Transformative Advances

• By Florian Matusek
• Jul 23, 2024

Over the past decade, machine learning has enabled transformative advances in physical security technology. We have seen some amazing progress in using machine learning algorithms to train computers to assess and improve computational processes. Although such tools are helpful for security and operations, machines are still far from being capable of thinking or acting like humans. They do, however, offer unique opportunities for teams to enhance security and productivity.

Opportunities in Security and Operations
In physical security, machine learning can speed up investigations by assisting security teams in finding relevant information quickly. Machine learning helps analyze data that hasn’t been organized or labeled. By identifying patterns or possible relationships between data, such tools help security teams, law enforcement, and other staff gain a better understanding of an incident or important correlations between trends.

On the operations side, machine learning helps organizations better leverage their physical security investments. Machine learning can help glean actionable insights from the system data to ramp up productivity. This often leads to improved operations.

For example, machine learning can help automate people counting, monitor traffic flow and enhance cybersecurity by identifying and blocking malware. It can also enable automation that helps organizations adhere to various industry standards and regulations through streamlined processes.

Deep Learning to Structure Data
One type of machine learning that has been particularly influential in the physical security world is deep learning. It uses task-specific algorithms to train computers to classify data.

Programmers begin with data sets that have been carefully organized or labeled. Then deep learning tools take unstructured data, such as hours of video footage, and turn it into structured data. The computer recognizes matching patterns or correlations that it can apply to other instances. As a result, teams can more easily find the specific items or events they are looking for.

For example, modern video analytics systems may use deep learning to “read” the letters and numbers on license plates. Other teams may use the tools to count how many people pass through a door or pick out a certain model of a car that passed by on a busy road. An operator can ask the system to display video that includes a red truck with a specific license plate or a person wearing blue jeans, a plaid shirt, and a brown baseball cap. Quickly sorting through data helps speed up processes and allows teams to operate more efficiently.

Combining Video Analytics with Automation
Using machine learning, systems can compile data from cameras and use video analytics to detect specific activities or items. Then, through automation techniques, the system can respond in certain ways if such an event occurs.

For example, when an intruder is detected, the system is programmed to alert the security team. Video and sensor data help track the intruder’s progress on a map, so responders find them more easily. The system may take additional actions, such as locking interior doors and notifying law enforcement. Throughout the process, it is important to note that every action follows criteria established by the programmers who set up these workflows in the system. The automation tools simply follow the steps based on the inputs received and identified.

Unification, Privacy and Human-centered Design at the Core
Unified physical security systems that leverage machine learning can collect and interpret a wide variety of data from many sources. Likewise, open architecture systems give security professionals the freedom to explore applications from various manufacturers. As new solutions come to market, teams can try them out and select the ones that best fit their objectives and environment.

Regardless of the applications selected, it is important to use the technology in ethical ways. Confirm with your manufacturer that the data your deep learning system is trained on is properly sourced and allowed to be used. People must give consent, and the manufacturer should ensure their right to privacy is respected.

While machine learning has come a long way in the last few decades, it is important to remember that it is not magic. To be useful, machine learning technology must be combined with human-centered design that is grounded in real-world customer problems.

When exploring options, begin by clearly identifying the challenges you want to address or the outcomes you seek. Then explore if machine learning is the right tool for that. It is not about simply implementing the latest application but understanding how that solution will impact your goals.

Likewise, humans need to be the final decision-makers and confirm that best practices are in place in all situations. While machine learning systems can streamline processes, sort through data, and help ensure procedures are properly followed, they do not replace human expertise.

This article originally appeared in the July / August 2024 issue of Security Today.