When Surveillance Turns Around: How a Canadian Engineer Built an AI System to Track Police Vehicles

Introduction: A Small Experiment With a Big Message

A Canadian software engineer built his own version of the artificial intelligence cameras used to scan passing vehicles, but his project came with an unusual twist: instead of tracking ordinary drivers, his system was designed to identify police vehicles and record where and when they appeared. The experiment was intended to demonstrate how accessible automated surveillance technology has become. Yet it also attracted the attention of the very authorities whose movements the system was designed to monitor. Before long, police officers arrived at his door, raising questions about privacy, public safety, technological power, and the limits of surveillance in modern society.

The engineer behind the project, Anthony Sistilli, is a Canadian software engineer and YouTuber interested in demonstrating how emerging technologies work in practice. His experiment focused on a form of automated monitoring that has become increasingly common in public spaces: cameras equipped with artificial intelligence that can recognize vehicles, read license plates, and associate those plates with particular locations and times.

Companies such as Flock Safety provide systems that allow organizations to collect and search information about vehicles passing in front of their cameras. These systems can help investigators identify vehicles connected to criminal activity, reconstruct movements, and locate cars of interest. However, the same capabilities also create concerns about how much information can be collected about ordinary people who may have done nothing wrong.

Sistilli decided to turn the technology around. Rather than using a camera to record the movements of members of the public, he built a prototype intended to detect police vehicles, recognize their license plates, and log the time and place of each sighting. He reportedly assembled an initial version using an ordinary camera, a laptop, and AI-assisted software in approximately twenty minutes. The system was imperfect, occasionally failing to detect vehicles or read their license plates, but it demonstrated the basic principle.

The significance of the experiment was not that it introduced a revolutionary technical breakthrough. Instead, it illustrated how tools once associated with governments, law enforcement agencies, and well-funded technology companies can increasingly be reproduced by individuals with relatively accessible equipment and software. Automated surveillance is no longer necessarily confined to institutions with large budgets or specialized technical teams.

The reaction to the project made its message even more striking. Police officers reportedly visited Sistilli because of concerns about what a database of police vehicle sightings might reveal. Such information could potentially expose officers' movements, routines, and locations, creating risks for their privacy and safety.

Those concerns deserve serious consideration. A system that systematically records where police vehicles appear could be misused, particularly if its data were combined with other information or made available to people seeking to intimidate, harass, or harm officers. Yet the situation also reveals an uncomfortable contradiction. Ordinary members of the public have raised similar concerns for years about automated cameras that record their vehicles and movements.

By directing the same technology toward police vehicles, Sistilli created a practical demonstration of a broader social question: does surveillance become more acceptable simply because it is used by an institution with authority, or should the same principles of privacy, accountability, and proportionality apply regardless of who operates the camera?

His experiment offers a useful starting point for examining the growing role of artificial intelligence in public surveillance. It highlights the relationship between technology and power, the challenges of protecting privacy in an increasingly data-driven society, and the possibility that tools designed for monitoring the public may also be used to monitor the institutions responsible for enforcing the law.

1. The Technology Behind Automated Vehicle Surveillance

To understand the significance of Sistilli's project, it is necessary to examine how automated license plate recognition works and why it has become an important component of contemporary surveillance systems.

At its most basic level, an automated vehicle monitoring system combines cameras, image-processing software, optical character recognition, and a database. Each component performs a different task, allowing the system to transform images of passing vehicles into searchable information.

The camera captures images or video of vehicles moving through a particular location. Computer vision software then identifies vehicles within those images and attempts to locate their license plates. Once a plate has been detected, optical character recognition, commonly known as OCR, analyzes the characters and converts them into machine-readable text.

The system can subsequently associate the recognized plate with information such as the time of detection and the location of the camera. Depending on the system's configuration, it may also record characteristics of the vehicle, including its color, make, model, or direction of travel.

The resulting information can be stored in a database and searched later. An investigator, for example, might search for a particular license plate to determine whether the vehicle was recorded near a location at a certain time. If several cameras are connected to the same network, records from different locations may help establish a sequence of sightings.

This process can reduce the amount of manual work required to review hours of video footage. Instead of asking an employee to watch recordings and identify individual vehicles, a user can search a database using a license plate or other available information.

The technology's practical appeal is straightforward. Vehicles leave traces of their movements whenever they pass monitored locations. By automatically collecting those traces, surveillance systems can make it easier to reconstruct where a vehicle has been and when it was there.

However, the effectiveness of such systems depends on several technical conditions. Cameras need sufficiently clear images, license plates must be visible, and software must correctly distinguish characters that may look similar. Poor lighting, weather, motion blur, unusual plate designs, obstructed views, and camera placement can all affect performance.

Sistilli's prototype reportedly experienced some of these limitations. It sometimes missed vehicles or failed to read license plates accurately. These shortcomings are important because they demonstrate that automated recognition is not the same as perfect recognition.

Nevertheless, a prototype does not need to perform flawlessly to demonstrate a technological possibility. If an inexpensive setup can identify some vehicles and create a basic record of their appearances, it may be enough to show that similar systems are within reach of technically capable individuals.

The wider significance lies in the combination of relatively accessible hardware and increasingly capable software. A camera provides the visual information, a computer processes it, and AI-assisted tools can help developers build the software needed to connect the components.

What once might have required a specialized engineering team can now be approached as a small personal project. The result may be less reliable than a commercial product, but the underlying capability is no longer restricted to large organizations.

This change has important implications. The availability of surveillance technology is expanding at the same time that societies are debating how much monitoring is appropriate in public spaces. The cost of creating a monitoring system may be falling faster than the development of widely accepted rules governing its use.

As a result, the central question is no longer simply whether surveillance technology exists. It is increasingly about who can deploy it, what information they can collect, how long they can retain that information, and what limits should apply to its use.

2. From Corporate and Government Systems to Individual Experiments

For many years, large-scale surveillance systems were associated primarily with governments, police departments, transportation authorities, and private organizations capable of purchasing specialized equipment. These institutions generally had the financial resources, technical expertise, and operational infrastructure necessary to collect and manage large volumes of data.

The development of more accessible computing tools has gradually changed that situation. Individual developers can now experiment with image recognition, machine learning, data storage, and automated analysis without needing to build every component from scratch.

Cloud services, open-source software, widely available cameras, and AI-assisted programming have reduced some of the barriers to experimentation. A developer may be able to adapt existing libraries, use pretrained recognition models, and connect several software components to create a working demonstration.

This does not mean that every individual can immediately reproduce a sophisticated commercial surveillance network. Reliable operation at scale requires considerably more than a camera and a laptop. A system intended for continuous use may need secure storage, dependable networking, careful calibration, error handling, maintenance, and procedures for managing large quantities of data.

Commercial systems may also offer integrated dashboards, technical support, access controls, and other features that a quick personal prototype lacks. A demonstration that works under limited conditions should not be confused with a fully developed operational system.