Revolutionizing Safety: The Journey to AI-Driven Security

Dmitry Sokolowski
April 3, 2024
AI Explained
We Care About School Safety
Exclusive Offer
for K-12 Public Schools
per stream, per month
Applicable for '24/'25 school year
Contact Us
Offer valid until September 1, 2024
One-time installation fee will apply

The Good AI

As we stand at the precipice of a technological revolution, the doom and gloom AI stories seem to dominate Hollywood, making it even harder to distinguish the truth from science fiction. Hence, I’d like to write my own AI story as I fundamentally believe that technology - done responsibly - can make the world a better place, specifically I want to share the vision of AI that is used for a noble cause – protecting our communities. But first, allow me to introduce myself and the journey that brings me here.

My career began as a software engineer, consulting for the US government, focusing on modernizing systems and adopting Cloud technologies. My journey took me to the innovation hub of Silicon Valley, where I had the privilege of working on diverse projects, from consumer products to cutting-edge AI systems at tech giants like Apple, Amazon, Uber, and Facebook.

A pivotal moment came in 2018, following a harrowing experience during the active shooting event at the YouTube campus. This incident ignited a passion in me and my co-founder to create something meaningful technology that could preempt shootings and other violent events. Thus, VOLT AI was born with a mission to save lives and enhance the safety and security of our communities, ranging from educational settings for our children to their parents' workplaces.

In this series of posts, we will dive deep into the interactions between AI and humans, particularly in physical security. Join me as we explore this fascinating and critical aspect of our modern world.

The Most Useless Security Invention

The status quo of physical security largely relies on a triad of gates, guards, and guns – a methodology rooted in visible deterrence and reactive response that tracks its history back to the invention of guns, but even before that to the tribal lifestyle. Sure, the introduction of guns increased the reach of response, but it didn’t bring any fundamental changes to the reactive nature of traditional security measures.

Finally, in the early 1950s, advancements in motion picture and VTR technology gave us something novel - a device that was going to revolutionize the security industry, something worthy of protecting Her Majesty the Queen during the coronation - a surveillance camera.

At last, security teams could see everything, everywhere, all at once, record all events, capture the faces of suspects, rewatch incidents multiple times, and it changed .. not much, as it turns out.

First, security guards, the human element in this equation, face inherent limitations. Vigilance can wane over long shifts, and human reaction times may not be swift enough to counter fast-unfolding events effectively. Moreover, guards are not infallible and can be prone to errors in judgment or can miss critical cues.

Second, as the number of security cameras grew - there are now over one billion surveillance cameras in the world - it simply outpaced the number of people that could physically monitor them in real time. As a result, only 2-5% of security camera feeds are being watched live.

As such, the traditional security approach is still predominantly reactive: guards with guns respond to incidents after they occur, too often when it’s already too late. Moreover, traditional security measures are often not equipped to deal with sophisticated or unconventional threats, such as insider threats. In the meantime, workplace violence and active shootings are at all times high, and security cameras are deemed useless when it comes to incident prevention. But do they have to be?

Deus Ex Camera

The explosion of big data and a leap in computing power in the early 2010s brought us a new wave of tech revolution - the rise of AI, and by ‘AI’, I mean robust Deep Nural Network models, capable of consuming large amounts of text and visual data, learning its patterns, and then making predictions about new data based on that knowledge.

The application of this technology has been vast - from recommendation engines (think of Netflix’ “You might also like..”) to financial models to visual data interpretation. One of the most famous examples of the latter is Google’s AI, which was exposed to 10 million randomly selected YouTube videos and learned to identify cats in three days.

Today, Vision AI models are powering new industries, such as AI content creation, autonomous vehicles, and robotics. However, security camera surveillance is one of the least talked about but real-world applications. AI is giving security cameras a new meaning and a chance to prove themselves as a proactive security tool, capable of protecting people by alarming them about a visual threat.

It means that you no longer need to keep adding more guards, guns, and gates to improve security posture nor need an army of camera operators to monitor your cameras around the clock. Instead, you can focus on responding to and resolving the incidents in real-time, equipped with the relevant visual intelligence.

Below are a few examples of what AI is capable of doing with security camera footage:

Behavior Analysis: AI algorithms can analyze behavior patterns, identifying anomalies that could indicate suspicious activities. This proactive approach allows for early detection of potential threats, such as someone loitering in a sensitive area or exhibiting unusual behavior.

Object Recognition: Advanced AI systems can detect and identify objects that may pose security risks, such as unattended bags or weapons in public areas. This technology extends the capabilities of surveillance beyond mere visual recording.

Crowd Analysis: In scenarios with large crowds, AI can analyze crowd dynamics to identify potential risks, like crowd crushes or rapidly evolving disturbances, enabling quicker response to emerging situations.

License Plate Recognition: AI is used to read and catalog license plates, which is valuable in monitoring vehicle traffic, tracking stolen vehicles, or enforcing parking and traffic regulations.

Thermal Imaging and Night Vision: AI enhances thermal and night vision technologies, allowing for effective surveillance in low-light or no-light conditions, and can even detect temperature anomalies that could indicate fires or chemical spills.

Facial Recognition: AI-powered cameras are increasingly capable of identifying individuals through facial recognition technology. This application is beneficial in access control and identifying persons of interest.

Perception System: The Next Frontier of Vision AI

When I embarked on a journey to create the most robust AI system to protect our communities, I quickly realized that incident data is one of the biggest limitations of the current state of smart cameras. Despite an astronomical number of security camera footage generated by the billion cameras worldwide, security threats are relatively rare. In addition, security incidents typically unfold through space and time, so every camera would only capture one piece of the puzzle, but no camera would have the whole picture.

And this is how the VOLT AI Perception System was born. At its core, it's an advanced artificial intelligence designed to enhance surveillance beyond the typical point camera capabilities by fusing signals across all cameras to create a comprehensive understanding of space in time.

To better picture the difference between a point camera and an AI perception system, let’s do a little experiment. First, close your left eye - what did you notice? Your field of view was reduced, and the visual center shifted to the right. If a car came towards you from far left, you’d notice it with a delay.

When you have both eyes open, the car on your left will be perceived by your left eye a few seconds sooner; the information will be communicated to your brain, then back to the right eye to watch out for it, starting an ongoing feedback loop between your retina and neocortex. You’re tracking the car as it passes by, learning more about its speed and direction as you observe it, continuously evaluating if you need to adjust your own speed and position to avoid it.

Now, apply the same concept to security cameras: instead of getting a signal from one camera (one ‘eye’), the AI ‘brain’ is simultaneously tuned into all of them (hundreds of ‘eyes’). This multi-camera - single brain communication is the foundation of VOLT’s perception system. What does it mean for security surveillance?

Multi-Camera Threat Tracking in Real-Time: VOLT's system can seamlessly track a threat or incident across multiple cameras and spatial environments. For instance, if an unauthorized individual enters a facility, VOLT's system can follow their movement across various camera feeds, providing a continuous and comprehensive view of their actions, unlike single-camera systems that lose track once they move out of their field of view. This capability also eliminates the need to do incident recreation, which takes hours to sort through the historical footage.

Event Reasoning Beyond Camera Views: The system's ability to reason about events extends even to those not directly visible in the camera feeds. For example, suppose a door in a restricted area is opened, but the camera's view is obstructed. In that case, the system can infer the event from other data points, such as access control logs or motion sensors, providing a more complete security picture.

Holistic Environmental Understanding Over Time: VOLT's AI system develops a comprehensive understanding of an environment's characteristics, learning from patterns and changes over time. This could include recognizing the usual foot traffic patterns on campus and detecting anomalies that might indicate a security threat or operational issue—for example, students running through a hallway where such behavior is unusual.

Enhanced Accuracy with Multi-Camera Analysis: VOLT's system synthesizes these viewpoints to increase accuracy when multiple cameras capture the same scene. For instance, in a crowded event, different camera angles can be used to more accurately identify and assess the behavior of individuals, reducing the chances of false positives.

Security Posture Analysis and Optimization: VOLT AI can analyze and reason about the overall security posture, including the effectiveness and optimal placement of cameras. By assessing camera coverage and the quality of detections, the system can recommend adjustments to improve overall security. For instance, it can suggest repositioning cameras for better coverage or upgrading to higher-resolution models in critical areas.

Predictive Analysis for Proactive Security: Leveraging historical data and pattern recognition, the system can predict potential security threats or operational disruptions, enabling proactive measures. For example, if it identifies a pattern of unauthorized access attempts at a certain time, additional security measures can be implemented in advance.


I'm excited about the potential of Vision AI to make our communities safer, and I'd love to hear from you. Whether you're an engineer, a security expert, or a parent like me who's passionate about public safety, your insights are invaluable. After all, the true value of any technology lies not in its sophistication but in the trust and belief of those who use it. Let's connect and share thoughts and questions—your perspective matters.

Author: Dmitry Sokolowski

AI Explained

Subscribe to our newsletter