Improve License Plate Recognition-Based Entry/Exit Systems

Match entering and exiting vehicles with greater confidence, even when plates are hard to read. We use vehicle fingerprinting and contextual data to provide highly robust vehicle matching.

Vehicle Match Banner

Comprehensive Approach to Matching Vehicles

Go beyond low-quality LPR reads and fuzzy string matching. When plates don't match, we use a wide collection of signals to reliably match vehicles.

Comprehensive vehicle matching diagram

High Quality LPR

Difficult plates are re-processed with state-of-the-art AI (LPR cameras often use low-quality LPR due to resource constraints).

Vehicle Fingerprinting

Unique identity cues are encoded with learned visual representations (damage, stickers, accessories, mounts) for identification beyond license plates.

Vehicle Attributes

Type, color, make/brand, body style, and visible equipment are extracted to strengthen candidate ranking and resolve ambiguous reads.

Contextual Signals

Time windows and historical parking stay-times are incorporated to improve precision.

Confidence Scoring

A single scoring engine that takes into account all of the above factors produces a single, holistic matching score.

How it Works

1

Local LPR Read

A plate read is produced by local LPR on the edge.

2

Cloud Matching

The most likely match is determined by a cloud-matching algorithm. This includes additional processing to extract additional attributes that can be used to match vehicles.

3

Accept or Review

High-likelihood matches are accepted; ambiguous cases can be sent to local operators for manual review.

Measurable Impact

Higher matches, lower error rates vs naive matching techniques.

Match Rate

Before 90%
After 95%

Error Rate

Before 5%
After 1%