A vehicle signature, not a plate read.
Plate recognition fails the moment a plate is missing, obscured, angled, or too low-resolution to resolve. Identify Car sidesteps that dependency entirely by working from the vehicle itself.
The model reads body lines, materials, trim, controls, and lighting across both interior and exterior views, then compares those features against a learned representation of vehicles to surface the closest matches.
Because the comparison is similarity-based rather than rule-based, a fragment of a frame still carries signal. A rear bumper corner, a wheel arch, a section of dashboard: each narrows the field.
A blurred corner of a bumper.
The input below is low-resolution, blurred, and shows no full-body view. There is no plate, no badge, and no complete silhouette.
Ranked candidates, operator judgment.
The workspace puts the source image beside the candidate set so an analyst can compare directly. The top-scoring candidate is not automatically the answer. In the reference run above, the operator selected candidate #8 after review.
That design reflects how vehicle identification actually works in casework. Similarity scores narrow the field; a human decides, and the reasoning stays visible in the record.
- Identified vehicle: make, model, and year for the selected candidate.
- Candidate set: ten options ranked by visual similarity.
- Match score: similarity confidence for each candidate.
- Selection: which candidate the operator confirmed, retained for the case record.
Identification plus location.
In the stolen vehicle case, a partial bumper photograph was run twice: Identify Car pulled make, model, and year, and Find Region placed the frame two counties over. The vehicle was recovered the same week.
Running both against one image is the common pattern. What the vehicle is narrows the lookup; where the photograph was taken narrows the search area.
Common questions.
Can it work from a partial or blurred image?
Yes. Partial interior and exterior signals are enough to produce ranked candidates: a bumper corner or a section of dashboard can carry the identification.
Does it read license plates?
No. It works from the appearance of the vehicle, which is what lets it run on frames with no visible or legible plate.
What does it return?
Make, model, and year as a ranked candidate list, each scored by visual similarity.
Can it be combined with location capabilities?
Yes, and it usually is. Pair it with Find Region or Find Street to get identity and location from the same frame.
