How Does AI Detect Road Cracks Before They Become Potholes?

Every pothole has a backstory. Long before there's a hole in the road, there's a crack  thin, easy to miss, and usually ignored because it doesn't look urgent. That's the exact window where the cheapest possible fix is available, and it's also the window that most road maintenance programs miss entirely. By the time a crack is reported, it's often already a pothole.

AI-based crack detection exists to close that gap. Instead of waiting for cracking to fail into a pothole, computer vision systems can scan road surfaces continuously, catch cracking at its earliest visible stage, and route it into a repair queue while the fix is still a quick seal job rather than a full-depth patch.

This piece explains how a crack actually becomes a pothole, how AI identifies and classifies cracking before that happens, and why catching this early stage is one of the highest-leverage things a road maintenance program can do.

‍

How Does a Crack Actually Turn Into a Pothole?

A pothole doesn't appear out of nowhere. It's the final stage of a process that usually starts small and progresses predictably if left unaddressed.

• A surface crack forms. Traffic loading, thermal expansion, or aging asphalt creates a hairline crack in the wearing course.

• Water gets in. Rain seeps through the crack into the layers below, weakening the base and sub-base that the surface depends on for support.

• Traffic keeps stressing the weak point. Every vehicle passing over the softened area widens the crack and deepens the damage underneath.

• Freeze-thaw cycles accelerate the failure. In colder climates, water trapped in the crack freezes, expands, and forces the asphalt apart  repeating with every temperature swing.

• The surface collapses. The weakened chunk of asphalt breaks free under load, and what was a crack is now a pothole.

The critical insight here is that steps 1 through 4 can take weeks or months and during nearly all of that window, the fix is cheap: a crack seal or surface treatment. Step 5 is where the cost jumps dramatically, requiring full-depth patching or worse. Catching the problem anywhere before step 5 is the entire game.

‍

Why Do Cracks Get Missed Until It's Too Late?

If catching cracks early is this valuable, the obvious question is why it doesn't happen more often. The answer comes down to how most road networks are actually inspected.

• Manual inspection can't keep up with the scale. A road agency responsible for hundreds or thousands of kilometers simply cannot walk or drive every stretch often enough to catch cracking while it's still minor.

• Cracks aren't visually dramatic. Inspectors and the public alike tend to notice potholes, not the thin lines that precede them  human attention is naturally drawn to more obvious damage.

• Inspection cycles are too infrequent. Many agencies survey a given road segment once every one to several years, which is far slower than the timeline over which a crack can progress into a pothole, especially in wet or freeze-thaw climates.

• There's no consistent scoring standard. Even when cracks are noticed, manual assessment varies by inspector  one person's "monitor later" is another's "seal now," making prioritization inconsistent across a network.

The result is a maintenance system structurally biased toward reacting to potholes rather than preventing them, simply because cracks are much harder to systematically catch in time.

‍

How Does AI Actually Detect and Classify Road Cracks?

AI-based crack detection follows the same underlying collection approach used for broader pavement and defect monitoring, but with models specifically tuned to identify cracking patterns at an early stage.

1. Collect: Capture Pavement Imagery at Driving Speed

Dashcams mounted on any regular vehicle patrol cars, maintenance trucks, buses, delivery fleets  capture GPS-tagged video of the road surface as part of routes already being driven. No dedicated survey vehicles or lane closures are required, which is what makes frequent, network-wide coverage practical in the first place.

2. Detect: Identify Cracks and Classify Their Type

Computer vision models scan each frame to identify pavement distress and classify it by crack type, since different crack patterns indicate different underlying problems:

• Longitudinal cracks — running parallel to the road, often from paving joint issues or aging
• Transverse cracks — running across the road, typically from thermal contraction
• Alligator (fatigue) cracking — interconnected cracks resembling a reptile's skin, usually indicating structural fatigue from repeated loading
• Block cracking — large, rectangular crack patterns from asphalt shrinkage
• Edge cracking — cracking near the road's edge, often linked to poor drainage or lack of shoulder support
• Reflective cracking — cracks that mirror joints or cracks in an underlying pavement layer

Identifying the type matters because alligator cracking, for example, signals a structural problem that a simple surface seal won't fix — while a single transverse crack might be a straightforward, low-cost repair.

3. Measure: Quantify Length, Width, and Density

For each identified crack, the AI measures:

• Width — a key indicator of how far the damage has progressed and whether water infiltration risk is high
• Length and extent — how much of the road segment is affected
• Density — how many cracks are clustered in one area, which often signals broader structural fatigue rather than an isolated defect
• Pattern progression — comparing against prior survey passes to see whether a crack is stable or actively widening

4. Score and Prioritize: Convert Detection Into Action

Every detected crack gets a severity score and exact GPS location, feeding into a prioritized maintenance list. This is where the real value shows up  instead of a raw list of defects, agencies get a ranked queue that separates "seal this within 90 days" from "this needs monitoring next survey" from "this is already progressing toward a pothole and needs attention now."

‍

What Does an Early-Stage Crack Severity Framework Look Like?

Why Does Catching Cracks Early Save So Much Money?

The cost curve for pavement repair is not linear  it's closer to exponential the longer damage is left unaddressed. Engineering cost estimates commonly show that deferred repair costs 3 to 5 times more than the same defect fixed early, and that multiplier only grows once a crack has become a full pothole requiring full-depth reconstruction.

This is the core economic argument for AI-based crack detection: it's not really about finding more damage  it's about finding the same damage earlier, when the fix is still cheap, so the multiplier never gets a chance to compound.

‍

How Does This Fit Into a Broader Predictive Maintenance Strategy?

Crack detection is most powerful when it's not a one-time survey but a repeating monitoring cycle:

1. Baseline survey establishes current crack locations and severity across the network.
2. Repeat surveys — monthly or quarterly on high-priority corridors  track whether existing cracks are stable or progressing.
3. Trend alerts flag stretches where cracking is worsening faster than expected, prompting investigation into underlying causes like drainage or base failure.
4. Verified repairs — a follow-up survey after a crack is sealed confirms the fix held, closing the loop between detection and resolution.

This turns crack monitoring from a snapshot into a trend line  which is what actually lets an agency get ahead of pothole formation instead of just detecting cracks slightly earlier and still reacting to each one individually.

‍

How RoadVision AI Helps Cities Catch Cracks Before They Become Potholes

RoadVision AI's Collect–Process–Execute–Deliver pipeline applies the same dashcam-based survey approach used for pothole detection to earlier-stage crack monitoring. RoadGPT, RoadVision AI's vision-language model, reviews every frame of footage to identify and classify cracking by type, measure severity, and geotag each finding  all from footage captured by vehicles already on the road, with no dedicated survey fleet or lane closures required.

The result is a live GIS dashboard showing crack progression across the network, an auto-generated maintenance plan that separates urgent structural cracking from routine sealing candidates, and trend tracking that flags stretches deteriorating faster than expected. When a crack is sealed, the next survey pass automatically confirms the repair held  giving agencies a genuinely closed loop from early detection through verified resolution.

RoadVision AI has surveyed more than 3 million kilometers of road globally and detects 65+ pavement defect types across deployments in 17+ countries  helping road agencies catch damage while it's still a crack, not after it's already a pothole.

Want to see how much cheaper prevention is than repair? Get in touch with RoadVision AI to learn how AI-based crack detection can help your road network catch problems before they become expensive ones.

‍

Frequently Asked Questions

1. How does AI tell the difference between a minor crack and a serious one?

‍AI models measure crack width, length, density, and pattern type, then compare these against known severity thresholds. Interconnected patterns like alligator cracking are flagged as more serious than an isolated hairline crack, since they usually indicate deeper structural fatigue.

2. Can AI really catch cracks before they turn into potholes?

‍Yes — that's the core value proposition. Since cracking is the precursor stage to pothole formation, continuous AI monitoring can flag a crack while it's still in the cheap-to-fix window, well before water infiltration and traffic loading cause it to collapse into a pothole.

3. What equipment does a city need to run AI crack detection?

‍Typically just a dashcam mounted on vehicles already in use  patrol cars, maintenance trucks, or buses  or a smartphone-based collection app. No specialized survey vehicles or road closures are needed.

4. How often should road segments be surveyed to catch cracks early?

‍This depends on traffic volume and climate, but high-priority corridors benefit from monthly or quarterly surveys, since cracking can progress from minor to severe within a single season, especially in freeze-thaw climates.

5. Does crack type actually matter, or is a crack just a crack?

‍Crack type matters significantly. A transverse crack from thermal contraction may just need sealing, while alligator cracking often signals structural fatigue that requires a deeper fix. Treating all cracks the same risks under-fixing serious structural problems.

‍

RoadVision AI is an AI-powered road infrastructure monitoring and engineering platform helping road agencies and city governments worldwide detect, prioritize, and verify road maintenance work  from early-stage cracking to potholes and pavement markings. Learn more at roadvision.ai.

‍