Computer vision in automotive service is the use of cameras and AI models to read a vehicle and its tires as data: a VIN, a license plate, a DOT code, a tire size, a tread depth across the full width of the tread. The camera captures an image. A model turns what is in that image into a value that lands in an inspection record.
Most topics on automotive computer vision describe vehicles being built or self-driving vehicles. This is about the years in between, when a vehicle comes into a bay, onto a lift, through a return lane or across a forecourt, and somebody has to write down what condition it is in. Anyline has spent more than ten years on that problem and passed 100 million digital tire inspections worldwide in a single year in 2025.
The three jobs computer vision does in a service bay
Computer vision does three different jobs during a vehicle inspection, and they have almost nothing in common except the camera.
| Job | What it reads | How you know it is right | What it feeds |
|---|---|---|---|
| Identification | VIN, license plate, DOT code, tire size, make and model | A check digit or a database confirms it | Vehicle history, recall status, correct part ordering, registration |
| Measurement | Tread depth across the tread width, per wheel position | Method and sampling density, because no registry of true values exists | Replacement decisions, service recommendations, quote lines |
| Condition indication | The shape of the wear across a tire, per position | The technician judges it against the vehicle in front of them | Alignment, inflation and suspension conversations |
Confusing the three is the most common error in this category, and it produces the two failure modes that get vendors into trouble: quoting identification speed as though it were measurement accuracy, and claiming a system diagnoses a fault when it has recorded a reading.
Computer Vision in Automotive: How It Reads a Vehicle in Service
Service-bay computer vision has to work against the chaotic environment, which makes it a harder engineering problem than it looks. In addition, a tire sidewall isn’t easy to read. The lettering is black rubber on black rubber, and the technician can be reading it at various angles.
Torn, wet, shiny, warped, obscured. Those five conditions arrive together on a sidewall, and the model still has to return one value a technician will act on. This is why our engineers train the network on thousands of damaged, obscured, and badly lit images rather than clean ones. Hence, our sidewall reader works across different lighting and weather conditions instead of only in the conditions it was demonstrated in. With TireBuddy, technicians can now simply point, align, and autocapture all the data from their smartphone.
AI-Powered Tire Sidewall Scanning for Instant Tire Identification
Capture every critical tire detail: make, model, size, or DOT production date in seconds, using any mobile device. Built for dealerships, fleets, and tire retailers that need fast, accurate tire data at scale.
How computer vision identifies a vehicle and its tires more efficiently
Identification is the most settled application of computer vision in automotive service, and the accuracy argument is already over. Cox Automotive reports that Anyline’s reading has reached above 99% accuracy, which is far more accurate than manual entry accuracy rates.
Speed genuinely matters here, because identification is transcription and transcription is the part of an inspection that adds nothing a customer can see. Anyline’s DOT reader captures the 17 black digits on the tire sidewall faster and more accurately than the human eye, and Discount Tire completes tire assessments typically in under five seconds per tire across almost 1,100 stores.
Success Story: Discount Tire
Anyline collaborates with Discount Tire to create an entirely new computer vision solution that could scan the tire DOT number in real-time with a phone camera.
The scale tells the same story elsewhere. RPM processes 60,000 vehicles monthly across 30 countries and 20,000 carriers, aiming to read 750,000 VIN numbers annually. NAPA digitized VIN capture across its service network. Continental uses mobile license plate reading to follow vehicles through their lifecycle.
What separates identification from everything else on this page is that the answer can be checked. A VIN carries a check digit, a plate resolves against a register, and a DOT code has a defined structure the reader can validate against. The system can tell you it is confident and be held to it.
How computer vision measures tread depth
Measurement has no external answer key, which is why the method matters more than the headline figure. TireBuddy has a precision of 1/32 inches or down to 0.5 mm. The model builds a three-dimensional tread model from around 100,000 points across the tire.
Before Discount Tire moved to digital tire inspections, its technicians manually measured 12 different spots to understand tread depth. Twelve readings, taken by hand, in a hurry, by different people, describe a tire far less completely than a full-width profile. Three readings barely describe it at all. Three is what a gauge realistically gets in a busy bay.
The gap between measuring and asking shows up in the field data. During National Tire Safety Week 2026, Anyline measured tires across the United States and found one in seven was unsafe. Field measurement produced that figure. Most published tire-safety statistics come from asking drivers what they believe about their tires, which measures confidence rather than tread, and the two rarely agree.
Our guide to what each tread number change covers includes which measurement to record and why.
What a full tread profile indicates about the rest of the vehicle
A tread depth reading answers a question about the tire. The shape of that reading across the tread width asks a question about the vehicle, and this is the part of computer vision in automotive service that is underused.
A single number can’t capture a shape. What matters is how the depth varies across the tire: if the center is worn deeper than the shoulders, that points to an inflation problem; if the shoulders are worn deeper than the center, that points to an alignment problem; and if one edge is worn more than the other, that points to a suspension problem. The real diagnosis isn’t in any one measurement, but in how the numbers differ across those 20 centimeters of rubber. Three gauge points cannot resolve a 0.5 mm gradient over that distance. A full-width profile can, and it can do it at every wheel position on the same visit, which is what turns four tires into a picture of one vehicle.
The tire is the only component on a car that keeps a physical record of what the rest of the car has been doing wrong. Reading that record properly is a diagnostic act, and it belongs to the technician. Anyline captures the profile and puts it on the record. The person standing at the vehicle decides what it means, orders the alignment check, and explains it to the customer.
Our guide to reading uneven wear covers the patterns and their causes, including where a wear signature has a root cause two steps upstream of the tire.
Where the reading has to land
A correct reading that never reaches a customer-facing record has produced nothing, and this is where most inspection intelligence projects actually stall. The measurement is solved. The handoff into the system the shop already runs is the part that gets underbuilt, because it looks like plumbing.
Anyline ships the TireBuddy ToolKit as an SDK for exactly this reason, so a reading appears inside the app a technician is already holding rather than in a second app nobody opens. André Ströttchen, Dealership Manager at Timmermanns GmbH, describes what it produces when the handoff works:
Currently, we use the Anyline application at all service stations for recording tire data during wheel changes. This provides us with a large database which is then used by our parts sales department for targeted marketing.
Nothing in that sentence is about cameras. The same measurement, taken the same way, at every station, until it becomes an asset the business can sell from.
Predictive maintenance needs a record before it needs a model
Predictive maintenance is the most over-promised application in this category, because prediction requires a history that almost nobody has assembled. No model forecasts when a tire will cross a threshold from a single visit. It needs the same vehicle, measured the same way, several times, with the wheel position recorded so the readings line up.
Building that history is a records problem long before it is a machine learning problem. The operators who will have useful prediction in three years are the ones taking consistent measurements today, and the boring part is the part that cannot be bought later.
Two ways to capture a vehicle in service
Vehicle capture in service has split into two hardware models, fixed arch and handheld device, and the split follows the moment of capture rather than the quality of the technology. In January 2026, Cox Automotive’s vAuto integrated UVeye, whose underbody, tire, and 360-degree exterior cameras are installed at hundreds of US dealerships, fleet sites, and auction lots. A vehicle drives through once and comes out with a condition report.
That works wherever every vehicle passes a single point under power: drive-through inspection lanes, auction entries, rental returns at a fixed gate. It works less well where a vehicle stops and never drives through. On a lift with the wheels off. At the curb. In a mobile van’s parking space. On a wholesale lot with no power drop. Those moments belong to the device the technician already carries, which is also the only option when the vehicle is in pieces.
Cox Automotive runs both, and that is the tell. The question was never which camera is better. It is which capture moment you are trying to instrument, and most operations have several.
The bottleneck with fixed-arch computer vision scanners
The fixed-arch tire scanner is not a small purchase. A drive-through installation means construction, cameras and lighting rigs mounted into a lane, sensors calibrated to a fixed geometry, often a dedicated structure built around the vehicle’s path. Costs run from the tens of thousands into six figures per lane, before the network, the maintenance contract, and the staff time to keep it running. That expense only pays back if enough vehicles pass through that one lane. A dealership with steady service-drive volume can justify it. A mobile inspection fleet, a wholesale lot with scattered arrivals, or a service bay that isn’t built around a single traffic pattern cannot.
The Advantage of Mobile Capture: Freedom from Fixed Arch Bottleneck
Mobile capture inverts the economics. The camera is a phone already in the technician’s pocket, and the “installation” is an app. There is no lane to build, no calibration rig, no fixed site to commit to before a single vehicle rolls through. The investment is recovered the first week it’s used, because there was barely an investment to recover. And the technology behind it isn’t new or unproven: it’s the same computer-vision tire analysis running on a fixed arch, repackaged for a handheld device that already exists in every technician’s hand. The capital cost was never in the imaging science. It was in the concrete, the steel, and the fixed point of capture — and mobile capture simply removes that part of the bill.
What computer vision in service still cannot do
The honest limits are worth stating, because vendor material rarely does. Three hold consistently.
The first is edge cases. Getting a model to handle the common case is fast. Getting it to handle the last fraction of real-world conditions takes far longer than the first version suggests, and every deployment finds its own long tail in its own bays.
The second is judgment. A system that captures a sidewall image has not assessed whether that tire is repairable, and no responsible vendor should suggest otherwise. Repairability depends on where the injury sits, how large it is, and what the manufacturer permits, and it is decided by a trained person with the tire demounted. The captured data supports the inspector process; however, it does not make the decisions on how to go forward.
The capture stack in a service bay
A service-bay computer vision system runs six stages, and knowing where each one lives explains most of the deployment questions buyers ask. Camera capture comes first, then preprocessing to correct for angle, glare, and motion. Detection and segmentation isolate the region that matters, whether that is a sidewall, a tread face or a plate. Classification or optical character recognition converts that region into a value.
A confidence step then decides whether the value is good enough to keep or whether the technician is asked to capture again. The record write puts it somewhere a person or a system will see it.
Where those stages run is a commercial decision as much as a technical one. Anyline processes on the device rather than in the cloud, which removes the network dependency in a basement parking garage and keeps images off a server. The company holds ISO/IEC 27001:2013 certification, and on-device processing is what makes GDPR compliance straightforward rather than contractual. It also means the six stages are complete on hardware the business already owns, on the phones and tablets in the building
Common questions about computer vision in vehicle inspection
What is computer vision in automotive service?
Computer vision in automotive service uses cameras and AI models to read a vehicle and its tires as data during an inspection. It captures identifiers that already exist on the vehicle, such as a VIN, a license plate, a DOT code or a tire size, and it measures things that have to be measured, such as tread depth. Each reading lands in the inspection record rather than in a technician’s memory.
What can computer vision read on a tire?
Four things, and they are three different jobs. Tread depth is a measurement. The DOT code, the tire size and the make and model markings are identifiers already printed on the sidewall. The shape of the wear across the tread is an indication of something happening elsewhere on the vehicle. Anyline captures all four and writes them to the record against the correct wheel position.
Is computer vision accurate enough to measure tire tread depth?
Anyline measures tread depth with precision down to 0.5 mm. That figure rests on sampling density, because the model builds a three-dimensional tread model from around 100,000 points across the tire. A handheld gauge realistically gets three or four readings on the same tire in a busy bay. No registry of true tread depths exists to check any measurement against, so ask a vendor how many points it samples before comparing accuracy claims.
What does uneven tire wear indicate?
Uneven wear points at inflation, alignment, or suspension, and which one depends on where the wear sits across the tread. A full-width profile shows that shape at every wheel position on a single visit, which a three-point gauge reading cannot do. The technician reads it against the vehicle and decides what happens next. The system records the profile and leaves the diagnosis to the person standing at the car.
You can see our full guide on tire wear here.
Does computer vision replace the technician?
No. It removes transcription, which is the part of an inspection nobody sees, and everybody pays for. Judgment stays with the technician: whether a tire is repairable, whether a wear pattern warrants an alignment check, what to recommend and how to explain it. What changes is that the judgment now sits on top of a consistent record taken the same way every time.
Does digital tire inspection need special hardware?
No. Anyline runs on the phones and tablets a business already owns, with a recommended device list covering recent iPhone, Samsung Galaxy, Pixel, and iPad models. Nothing gets mounted, powered, or calibrated in the bay, which is what makes the technology workable on a lift, at the curb, or in a mobile service van.
