License plate capture: pixels, angle and motion exposure
Design readable plate images before tuning recognition.
Updated
Principle and limits
Recognition depends on the captured characters. Define plate formats, lane width, capture distance, speed, viewing angle and night illumination. A general overview camera may not deliver suitable plate exposure. Avoid a universal plate-pixel threshold: character layouts, algorithm versions and capture conditions differ. Use the selected recognition supplier’s documented requirements, then verify them on representative regional plates.
Worked example
For a front-facing plate 0.52 m wide in a 4 m scene recorded at 1920 horizontal pixels, estimated plate width is 1920×0.52/4 ≈250 pixels. This assumes a perpendicular plane without significant distortion. For transverse motion, blur ≈ speed × exposure × pixel density. At 10 m/s, 1/500 s and 200 px/m, blur is about 4 pixels; 1/1000 s halves it and halves exposure time.
Installation and acceptance
Stage daytime, headlight, wet-road, queue and night cases at operational speeds. Record readable-image rate separately from correct-read rate and permitted-access rate. Include unreadable and partially hidden plates. Retain an overview association where required, and define manual handling rather than opening a barrier on uncertain recognition alone.
Define the denominator before claiming accuracy
Define eligibility in advance: countries, plate classes, lanes, speeds, direction and environmental conditions. Include motorcycles, damaged or dirty plates only according to the stated scope; log exclusions instead of silently dropping failures. A correct plate read can still lead to a wrong access decision when permissions or schedules are stale.
Night commissioning is a separate test
Focus and aim at the capture zone, then drive through it at representative speeds with dipped and bright headlights. Reflective plates can saturate while the surrounding vehicle looks dark. Adjust illumination and exposure using the plate crop from the actual recorded or analytic stream, not general scene brightness. A daytime success is not a night acceptance.
Measure plate pixel width, contrast, blur and clipping across near and far lane positions. Oblique views reduce useful plate sampling and can distort characters; required limits come from the selected capture and recognition system. Do not present one universal pixel count or mounting angle as guaranteed for every country and algorithm.