Home / How accurate is license plate recognition
LPR · AccuracyHow accurate is license plate recognition?
Very — when the camera is set up well. Almost every missed read traces back to one of ten causes, and most are fixable.
Free LPR software · one camera free for life · plate recognition included · about 100 MB
Accuracy is mostly decided before the software runs.
Modern deep-learning LPR reads the large majority of plates correctly when the camera is well placed. When people report poor accuracy, the cause is almost always the image the software was given — too small, too blurred, too angled or too bright — rather than the recognition itself.
What lowers read rates — and the fix.
| Cause | What you see | Fix |
|---|---|---|
| Too few pixels on the plate | Characters merge together | Zoom in so the plate is ~130–150 px wide |
| Motion blur | Smeared characters on moving cars | Fix the shutter at 1/500–1/1000 s or faster |
| Steep horizontal angle | Squashed, slanted plates | Stay within ~30° of straight-on |
| Camera mounted too high | Plates foreshortened | Mount at 1–2 m, tilt under ~30° |
| Headlight glare | White blur at night | Infrared + short exposure; read rear plates |
| No infrared | Dark, noisy night frames | IR camera or external IR illuminator |
| Backlight / sun | Silhouetted plates | Reposition or use wide dynamic range |
| Dirty or damaged plates | Missing characters | Syntax checking helps; nothing reads a missing letter |
| Unusual plate formats | Wrong characters on foreign plates | Use software trained on your countries |
| Smart codecs | Stream connects, frames glitch | Use standard H.264/H.265, not H.264+ |
How to measure your own read rate.
- Let the system run for a day.
- Count the vehicles in the recorded video.
- Count the plates read correctly in the log.
- Divide. Then change one thing — zoom, shutter, angle — and measure again.
Measuring on your own site beats any number on a brochure. Camera setup rules.
Accuracy: quick answers.
How accurate is license plate recognition?
With a well-positioned camera, modern software reads the large majority of plates correctly. Poor accuracy usually comes from image problems: too few pixels, blur, angle or glare.
Why does LPR misread 0 and O or 8 and B?
They look alike in small or blurred images. Good software checks the plate's expected layout to correct them, and more pixels on the plate removes most confusion.
Does a better camera improve accuracy?
Position and settings matter more than price. A mid-range camera zoomed onto the lane with a fast shutter beats an expensive one watching the whole yard.