Make, model, and year from one image.
Upload a photo of a vehicle, even a partial or blurred one. Raven ranks candidate vehicles by visual similarity for your review. No plate or badge required.
Available to verified agencies. Demo on request.
- 01Start with one blurred frame. No plate or badge in view.
- 02Raven reads the vehicle itself: body lines, materials, and trim.
- 03It compares that signature against a learned representation of vehicles.
- 04The closest ten come back as candidates, ranked by visual similarity.
- 05All ten score within two points, so an investigator reviews the list against the frame.
- 06The investigator confirmed candidate 1, the top match: a 2022 Chevrolet Silverado 1500.
Illustration. The point cloud and outlines are drawn; the candidates, scores, and pick are from the result below.
The result, in Raven.
Identify Car ranked ten candidate vehicles within two points of one another. After review, the investigator confirmed the top match: a 2022 Chevrolet Silverado 1500, at 79% similarity.
What you get back.
Ranked candidates with the images behind each match. Raven presents options for review, not one automatic answer.
- Make, model, and year
- For each candidate, ranked by visual similarity to your frame.
- Similarity score
- Orders candidates against one another. It is not a probability that a candidate is correct.
- Reference images
- The image behind each candidate, to compare with your frame.
- Review state
- Confirmed, possible, or dismissed, recorded by the investigator for each candidate.
Identification plus location.
Run the same frame through Find Region. What the vehicle is narrows the lookup; where the photo was taken narrows the search area. In one stolen vehicle case, a partial bumper photo went through both, and the vehicle was recovered the same week.
Powered by Graylark's proprietary computer-vision models.
Common questions.
Can Identify Car work from a partial or blurred image?
Yes. The model reads partial visual signals across interior and exterior (body lines, materials, trim, controls, and lighting), so a corner of a bumper or a section of interior can be enough to produce ranked candidates.
Does it read license plates?
No. Identify Car works from the appearance of the vehicle itself rather than plate text, which is what allows it to operate on frames where no plate is visible or legible.
What does it return?
Make, model, and year as a ranked list of candidates, each scored by visual similarity, so an operator can weigh the options rather than accept a single guess.
Is the top-ranked candidate the identification?
Not automatically. Identify Car returns ranked candidates scored by visual similarity, not a definitive identification. In the worked example on this page, all ten candidates scored within two points, and the investigator confirmed the top match only after review. Results are investigative leads and are intended to be corroborated through normal case procedure before action is taken.
Can it be combined with location capabilities?
Yes. In one case study a partial bumper photograph was run through Identify Car for make, model, and year, and through Find Region to place it two counties over. The vehicle was recovered the same week.
For how vehicle identification began as a research side quest, read Why We Built GeoSpy and Why It Became Raven. To see where it fits in a stolen vehicle investigation, read Geolocating a Suspected Stolen Car From a Single Social Media Post.
See Identify Car in Raven.
Book a walkthrough with your team. Raven is available to verified agencies and investigative teams.
