Vehicle Identification
CarScope
The vehicle’s identity, down to its DNA. All visible vehicle data — extracted instantly.
The problem
Every claim starts with a plate typed in by hand.
Someone opens the photographs, works out how many vehicles are in them, reads the plate off one and the VIN off another, then types both into the policy system.
Then they check that the car in the photographs is the car on the contract: right model, right trim, right VIN. It is minutes per file, so most organisations sample.
In and out
What the car is, read off the photographs.
- Vehicle photographs
- Ordinary phone photography, several angles; more than one vehicle per batch is fine.
- Close-ups
- A VIN close-up, a window, a tyre, the cluster: each adds its own fields.
- The vehicle, identified
- Make, model, year, body type, color and trim, with engine, transmission and equipment detail.
- Plate and VIN
- Plate and VIN taken from the picture, returned empty when they cannot be read cleanly.
- Parts, masked
- A polygon per part, attached to the vehicle it belongs to, retrievable image by image.
One clear image profiles the vehicle; a close-up of the VIN, the glass or a tyre adds the deeper reads.
Capabilities
Licence Plate and VIN Detection
Instantly capture and validate identification markers from images.
Car Profiling & Glass Scan
Identify make, model, trim, color, and scan windshield markings.
Exterior Part Segmentation
Pixel-accurate masking of over 30 distinct exterior body parts.
The hard part
An invented plate is worse than an empty one.
An unreadable plate
Asked to read a blurred plate, a system returns a plausible one. This one tests what it read against the shape a real plate takes, and discards what fails.
One read, one car
A plate in one photograph and a trim in another belong together only if the cars do. A vehicle that cannot be isolated is left unread.
Parts inside parts
A window sits inside a door, a light inside a bumper. A generic window is not an answer; the record names which door's window, resolved by shape.
Where a person decides
It removes the transcription. The judgement stays with your expert.
An empty field is an answer
An unreadable plate comes back empty. An empty field reaches a person to check; an invented one reaches a payment.
Nothing found, or not assessed
The record separates them and names the photographs it could not assess, so absence never reads as a clean result.
Every field stands alone
A tyre whose production date cannot be read still returns its size and tread depth: each field stands on its own.
Who it is for
Insurance
Every claim file identified the same way, so a plate or trim mismatch surfaces before assignment.
Fleet and rental
Mileage, tyre condition and panel-level detail captured at handover, from the photographs the branch already takes.
Inspection and remarketing
The listing spec sheet: model, year, color, trim, glass and tyre detail, assembled from photographs rather than typed.
How it fits
Identification can go live before anything else does.
Send the photographs
Submit a file's images and take back a standardised record, with no per-model configuration to maintain.
Match against your record
Compare the fields read from the photographs with the plate, VIN and model on file.
Route on confidence
Accept the clean identifications; send the empty and the borderline fields to the person who would have typed them anyway.
The same fields are what ArchDoc's document extraction is compared against, so a paperwork mismatch surfaces as a discrepancy.
Questions we get asked
What is CarScope?
CarScope is Archmir's vehicle identification workflow. It takes photographs of a vehicle and returns a structured record: plate and VIN where they are legible, make, model, year, body type, color and trim, windshield markings, and every visible exterior part masked and named.
What do I need to send?
Ordinary phone photographs. A single clear image already profiles the car; more angles, and close-ups of the VIN, a window or a tyre, add the identifiers and the deeper reads.
Does it replace the claims handler?
No. It removes the transcription pass, so a handler starts from filled fields instead of a folder of photographs. Where files clear automatically, that threshold is yours to set.
How do I know it is right?
Every value carries a confidence score, and an identifier that cannot be read cleanly comes back empty rather than filled in. We publish no accuracy figure; measure it on files you have already closed.
What does it not do?
Damage assessment is a separate workflow. CarScope will not guess an identifier it cannot read, and it reports the car in front of the camera, not its service history or its papers.
Where to start
Run a batch of files you have already closed and compare what comes back with what your team keyed in.
Initialize CarScope