers to a weights matrix. Then, the validity of
the neural network is examined . Next, the weights are adjusted based on how well the
neural network performed and the validity of the results. This process is repeated until the
validation error is within and acceptable limit. There are many ways to train neural
networks. Neural network training methods generally fall into the categories of: supervised,
unsupervised, and various hybrid approaches.

Supervised training is accomplished by giving the neural network a set of sample data along
with the anticipated outputs from each of these samples. Supervised training is the most
common form of neural network training. As supervised training progresses, the neural
network goes through a number of iterations until the output of the neural network
matches the anticipated output, with a reasonably small rate of error. An access control
database and building automation device nodes would be the perfect candidates as a
source for supervised training.

Unsupervised training is similar to supervised training, except that no anticipated outputs are
provided. Unsupervised training usually occurs when the neural network is being used to
classify inputs into several groups. The training involves many iterations, just as in
supervised training. As the training progresses, the classification groups are constructed by
the neural network. There are several hybrid methods that combine aspects of both
supervised and unsupervised training. One such method is called reinforcement training. In
this method, a neural network is provided with sample data that does not contain
anticipated outputs, as is done with unsupervised training. However, for each output, the
neural network is told whether the output was right or wrong given the input.
Understanding the how to properly train a neural network is absolutely essential for access
control, PSIM, internet of things, and building management system integration. A functional
neural network integration can be trained with these three methods: backpropagation,
simulated annealing, and genetic algorithms. Fortunately, Ai-tronics Systems provides an AI
engine which includes an SDK and training tools. Once the neural networks is trained, it
must be validated to see if it is suitable.

Validating Neural Networks

The final step, validating a neural network, is very important because it allows you to
determine if additional training is required. To correctly validate a neural network , validation
data must be set aside that is completely separate from the training data. As an example,
consider a classification network that must group facility personnel and visitors into different
classification groups. Your provided with 10,000 sample records from the access control
database. For this sample data, the group that each element should be classified into is
known. For such a system, you would randomly divide the sample database records into
two groups of 5,000 records each. The first group would form the training set. Once the
network was properly trained, the second group of 5,000 records would be used to
validate the neural network.

The approach of separating data into groups is a good one because it helps keep the
validation data organized. It is also a good idea to always keep the validation data set up-
to-date. This brings up an important question. What happens if the neural network that
you have just finished training performs poorly on the validation data? If this is the case,
then you must examine possible causes.

If the validation is performing poorly, it is likely that there was data present in the validation
data set that was not available in the training data. The way this situation should be
rectified is to try a different random approach to separating the data into training and
validation data. If this fails, you must combine the training and sets into one large training
set. New data must then be acquired to serve as the validation data.

In some situations it may be impossible to gather additional data to use as either training or
validation data. If this is the case, then you are left with no other choice but to combine all
or part of the validation set with the training set. While this approach will forgo the security
of a good validation, if additional data cannot be acquired this may be the only alternative.

Real World Access Control and Barriers to AI Adoption

AI technology is still new to the access control industry and legacy systems have yet to exploit the
advantages of AI or IP technology. There are potentially a number of reasons for this.

➢ Existing proprietary hardware/software

➢ Manufacturer reluctance to build AI based products
➢ Legacy Infrastructure
➢ End user reluctance to learn AI-base solutions
➢ Impact of AI-based Access control on installers

Existing proprietary hardware/software

Legacy access control systems are dependent on having each device (i.e. Door, Lock, and
Card reader) is hard-wired with RS-485 cable into a central server. Furthermore, they are
usually proprietary systems, which confine the end-user to a single vendor of hardware and
software.

Manufacturer reluctance to build AI based products

While this may be a controversial opinion this may well be a reason for slow adoption. If we
look at the commercial aspects for a vendor of one of the main enterprise access control
systems we can assume that millions of dollars of investment will have been sunk into the
R&D of proprietary technology that will have thousands of established site deployments
across the world. This a significant amount of time must pass before the vendor gets a
return on investment from thousands of world-wide hardware and software deployments.
Legacy Infrastructure

If an access control customer wants to expand upon an existing system, the process can
be complicated as centralized controllers are typically build to accommodate a maximum
number of doors. As a result there are built-in limitations to expansion. This coupled with
the cabling infrastructure being run in a serial configuration, with localized power
requirements means that upgrading a system that is many years old can be expensive.

End user reluctance to learn AI-based solutions

End users adopted network based IP video systems over a long period of time. The drivers
of this where picture quality, flexibility, storage and the improved transmission of
compressed images. These innovations meant that an end-user could have a scalable
system with improved image quality on an open communication platform that provided
flexibility and vendor choice. AI has gain traction among camera manufactures but is still
viewed with caution on the part of the end-user. On the access control systems side the
level of AI market awareness is still quite low.

At the time of this writing, these benefits have not appeared across the major access
control vendors and a typical scenario where an end-user who is currently using
manufacture A across her facility acquires another site with manufacture B already installed
means he has to do a complete refit with an update to the access control database as well
as a retrofit to manufacture A. This could also mean making a significant investment in a
new overarching control software such as PSIM which typically has biased vendor based
integration and does not include an AI system.

Impact of AI based access control on installers

AI-based access control is a simpler way to expand and integrate a security device with
non-security devices such as building management systems and IoT devices. On a new
site extra functionality can be added as an interface between the access control database
and the controller/reader devices. Since IP networks are now ubiquitous in offices, retail
and other facilities, the cost of adding an IP and AI controller is minimal, as opposed to
multiple serial connections wired into a central server. In addition to this benefit, large
amounts of time and materials can be saved by deploying the use of Power-over-Ethernet
(PoE) to power the systems controllers rather than user separate power at the door.
Furthermore, the traditional type of cabling means that there are several possible points of
failure on a typical system the could occur if one or more of the remote units failed. This
problem is solved with an AI-based system which eases the management of connecting
doors and credentials. Finally, AI integration provides another advantage to the installers as
the access control system can be placed on the same network as the building
management and video management systems. With a neural network manager, these
systems can work seamlessly alongside each other, unlike legacy systems which often
require a physical interface to communicate.

Impact of AI-based access control on end-user

With AI-based solutions, the ability to remotely manage a system becomes easier as
access control makes the transition from proprietary to open integrated systems, based
on proven techniques and international industry standards. Market sectors are beginning to
move toward application programming interfaces (APIs) which can be used or shared by
different manufactures. Even though there are AI development offers such as Google’s
Tensorflow, the market and level of awareness indicates that there is plenty of room for
improvement. In addition, communication protocols such as BACnet; which has sixty-
percent market share has an access control component as a part of the specification but
has features that are not functionally complete. New users or credentials can be added
instantly to a system from another part of the world under a centralized management
function, thereby reducing or eliminating the need for system administrators to be at each
site location. This reduces costs and increases the level of security.

End users can also benefit from the latest technologies which bring together identity
management with video surveillance which ensures the correct physical identities of
employees, contractors, suppliers and visitors which can be classified into a neural network
which can be properly authenticated and have the right access to the right areas, for pre-
specified durations of time. This gives the end user visibility of their site 24/7 and full audit
trails of their sites. This is a huge advantage to corporate security people in delivering full
compliance and risk management.
Conclusions
In conclusion there is an increasing demand for AI-based access control systems from both
end-user and installers alike., as each seek to realize the benefits they can enjoy from
integrating their video systems, and building management systems with a well trained
neural network. In addition to these benefits, the increasingly larger role that IT
departments play in selecting physical security means that AI-based access control
systems inevitably go down the same path as video surveillance systems and become
more network and AI-based in the future. This means integrating an AI interface to the
access control database layer which will bring new and exciting opportunities as installers
will begin to appreciate easier installations and increased integration possibilities with other
systems. Furthermore, installers will be free to add access points to their customers
systems seamlessly due to the scalability of supervised neural networks and will become
more vendor agnostic as they can choose ‘best in class’ products from different
manufactures to provide end users with more functionality.

Finally, end-users will be able to take advantage of affordable, flexible open platform
systems that can help them secure and protect their physical identities and manage access
compliance and risk.

Ai-tronics Systems, LLC

222 w Merchandise Mart Suite 570
Chicago, Illinois
+1(708) 715-7118
http://ai-tronics.com
© 2018