In February 2026, a security camera in Sherwood, Arkansas read a license plate one character wrong, and an innocent couple ended up ordered out of their car at gunpoint while their six-week-old baby sat in the back seat. That was not a rare glitch. An Institute for Justice review published in July 2026 documented at least 26 cases since 2018 of innocent people being pulled over, detained at gunpoint, or jailed because of automated license plate reader errors, with the majority happening since 2023.
License plate recognition is one of the most widely deployed camera technologies in physical security, and one of the most widely misunderstood. It is a powerful identity and forensic tool. It is also frequently mistaken for something it is not: a system that detects threats. This guide explains what license plate recognition actually does, where it fails, and why the same cameras running an LPR system are usually blind to the events that hurt people and property.
What is license plate recognition?
License plate recognition (LPR), also called automatic license plate recognition (ALPR) or automatic number-plate recognition (ANPR), is a camera-based technology that captures an image of a vehicle's plate and converts the characters into searchable text. The system pairs a specialized camera, often with infrared illumination, with optical character recognition software. It records the plate number, a timestamp, and usually the camera location, then checks that string against a database or hotlist.
The output is not a warning that something is wrong. The output is a record: this plate was at this place at this time. That record can be matched in real time against a list of stolen vehicles or flagged plates, or stored and searched later during an investigation. In the United States, where the Federal Highway Administration counts more than 250 million registered vehicles, LPR lets operators check plates at a scale no human could match.
That is the important distinction to hold onto. A plate reader answers the question which vehicle was here. It does not answer the question is something dangerous happening right now.
How license plate readers work, and where the accuracy problem hides
A license plate reader follows a fixed pipeline: capture the image, locate the plate in the frame, read the characters with OCR, and store or match the result. Vendors commonly advertise better than 90 percent read accuracy, but that figure describes controlled conditions, not a rainy Tuesday night in a poorly lit lot.
The gap between the marketing number and real-world performance is where the trouble lives. Flock Safety, now the leading provider in the market, states its cameras accurately capture 93 out of every 100 plates. Flock's own materials also describe more than 20 billion plate reads per month. Even taking the 93 percent figure at face value, that implies well over one billion inaccurate reads every month, and the overwhelming majority of scanned plates belong to ordinary people who are not connected to any investigation.
The errors are mundane and consistent. Systems confuse an O for a 0 and a 2 for a 7. Dirt, a tinted plate cover, a temporary paper tag, snow, glare, motion blur at highway speed, and non-standard or vanity plates all degrade the read. Shadows and direct sunlight alone can cut recognition accuracy by 20 to 30 percent. When a machine misreads a character, or a person enters the wrong plate into a hotlist, the consequence is not an abstract data-quality issue. In nearly two-thirds of the wrongful-stop cases the Institute for Justice catalogued, officers had already drawn their weapons before realizing the alert was wrong.
Reading a plate is not the same as detecting a threat
The single most important thing a security director can understand about LPR is that it reads text on the back of a car and nothing more. It does not watch behavior. Point a plate reader at your entrance, your parking structure, or your loading dock, and it will faithfully log every vehicle that passes. It will not notice a person carrying a rifle across that same lot, a worker collapsed between two parked cars, smoke rising from a dumpster, or someone scaling the perimeter fence after hours.
This is the reactive-versus-proactive divide that defines modern physical security. A plate reader, like traditional recorded video, is fundamentally a look-backward tool. It helps you reconstruct who was present after an incident is already over. That has real investigative value. It has almost no preventive value, because by the time the plate data matters, the harm has usually already happened.
The graphic below lays out the split in plain terms: what a plate reader delivers, and what it structurally cannot.
Notice what is missing from the left column. Every item in the "what it misses" list is a live event that a security team needs to respond to inside a window of seconds, and none of them involve a license plate. That is not a flaw in any particular vendor's LPR. It is the nature of the technology. Optical character recognition reads characters. It was never built to interpret a scene.
License plate recognition vs. AI behavior detection
The clearest way to think about these two technologies is that they answer different questions, and most facilities need both answered. LPR tells you the identity of a vehicle. AI behavior detection tells you whether a person or object in view is doing something that requires a response. The table below maps the practical differences.
| Capability | License Plate Recognition | AI Behavior Detection |
|---|---|---|
| Primary output | A plate number, time, and location record | A real-time alert that a defined threat is occurring |
| Core question answered | Which vehicle was here? | Is something wrong right now? |
| Detects a visible weapon | No | Yes |
| Detects a fall or person down | No | Yes |
| Detects fire or smoke | No | Yes |
| Detects a perimeter or rooftop breach | No | Yes |
| Detects loitering, tailgating, crowding | No | Yes |
| Depends on a readable plate | Yes, and misreads are common | No |
| Uses facial recognition or personal identity | Ties a plate to a registered owner | No facial recognition; analyzes behavior, not identity |
| Runs on existing cameras | Often needs purpose-built plate-capture cameras | Layers onto the cameras you already have |
| Primary value | Investigation and access records after the fact | Prevention and response before harm occurs |
The last two rows carry the most weight for a facility budget. Purpose-built LPR usually requires cameras positioned, focused, and shuttered specifically to freeze a plate on a moving car. AI behavior detection runs on the standard cameras already mounted around your property, which is why adding it is a software decision rather than a hardware project. And because behavior detection reads what is happening rather than who someone is, it does not rely on facial recognition or any biometric database, which sidesteps the privacy and civil-liberties questions that follow plate-reader networks everywhere they go.
When license plate recognition is actually the right tool
License plate recognition earns its place in specific, identity-driven use cases, and pretending otherwise would be as misleading as overselling it. LPR is the correct tool when your problem is genuinely about vehicles and access rather than about threats. Consider it when the job is one of these:
- Automated access at a gate or garage. Reading a resident or employee plate to raise a barrier arm or process parking is exactly what LPR is built for.
- Toll and payment enforcement. Billing a registered owner for a toll or an unpaid stay is a plate-matching task, not a threat-detection task.
- Watchlist and VIP recognition at controlled entries. Flagging a known banned vehicle or recognizing an expected visitor at a single chokepoint is a reasonable LPR application, provided a human verifies every hit before acting.
- Post-incident investigation. A stored plate history can help reconstruct which vehicles were present around the time of an event.
What ties those together is that none of them is a real-time safety function. In every one, a plate is the answer to the question being asked. The failure mode that produces wrongful stops and jailed innocents is not using LPR for access control. It is treating a raw plate hit as probable cause and acting on it without verification, or expecting a plate reader to protect a space it was never designed to watch.
How AI behavior detection closes the gap on your existing cameras
AI behavior detection covers the events LPR cannot see, and it does so without ripping out your infrastructure. Instead of reading text on a plate, computer vision analyzes the live video your cameras already produce and flags defined threats as they unfold: a visible firearm, a fall, unauthorized access, loitering, crowding, a vehicle in a restricted zone, smoke or fire, and movement across a perimeter or fence line after hours. When the system identifies one of these, it alerts your team within seconds so a human can verify and respond while there is still time to act.
This is the same shift from passive to proactive that separates AI security cameras from traditional CCTV. A plate reader and a passive recorder share the same limitation: both are excellent at telling you what happened, and useless at telling you what is happening. Behavior detection changes the tense. It turns a wall of screens no one can watch continuously into a system that raises its hand the moment something matters.
The practical and legal stakes are real. Passive systems that only record are increasingly cited in negligent security lawsuits, where a plaintiff argues that a property owner had cameras, captured the incident, and did nothing to prevent it. A camera that reads plates but cannot recognize a weapon or a collapse does not close that gap. A system that detects the threat and alerts staff in real time is a materially different posture, both operationally and in front of a jury.
Key Takeaways
- License plate recognition is an identity and forensic tool. It records which vehicle was present, not whether a threat is happening.
- Real-world accuracy trails the marketing. Even at a vendor-claimed 93 percent, the volume of scans produces more than a billion inaccurate reads per month, and misreads have led to at least 26 documented wrongful stops of innocent people.
- A plate reader cannot see a weapon, a fall, a fire, or a fence breach. Those events require behavior detection, not character recognition.
- LPR is the right tool for access, tolling, and post-incident investigation, as long as no one treats a raw hit as proof.
- AI behavior detection layers onto your existing cameras, uses no facial recognition, and alerts staff within seconds, covering the threats a plate reader structurally cannot.
Frequently asked questions about license plate recognition
Does license plate recognition detect crimes or threats?
No. License plate recognition reads and records plate characters and matches them against a database. It does not analyze behavior, so it cannot detect a visible weapon, a fall, a fire, loitering, or a perimeter breach. Detecting those events requires AI behavior detection, which analyzes what people and objects are doing rather than reading text on a vehicle.
How accurate is license plate recognition?
Vendors often advertise better than 90 percent accuracy under controlled conditions, but real-world results are lower and vary with lighting, weather, plate condition, speed, and angle. Even at a commonly cited 93 percent, the sheer volume of scans produces over a billion inaccurate reads per month industry-wide, and character misreads such as O for 0 have led to documented wrongful stops.
Is license plate recognition the same as facial recognition?
No. License plate recognition identifies vehicles by their plates and can link a plate to a registered owner through motor-vehicle records. It does not identify people by their faces. AI behavior detection is different again: it flags threatening behavior on camera without using facial recognition or any biometric identity at all.
Can I add threat detection to the cameras I already use for LPR?
Yes. AI behavior detection is a software layer that runs on standard existing cameras, so a facility can keep using license plate recognition for access and gates while adding real-time detection of weapons, falls, fire, and perimeter intrusion on the same infrastructure. The two technologies solve different problems and work well side by side.
Why do license plate readers cause wrongful stops?
Wrongful stops happen through two paths: machine errors, where the camera misreads a character or is fooled by dirt or a plate cover, and human errors, where an operator enters wrong data or acts on an unverified hit. An Institute for Justice review found roughly one-third of documented cases stemmed from machine error and two-thirds from human error, underscoring that a plate hit should always be verified before anyone acts on it.
Your cameras can already read a plate. Make them recognize a weapon, a fall, or a breach in real time, on the hardware you own, with no facial recognition. See how IntelliSee turns passive cameras into proactive protection.
Talk to Our Team