Getting smarter about perimeter and border protection with advanced analytics

John Merlino

A major challenge in providing perimeter protection is not simply deploying video to help secure a perimeter or border. Rather, it is obtaining relevant video after installation, which ultimately yields best situational awareness for the operator.

A significant amount of recorded and live video is never viewed. Why? High densities of IP cameras—common in Department of Defense services and federal agencies—provide live video 24/7, making it virtually impossible for surveillance operators to observe everything recorded. There is just simply too much footage to examine.

In the past 20 years, the security industry has shifted from proprietary, closed-analog, closed-circuit systems toward digital, open, IP-based networks. Why? New IP systems are more solutions-based and offer a host of benefits including partner eco-systems and lower cost of storage. Advances in storage technology enable us to stockpile more data and, more importantly, manage it better. So, not only do we have a lot of video at reduced cost, but with analytically enhanced video, IP video systems retain even more relevant footage.

Through analytics, artificial intelligence (AI) and machine learning (ML), we can whittle down huge amounts of data to just needed information, then use this information to take appropriate action in a far timelier fashion.

Analytically-enhanced video also enables us to store only pertinent video as data. AI and ML applications can now inject this meta-data and begin producing what can become valuable tools across both the public and private domains.

All of the above parameters help us become “smarter” about the way we use video for perimeter protection.

AI and ML: What they are and how they integrate into security solutions

Before we go further, let’s define AI and ML.

AI, the broader term, is a component of computer science that enables computers to carry out tasks normally completed by people and perform them in an intelligent way.

ML takes this concept one step further. ML is an application of AI that allows computers to analyze data and “learn” the best way to proceed. It’s basically an automation of analytical model building; the computer can “figure out” the optimum solution without being programmed to do so beforehand.

All of this has major implications for video surveillance, especially when it involves the wide expanses of perimeter protection. AI and ML software can not only pick out the types of images we seek but also scan through all of the video data, selecting footage that requires further review based on physical characteristics, movement, behavior and other criteria.

For example, the system can send alerts when there is movement in restricted areas, such as the perimeter of an airfield. AI can then determine whether this is an individual, a vehicle or an animal. It could also be a benign weather-related issue. Based on previous “experience,” ML can recommend an appropriate response.

AI and ML are “force multipliers,” which enable physical security applications to optimize the use of security personnel. With fewer operators needed to monitor video, organizations can operate more efficiently.

Analytics is, and will continue to, play a key role in perimeter and border protection.

Intelligent video: Getting smarter about perimeter protection

Another term for video analytics is “intelligent video,” which makes use of three basic types of computer technology:

  • Pixel-based analytics: Pixel-based analytics, the most basic format of the three, sends alerts when there is a loss of video quality (due to equipment tampering for example) or image movement is detected.
  • Object-based analytics: Object-based analytics are far more sophisticated in that they can recognize objects, such as cars, people, trees and buildings, among others. This technology can be broken into two main categories: object tracking and object recognition.
  • Specialized analytics: Specialized analytics use pixel- and object-based information to examine video for specific applications, such as license plate identification, facial recognition or fire detection.

These analytic tools have many applications for perimeter protection, such as:

  • Crossline detection: Crossline detection applications (also known as tripwires) create virtual “fences” or restricted areas, and rules can be generated for these boundaries. For example, an alert can be sent when a car crosses a pre-determined boundary.
  • Intrusion detection: Intrusion detection, another type of object tracking, can determine when unauthorized people or vehicles enter a prohibited zone. It tracks their movement from one zone to another and typically begins recording when movement is detected in a “hot” zone.
  • Object left behind: This type of motion tracking can also flag an “object left behind” when an object, which had been moving, becomes stationary and remains that way for an inordinate amount of time. This may be an explosive, a cash payoff or an illicit drop-off of classified information.
  • Loitering detection: Likewise, object tracking can log the amount of time people linger in a certain area. If this is deemed to be inappropriate or illegal loitering, intelligent video can pinpoint the time, location and possibly identify the people involved.

By alerting personnel to unusual movements or behaviors—and then categorizing these events—intelligent video enables security personnel to recognize potential security threats while they are still on the perimeter, allowing for a timelier and appropriate response.

As a deterrent to illegal border crossings, for example, intelligent video can provide several layers of detection. A common technology deployed for detection at range are thermal cameras. Analytics with ground based radars and others with optimized HD camera technology can be used to interrogate the detection areas and determine the level of threat and what further response is necessary.

Facial recognition and license plate recognition are two additional security measures that are becoming increasingly more viable and quite valuable in augmenting perimeter and border security The use of facial recognition analytics has the potential to revolutionize many border and perimeter security applications, enabling surveillance personnel to quickly identify suspicious travelers, including those on no-fly lists. Both of these analytics will be huge enablers in future AI, ML and deep learning applications providing the exact kinds of metadata these engines need to ingest to be fully realized.

Predictive analytics: Analyzing aggregated data to predict future events

Now that I’ve provided an overview of analytics and intelligent video, let’s discuss how analytics might be used to try to make future predictions about potential events or non-events.

We can’t access future information, but we can collect analytically-enhanced data to more accurately predict potentially imminent occasions. Sure, while this may sound like something straight out of Steven Spielberg’s’ Minority Report, it’s actually much more reality than Hollywood.

What we’re talking about is a concept called predictive analytics. This involves the use of meta-data, derived from ML and AI, to drive intelligence so security personnel can more accurately forecast events and behaviors. As you can imagine, this is particularly important to security personnel who strive (and really need) to act proactively rather than reactively.

The many shapes and forms of predictive analysis

Predictive analytics is used in a number of situations. For example, England and Los Angeles use predictive policing to try to forecast where crime is most likely to occur. Predictive policing works when algorithms crunch years of historical data and crime reports to determine crime hot spots. Law enforcement can then focus on these locations when patrolling.

The capabilities of predictive analytics, however, don’t stop there. A recent study, published in the Journal of Empirical Legal Studies, even determined that predictive analytics could help courts more accurately determine whether or not to release an offender during his or her arraignment before the next court hearing.

Researchers came to this conclusion by first gathering data from 28,000 domestic violence arraignment cases from a major metropolitan area between 2007 and 2011. They then used ML to analyze over 35 characteristics, such as age, gender, demographic data and previous convictions. Typically, the court would have granted bail to the majority of offenders, and 20 percent would have recommitted a crime. However, researchers found that if the court used the ML method and only released offenders that the model predicted wouldn’t commit a crime upon being released, the percentage who recommitted an offense would drop to 10 percent.

Predicting crime along borders and perimeters

In border security, the need for predictive analysis is greater than ever. After all, many borders extend hundreds, if not thousands, of miles. (The U.S.-Canada border extends over 1,500 miles, and the border between the U.S. and Mexico stretches roughly 1,900 miles.) Such a long distance, new research suggests, has unsurprisingly strained border patrol units; they recently struggled to meet their target Interdiction Effectiveness Rate. IER is the percentage of illegal border crossers who are either detained or turned away at the border during a given year. The IER increased from 76 percent at the end of fiscal year 2013 to nearly 83 percent at the end of fiscal year 2016. However, in 2017, that number decreased to 78.9 percent, which means border control is finding it increasingly difficult to protect borders from illegal crossings.

Predictive analysis can (and is) being used to more accurately determine where to place border officers based on risk profiles instead of just footfall or expected (or common) entry points. In turn, it can help security better allocate resources. Similar methods are also being used in other perimeter protection settings, such as airports, where natural perimeter protection methods are no longer enough—a topic we described in greater detail in a previous blog post.

According to a report, Los Angeles Airport is using a game theory application to randomize checkpoints, and the schedules of guards and canines to protect against criminals who may exploit security patterns and weak points. However, the collaborative application doesn’t simply shuffle schedules around. It weighs the costs and benefits of placing personnel in one location over another and takes into consideration real-world constraints, which users can enter into the system, to form schedules.

Predictive analytics can also provide relevant data to objectively screen individuals, identifying those who may be more high-risk than others based on behavioral patterns they’re displaying. The theory is that people who are about to commit a crime typically display certain key macro and micro behaviors. Macro behaviors may include trying to hide one’s face from a camera’s view while micro behaviors include such actions as perspiration or failing to make eye contact. While experienced professionals should be able to detect these movements on their own, machines can do so with a much greater degree of accuracy to more people in shorter time.

The world of AI, ML and predictive analytics is no longer relegated to just big screen movies like Minority Report, iRobot, Eagle Eye and countless others, whose plots all play off these concepts. Today, analytics is a part of everyday life, and it continues to rapidly advance, limited only by our willingness—and need—to explore new technologies to safeguard people, borders and infrastructure.

For more information on how to develop a comprehensive perimeter protection plan, check out this blog developed by Quang Trinh, Team Lead, Professional Services, Business Development, North & Central America, Axis Communications, Inc.

Developing a perimeter protection plan