License plate reader accuracy is not one number: capture rate (vehicles detected), read accuracy (plates read correctly) and alert accuracy (hotlist alerts that are correct) differ, and conditions such as night, rain, glare and plate condition change all three. Public agency reviews have reported high alert error rates in some deployments, so every agency should measure accuracy in a pilot on its own roads.
License plate reader accuracy means three different things
“How accurate is it?” has at least three different answers, and vendors rarely say which one they mean.
| What is measured | Question it answers | Why it matters |
|---|---|---|
| Capture or detection rate | Of the vehicles that passed, how many did the system detect? | A missed vehicle is a missed lead. |
| Read accuracy | Of the plates detected, how many were read correctly, character for character? | A wrong character can point to the wrong vehicle. |
| Alert accuracy | Of the hotlist alerts raised, how many were correct matches? | Wrong alerts cost officer time and can lead to wrongful stops. |
A vendor’s headline percentage often reports the first or second. Agencies live with the third. Ask which one a number describes and under what lighting, speed, distance and weather.
What public reviews have found
We did not find a peer-reviewed, independent accuracy study of current systems. What exists publicly are agency reviews of alerts, reported by news outlets. Two examples, both involving Flock Safety cameras and reported secondhand:
- Los Angeles Police Department: the department’s inspector general reportedly found a 32.3% error rate for stolen-vehicle alerts over a two-month period, with 161 false alerts.
- Roseville, California: police reportedly reviewed 1,427 alerts from 2023 and 2024 for vehicles flagged as stolen or linked to felonies and found misreads in 71% of them. That measures the share of reviewed alerts that were wrong, not how often cameras misread plates overall. Reports say none of the false alerts led to stops, because officers verify each one, and the vendor attributed the errors to the city’s setup.
The vendor says its cameras accurately capture 93% of the plates they scan. We have not verified these figures, and they come from news reports (Carscoops and SAN) that you should read for yourself. The lesson is general: capture accuracy and alert accuracy are different, and every agency should measure its own.
What causes misreads
Most misreads come from a small set of causes:
- Image quality: blur from speed or vibration, low resolution, poor focus or a dirty lens.
- Obstructions: tree branches, plate frames, tow hitches, bike racks and stickers.
- Plate condition: bent, faded, dirty or non-standard plates, and unusual fonts.
- Similar characters: 0 and O, 8 and B, 1 and I, 5 and S.
- Wrong state: a correct string read as a plate from the wrong state is a different vehicle.
- Camera placement: poor angle, distance or aim for the lane being watched.
- Hotlist quality: an accurate read of a stale or wrong entry still produces a false alert.
Night, rain and glare
Readers work at night and in rain using infrared illumination and fast shutter speeds, but results vary by site. Glare from headlights and sun, heavy rain or snow, fog and a dirty lens all reduce accuracy. A solar-powered camera also depends on its battery carrying the unit through the night and through cloudy spells; see solar-powered license plate readers for what to check.
Ask any vendor, including us, for real sample reads from nighttime and rain deployments at sites like yours, not only daytime demos. Intelvi AI records continuous video next to every read, so a reviewer can check a read against the footage and see what the camera actually saw.
How to test accuracy in a pilot
| Metric | How to measure it in a pilot |
|---|---|
| Capture rate | Count vehicles in a sample of footage by hand and compare with the vehicles the system detected, by lighting and weather. |
| Read accuracy | Compare the system’s reads for the same sample with the plates visible in the footage. Record correct, misread and missed, and the error types. |
| Alert accuracy | For every alert during the pilot, record whether it was a true match after verification. Divide true matches by total alerts. |
| Time to alert | Time from detection to a notification on an officer’s phone or dispatch console. |
| Uptime | Hours each camera was online versus expected, including after bad weather. |
| By condition | Report all of the above for day, night, rain and the hardest site, not just the average. |
Frequently asked questions
What does license plate reader accuracy mean?
Three different things: capture rate (vehicles detected), read accuracy (plates read correctly) and alert accuracy (hotlist alerts that are correct). Ask a vendor which one a number describes.
How accurate are license plate readers?
It depends on the measure, the conditions and the system. Vendor figures describe capture or read accuracy under their test conditions; agencies experience alert accuracy. There is no standard benchmark, so test on your own roads.
Do license plate readers work at night and in the rain?
Yes, using infrared illumination, but performance varies with glare, weather and site. Ask for sample reads from night and rain deployments at similar sites.
What causes a license plate reader to misread a plate?
Blur, obstructions, damaged or non-standard plates, similar characters, the wrong state, poor camera placement and stale hotlists.
What is a good false-positive rate?
There is no industry standard. Set an acceptable alert-accuracy threshold before the pilot, measure against it, and require human verification of every hit regardless.
Does continuous video help with accuracy?
It does not change the read itself, but it lets reviewers check any read or alert against the footage, which makes errors easier to find and fix. See how AI plate recognition works.