Commission license-plate capture for the lane, speed, and night

Source facts: recognition quality begins before the algorithm

Axis Communications’ License plate capture white paper distinguishes license-plate capture—the production of a readable plate image—from license-plate recognition, where software locates and reads the characters. It states that recognition rate and accuracy depend strongly on the captured image and that a sophisticated algorithm cannot read a plate that is not clearly visible.

The source explains that license-plate work uses different installation and exposure choices than general surveillance. Camera-to-vehicle angle, lane width, capture distance, vehicle speed, field of view, shutter time, gain, infrared illumination, focus, and the plate’s size in pixels all affect the result. Axis recommends minimizing the total viewing angle, accounting for the time a vehicle remains in the capture zone, and limiting exposure time to reduce motion blur. At night, reflective plates and infrared light create a different exposure problem from the surrounding scene; excessive gain or poor illuminator placement can wash out the characters.

More image data is not automatically better for edge recognition. The source notes that enough pixels are required to resolve characters, but excessive resolution can increase analysis time and contribute to missed plates in dense traffic. Product and analytics instructions for the specific deployment remain controlling. The published values and examples are design guidance, not proof for every plate format, jurisdiction, lane, camera, or recognition engine.

DSE recommendation: define and test a capture envelope

For each lane, draw the region in which a valid read is expected. Record its near and far boundaries, lane width, permitted direction, expected speed range, vehicle types, camera angle, mounting height, lighting, and the action a recognized plate can request. A camera that reads a parked test car does not prove performance for a moving vehicle at night.

1. Separate capture from decision. First confirm that recorded frames contain a sharp, properly exposed plate. Then measure whether the recognition engine extracts the correct characters. Finally test any lookup, alert, or gate workflow. This makes it possible to locate a failure instead of blaming “the LPR.”
2. Use an authorized, varied test set. Include representative passenger vehicles, trucks where relevant, front and rear plates as applicable, clean and moderately weathered plates, and normal mounting variation. Record expected values securely and avoid collecting unnecessary plate data.
3. Drive the real approach. Test the low, normal, and highest approved speeds; expected lateral positions; vehicle following distance; stops and rolling approaches; and both travel directions if supported. Confirm that a vehicle cannot bypass the intended capture zone through an adjacent path.
4. Test day and night separately. Exercise direct sun, shade, headlights, wet pavement or other material reflections, artificial lighting, and infrared operation. Review glare, overexposure, motion blur, focus, and day/night switching at the plate—not only the overall scene.
5. Measure outcomes. Track total passes, plates captured readably, correct full reads, partial reads, wrong reads, duplicates, missed events, and time from capture to action. Preserve examples of each failure category so tuning remains evidence-based.
6. Exercise the safe exception path. An unreadable, ambiguous, expired, duplicated, or unlisted plate should not silently become authorized. Verify the approved manual review, alternate credential, intercom, denial, alert, and audit behavior.

Where a plate triggers a gate, test the complete sequence with the gate safety system and access policy: approach detection, read, authorization, open command, vehicle passage, close behavior, tailgating or second-vehicle scenario, loss of network, loss of analytics, and stale allowlist. Recognition is one input to the opening workflow; it does not replace the operator’s safety devices or the applicable gate requirements.

Save the accepted camera image settings, analytics version, region of interest, lane map, test evidence, exception rules, and change owner. Revalidate after camera movement, focus or firmware changes, pavement or lighting work, analytics updates, lane reconfiguration, or repeated error patterns. Protect plate records, allowlists, and exported test data according to the organization’s approved access, retention, and privacy practices. A dependable system proves the plate image, the read, and the action as three connected but separately testable stages.

Official reference

• Axis Communications, License plate capture, December 2024.