How does license plate recognition work?
Plate recognition is a pipeline of steps. First, the camera captures an image of a moving or stopped vehicle. Software then locates the rectangular plate within the frame, corrects for the viewing angle, and separates the characters. A neural network or an older pattern matching method classifies each character and assembles a string, such as the letters and numbers on the plate. Many systems also record the vehicle image, a close crop of the plate, the time and the camera name.
Along with the text, good systems return a confidence score and sometimes the region or format. That score matters because recognition is probabilistic, not certain. Characters that look alike, such as the digit zero and the letter O, or eight and B, are common confusions, and dirt, bent plates, trailer hitches and unusual fonts add errors. Treat a read as strong evidence to check, not as proof, particularly before any action that affects a person.
Why do LPR cameras need a different setup?
A general surveillance camera is built to see a wide scene. An LPR camera is built to read a small, fast target. It uses a narrow field of view so the plate fills enough pixels across its width, a fast shutter so motion does not blur the characters, and a controlled exposure so reflective plate paint does not wash out. A useful relationship: the pixels across a plate equal the plate's real width divided by the width of scene that one pixel covers, so a tighter view gives more pixels on the plate.
Plates are retroreflective, designed to bounce light straight back toward its source. At night an infrared or white light illuminator beside the camera makes the plate bright, but if the exposure is too long the plate turns into a white blob and the characters vanish. LPR cameras therefore use short exposures and active illumination tuned to the plate. That is why a camera that looks fine for watching a lot can fail to read a single plate at the entrance.
Angle, distance and shutter speed: capturing a readable plate
Three things dominate capture quality. The first is angle: the further the camera is to the side or above the lane, the more the plate is skewed and the more the characters are squeezed. A modest angle is workable, a steep one is not. The second is distance and focal length, which set how many pixels fall across the plate. The third is motion: blur in the image is roughly the vehicle's speed multiplied by the exposure time, so faster traffic needs shorter exposures.
The practical response is to choose a dedicated capture point: a lane or gate where vehicles slow down, pass along a predictable path and face the camera at a manageable angle. Place the camera so headlights are not pointing into the lens, and so the plate is lit by the camera's own illuminator rather than by changing street lighting. Test at night and at the fastest speed you expect, then adjust. Capturing plates across a wide, open lot is much harder than at a controlled entrance.
Watchlists, allowlists and what you do with a plate read
A plate text is useful when it is compared with a list. An allowlist lets known vehicles through a gate or flags that a recognized vehicle has arrived. A watchlist, sometimes called a hotlist, raises an alert when a listed plate is seen. Logs of reads also allow searching afterwards: if an incident occurs, investigators can look for a plate or partial plate across a time range and retrieve the matching video.
Think about how reads connect to people and processes. Alerts should go to someone who knows what to do with them, and the reviewer should confirm the plate image against the text before acting. Partial matches, plates from other regions and temporary paper plates complicate matching. Open interfaces let reads feed gate controllers, parking systems and incident records, which is often where the real value comes from, more than from the read itself.
Privacy, legal and retention questions for LPR
Plate data identifies vehicles and, indirectly, the people who own or drive them, so it deserves careful handling. Laws on collecting, sharing and keeping plate data differ by country, state and local authority, and some places restrict private use or require signs and written policies. Confirm with counsel or the local authority what applies at your site before deploying, and write down who may view reads, who may export them and how long they are kept.
Good practice is to collect only what the purpose requires, set a retention period that matches it, restrict access by role, and keep an audit of who searched. Be clear in public-facing signage and policies about what you capture. Also plan for errors: a wrong read should never be the only basis for a decision about a person. Keeping the vehicle image alongside the text makes it possible to verify and correct.
How to choose an LPR camera or analytic
First decide whether you need a dedicated LPR camera or software on a general camera. Dedicated cameras are tuned for fast shutters, narrow views and illumination, and usually read more reliably at a gate or lane. Software on a general camera can work where vehicles are slow, close and well lit, but struggles at long distance or in glare. Ask where recognition runs: on the camera, on a local box or in the cloud, since that affects latency and what happens if the network drops.
Ask suppliers for a trial at your site, in your lighting, with the plate styles you expect. Check support for the plate formats in your region, how updates are handled, how reads and images are exported, and whether events can trigger gates, lights or alerts. Finally, compare the camera's installation needs, such as pole height, lane width and illumination, with what the site actually offers, and decide who will clean the lens and re-check alignment after storms, since a small shift in aim can quietly ruin reads.
Key terms
- LPR / ALPR
- License plate recognition, also called automatic license plate recognition: software that finds a plate in an image and converts its characters to text.
- OCR
- Optical character recognition, the step that classifies each character shape in the plate image as a letter or digit.
- Retroreflective
- A surface that returns light back toward its source, which makes plates very bright under a nearby illuminator.
- Confidence score
- A number the software attaches to a read showing how sure it is, useful for deciding which reads need human checking.
- Allowlist
- A list of approved plates the system can use to open a gate or mark a vehicle as expected.
- Watchlist
- A list of plates of interest that triggers an alert when one is read, also called a hotlist.
- Motion blur
- Smearing caused by a vehicle moving during the exposure, roughly its speed multiplied by the exposure time.
License plate recognition is available as an option on these trailers, not a standard feature. Because reads depend on a controlled angle, speed and light, the best results come from positioning the trailer so a lane or gate passes in front of the capture point, rather than relying on a wide scene camera. Video from the trailer is managed in AI VisionStream, and the operations center verifies alerts before they are escalated. Ask during scoping where vehicles will pass and how fast they move.
Common questions
Can any security camera read license plates?
Not reliably. Plate reading needs enough pixels across the plate, a fast shutter and controlled light. A wide scene camera often lacks these, especially at night or at speed. A dedicated LPR camera or well-placed narrow view at a lane gives far better results.
How accurate is license plate recognition?
It varies with camera placement, lighting, speed, plate condition and software. No system reads every plate every time. Look at confidence scores, keep the plate image with the text, and have a person confirm important reads before acting.
Is it legal to record license plates?
Rules differ by country, state and local authority, and may cover collection, sharing, signage and retention. Confirm with counsel or the local authority before you deploy, and set a written policy for who can view plate data and how long it is kept.