LPR (Licence Plate Recognition), also called ANPR (Automatic Number Plate Recognition), reads vehicle plates from video and turns them into searchable text. The pipeline finds the plate in the frame, isolates the characters, and runs OCR to produce the alphanumeric string, which the VMS can then alert on, log, or match against a list. It powers car-park access, tolling, watchlists, and forensic "find this vehicle" search.
Its accuracy is best understood as two numbers multiplied: the capture rate (did the system get a usable image of the plate?) and the read rate (did it read the characters correctly?). Roughly 98% capture times 95% read gives about 93% end-to-end, and controlled single-lane setups reach 95–99% while fast, free-flow, multi-lane traffic does worse. It is never 100%, and it depends heavily on a dedicated camera: fast shutter to freeze motion, infrared for retroreflective plates at night, and a tight angle.
The legal framing is what distinguishes LPR from face recognition: a plate is personal data but generally not Article 9 biometric data, so the gate is not biometric-consent law but retention, access, and sharing rules — California's SB 34 (Civil Code §1798.90.5), the UK's ANPR regime under the Surveillance Camera Code and ICO guidance, and similar. The real risk is the stored movement profile that mass plate reads create. This is engineering guidance, not legal advice — confirm specifics with counsel; the OCR model internals belong to the AI for Video Engineering section.
