Even with high per-lane LPR accuracy, real-world conditions (glare, dirt, damage, angle variations, obstructions) introduce variability that results in inconsistent reads between entrance and exit:

And when matching fails:

• Sessions remain unresolved
• Vehicle counting becomes unreliable
• Manual review increases
• Revenue can be at risk

Why use the license plate digits only when you can use the whole vehicle?

LPR accuracy is measured per lane, but Parking performance is defined between entrance and exit, where sessions close, revenue is secured, and vehicles are counted.

​

​

​

Deepmatch maintains consistent vehicle identity across lanes by combining plate intelligence with additional vehicle-level signals such as:

• Multi-model plate recognition
• Type, Make and model detection
• Vehicle's image details (fingerprint)
• Temporal and directional context

This multi-signal data fusion enables DeepMatch to assign a unique vehicle ID. Even when the plate is partially hidden, damaged, or inconsistently captured between entrance and exit.

DeepMatch enhances

entrance-to-exit matching by:

• Reconciling partial or inconsistent reads
• Resolving mismatches caused by real-world visibility changes
• Assigning vehicle-level identity across detections
• Providing matching confidence scoring

This reduces mismatches and improves session continuity across lanes.

What information is received

DeepMatch is delivered through an API. Each time a vehicle is detected, an event is pushed, containing:

• A persistent vehicle transaction ID
• Plate information (plate, state, confidence rate, MMR, C)
• Matching confidence score
• Boosted plate number (result of the matching)
• Full events history