Artificial intelligence has been a part of the global technology lexicon ever since HAL, the Heuristically Programmed ALgorithmic computer was first introduced as the sentient AI that controls the systems of the Discovery One spacecraft in Arthur C. Clarke's 1968 science fiction film classic 2001: A Space Odyssey. HAL was capable of speech, speech recognition, facial recognition, natural language processing, lip reading, interpreting emotional behaviors, automated reasoning, spacecraft piloting; and he was a heck of a chess player. However, in the film, HAL’s penchant for overthinking became man’s worst nightmare as the AI designed to pilot the mission and protect the crew eventually went rogue when he felt threatened and became a malevolent force.
Clarke’s genius, and his vision of what artificial intelligence and machine learning might look like more than five decades into our future, has manifested itself in everything from automated factories and automobiles, to our constant companions Siri and Alexa. With the evolving dynamics of cloud storage and the ability to harness and proactively employ an ever-increasing pool of big data, AI in the form of machine learning and deep learning has become a disruptive technological force in the physical security industry. Advanced AI and cheaper network resources have significantly impacted video surveillance, which has been among the biggest beneficiaries of faster processing and impactful analytics. Even building automation, fire systems, intrusion detection, and physical and network access control have AI built into many core competencies.
Taking A Step Back to Assess
Before we say that AI has changed everything, we should step back a moment and assess the status of AI today. It seems clear that AI has the potential to play an increasing role in making exterior and interior entrances more secure in the fairly near term. But while AI will be able to help in many security-related tasks, such as discerning people from objects at a facility’s perimeter and interior entrances, detecting attempted piggybacking, spotting and analyzing potentially lethal objects and dangerous people, and more, AI alone cannot take action to prevent unauthorized human entry or deny the entry of dangerous objects.
As we move to greater converged technology at the edge, we face the quandary of how AI may practically interplay with entry solutions such as revolving doors, turnstiles, and swing doors to accomplish risk-reduction goals. Any disconnects between the objectives of the building owner and building code regulations are another factor that can further complicate the security blueprint.
Consultant Ben Butchko, President and CEO of Butchko, Inc. and a former security engineer with ExxonMobil, warns that manufacturers’ goals must align with their end users’ needs when it comes to driving the development of embedded solutions with advanced sensors (cameras, microwave, LIDAR), operational analytics (facial recognition, tracking, object discrimination, pattern recognition), and active response (entry lockout, alert notification) that now prevent or deter risks detected at the entry.
“To this point, the security entrance must be part of the general building operations design, clearly separated from an architect’s complete authority. Most secured entries are specified in CSI Division 28, outside of building design since it is structural and falls under code compliance surrounding emergency egress as well as building capacity and throughput. Therefore, if this is to work, the rules for design and the merging of Division 28 and Divisions 8 and 11 must become refined, practical, and widely accepted,” Butchko says.
Because many security entrances do not have AI built into their technology, integrating intelligence into secured entrances requires a collaborative effort with a third-party solutions provider. Video analytics are increasingly deployed to address use cases such as people detection, piggybacking, dangerous object detection and facial recognition among other issues relevant to secured entrances. The increased integration of AI providers with traditional security entrance partners has resulted in improvements in:
- Ease of use and usability
- The use of machine learning to improve algorithms over traditional modeling and correlation approaches
- Integration with other systems and sensors
Plan for a Converged Solution
Salvatore D'Agostino is the CEO of IDmachines, a technology firm that supports the lifecycle management of physical security, surveillance systems, and Internet-connected devices. D’Agostino believes the successful deployment of video analytics has always been dependent on a range of system design factors, including camera placement, lens selection, and lighting conditions, and a range of operational factors, including performance expectations, operational procedures and requirements, and the degree to which human interaction is required.
Security entrances often combine a number of systems, sensors, and requirements to achieve their purpose, which is tailgating mitigation. When deployed, these entrances by their nature are an integrated solution combining access control, mechanical hardware, sensors, software algorithms and most importantly, design.
The addition of surveillance cameras around doors or security entrances is an example of adding video deployment for the sake of forensics: the ability to tie what took place at the entrance to an alarm condition (e.g., a forced or jammed (propped) entrance/exit, or a tailgating incident). This functionality can be further enhanced by analytics; for example, facial recognition could be used to determine which individuals set off the alarm condition. Analytics can also be proactive, determining that a crowd has gathered and then automatically activating additional security entrances, bringing them online and ready for credentials until after the crowd has passed through.
“From a design perspective there is an increasing demand, due to COVID-19, for touchless access. In this case, the integration of technologies and the use of machine learning can be leveraged to provide efficient, safe and secure access. Machine learning and AI are well adapted to leveraging data sets and, over time, gaining an understanding of conditions and matching them to access control and individual requirements,” D’Agostino continued.
“At present there are advances that are taking place from system and sensor integration. Examples include mapping elevator calls to individuals with rights in an access control system, and identification with a phone to grant and automate access to a given floor while avoiding crowding conditions that will be of increased concern in the COVID-19 era. These are more of what is sometimes referred to as OFAI (old-fashioned AI) in the sense that you could do this with a rule set (e.g. no more than 3 people in an elevator at a time), optimize the elevator calls based on the people in a queue and the floors they need to access. That being said, it is also a good example of where machine learning might be able to improve on those rules and algorithms over time and even provide a feedback loop to stagger access and arrival times to meet access and safety objectives,” D'Agostino explains, citing this use case for turnstiles and swing doors as well.
Replacing Human Effort
So, what can AI do alone at a secured entrance? Basically, replace human effort at tasks that would be tough for humans to do reliably: learn behaviors of staff, employees, and contractors, and identify people and monitor them more consistently than operators.
If a camera with AI is paired with a security entrance solution and monitoring it constantly, such as a turnstile, security revolving door, or mantrap portal, it could take the detection capabilities currently built into these entrances to the next level in terms of identity verification and anti-tailgating/anti-piggybacking. Currently, security entrances detect tailgaters by using near-infrared sensors – if it appears that two separate objects are breaking through the sensor beams, an alarm is generated. In security revolving doors and mantrap portals, near-infrared, “time of flight” technology is paired with optics to create a 3D image of the objects inside the door; algorithms and sampling data are used to determine whether there is one person or two. When these technologies reject a user “for no reason” (for example a person enters a door with a box of pizza and wears a backpack), that’s known as a “false rejection.”
Advanced AI can fill the gap by recognizing people (through learned movement patterns and spacing of features) and objects the way humans can, which can bring the false rejection rate to near zero. For example, it could know the difference between any inanimate objects being worn or carried through the entrance vs. living users. It can know intimately the identity of authorized users, regardless of clothing, current weight, hair color or facial hair, and the process of aging.
Butchko says the use of machine learning as well as deep learning has been used for many years in the “Big Data” world to identify trends and produce metrics regarding human intent. The use of “synthetic cognition” or AI is part of the drive for establishing ways to create the correlation to human patterns and the completion of tasks in the hopes of creating greater efficiencies in business practice.
“Within the security industry we see a trend by companies to leverage these specific engines to gain greater benefits from access control, VMS, intrusion detection, and tracking systems. It can be seen as a double-edged sword, because tracking learned behavior can help define potential vulnerabilities and unmask potential threats, yet it can also lead to privacy and discrimination concerns, especially when intent and analytic detail are not clear,” Butchko adds.
Proactive and Predictive Possibilities
D'Agostino sees the convergence of AI into security spaces not known for their reliance on analytic data reshaping the landscape. AI can be used as a proactive step against intrusion at a security entrance like a swing door or turnstile and integrated into the access control and video security systems to provide rich analytics and situational awareness.
“It has long been known that there are often patterns to human, and to the same extent, enterprise behavior. Access control, surveillance, and intrusion detection systems collect large amounts of data that is often stored and then deleted without much analysis. Enterprises are now more attuned to the ability to leverage this ‘big’ data. These are evolving now to common data formats, real-time analytics and predictive tools. There seems like there would be a similar evolution in the capabilities of physical security systems where it is not so much what is happening at a turnstile, swing door or entryway – but what is going to happen,” D’Agostino explinas. “This would leverage the existing systems, sensors and data collection capabilities and use big data, and analytics to drive management and monitoring. The more that physical security systems adopt standard data types, sets and structures (using syslog for logging is a simple example) and the more intelligent these systems become, the more intelligence can be put into the predictive analytics.”
About the Author:
Kurt Measom serves as Vice President of Technology and Product Support at Boon Edam and is also part of the company’s Enterprise Account Security Team. Kurt has been employed at Boon Edam for over 21 years serving in multiple roles including Vice President of Technical Services, Training and Quality. Over the past four years, Kurt has worked closely with the company’s Enterprise Sales Team as an advisor for security solutions to many Fortune 1000 companies and is currently a Lenel Certified Associate.