Most cities still find potholes the slow way: a resident calls, a crew drives the route, and a defect that formed weeks earlier finally reaches a work order. AI pothole detection changes the first step. Cameras on vehicles capture the road, computer vision flags potholes and cracks, and each detection carries a location and a photo. The value comes only when those detections become prioritized repairs, which is where work order software fits into the process.
AI Pothole Detection for Cities: How It Works, From Camera to Repair
Understand how vehicle cameras and AI find road defects, and how those findings turn into geotagged, ranked work orders crews can complete.
The Problem With Finding Potholes by Complaint
Reactive reporting leaves cities with a biased and delayed picture of road condition.
- Reports cluster on busy roads and in neighborhoods that call more often, so quiet streets go unseen.
- Duplicate calls for the same hole waste dispatcher time, while other holes are never reported.
- Defects grow between report and repair, and a small pothole becomes a larger patch.
- Manual windshield surveys are slow, subjective, and hard to repeat on a schedule.
- Liability questions arise when a city cannot show when it learned of a defect or how it responded.
How AI Pothole Detection Actually Works
Systems differ by vendor, but most follow the same technical path.
Where the Cameras Can Come From
Data collection strategy shapes coverage, cost, and image quality.
| Capture option | Strength | Consideration |
|---|---|---|
| Cameras on refuse trucks or buses | Frequent repeat coverage of regular routes | Coverage follows route patterns, so side streets may be missed |
| Cameras on public works pickups | Flexible and easy to direct | Depends on drivers covering assigned areas |
| Dedicated survey vehicle | Consistent, high-quality network survey | Higher cost and periodic rather than continuous |
| Smartphone or dash camera apps | Low entry cost for pilots | Mounting, lighting, and image quality vary |
What Affects Detection Accuracy
AI does not remove the need for judgment. Plan for these factors.
Ask any vendor how accuracy was measured, on what roads, and how your own team can review a sample before relying on results.
Detection finds the defect. A work order gets it fixed.
Manage the repair side with assignments, status, photos, and history in Oxmaint.
From Detection to Prioritized Work Order
A list of thousands of detections is not a plan. Rank them with a simple risk matrix based on two questions: how severe is the defect and how exposed is the road?
Local policy decides actual response times. Bike lanes, bus routes, and school zones are common reasons to raise priority.
What a Good Pothole Work Order Contains
- Street address or coordinates and the asset segment it belongs to
- Detection photo and later completion photo
- Defect type, estimated size, and severity rating
- Priority and target completion date from the matrix
- Assigned crew, materials used, and labor time
- Link to any resident request so the requester can be updated
Routing Repairs So Crews Waste Less Time
Detection data lets planners group nearby defects into one trip instead of sending a truck for each report.
A Practical Pilot Plan
Metrics That Show Whether It Is Working
- Days from detection to completed repair
- Share of detections confirmed by staff
- Repeat repairs at the same location within a set period
- Repair cost per pothole, including labor, materials, and equipment
- Resident requests per month compared with prior periods
- Percentage of the street network surveyed on schedule
Where Oxmaint Fits in the Workflow
Oxmaint is the maintenance management layer. It does not perform image detection; it manages the work that follows once defects are known.
To see how a detection feed could connect, book a demo and bring a sample of your data.
Which Road Defects AI Can and Cannot See
Camera-based systems are strongest on surface defects. Structural and subsurface problems still need other methods.
| Defect type | How well cameras see it | What staff should still check |
|---|---|---|
| Potholes | Generally well, when lighting is good and the hole is not full of water | Depth, edge condition, and whether the base has failed |
| Longitudinal and transverse cracks | Often detected, with severity depending on image resolution | Crack width and whether sealing or patching is appropriate |
| Alligator cracking | Usually recognizable as a pattern | Underlying base or drainage problems that a patch will not fix |
| Patches and utility cuts | Detected, but may be confused with new defects | Condition of older patches and who owns the cut |
| Settlement and voids | Limited, since depth is hard to judge from a flat image | Field inspection, especially near storm drains and trenches |
Connecting Detections to Your Maintenance Records
A clean handoff prevents duplicate orders and lost context. Settle these rules before the first survey runs.
Privacy, Data Ownership, and Governance
Vehicle cameras record public streets, which means faces, license plates, and private property can appear in images. Handle that deliberately.
- Ask whether the system blurs faces and license plates automatically, and where the original images are stored.
- Write a retention rule for raw images and keep only what is needed for defects and repair records.
- Confirm in the contract that the city owns the data and can export it in common formats.
- Check public records and legal guidance on how survey images may be requested and released.
- Limit who can view full-resolution footage and log access.
- Review cybersecurity expectations for any vendor-hosted platform with your IT team.
Why it matters beyond compliance
Public trust affects whether residents accept vehicle cameras on city trucks and buses. A short, plain explanation of what is captured and why helps.
Cost Considerations Beyond the Software Fee
Compare the whole workflow, not only the detection price.
Seasonal Planning: Freeze-Thaw and Wet Weather
Pothole formation follows weather. Water enters cracks, freezes, expands, and leaves voids once traffic loads the weakened pavement.
- Survey before winter to find cracks that will let water in and seal what budget allows.
- Survey again after thaw, when new potholes appear quickly and crews face the largest backlog.
- In wet climates, repeat surveys after heavy rain seasons and compare against the previous pass.
- Stock patching material ahead of the peak and confirm cold-mix or hot-mix availability.
- Use temporary patches only as an interim step and schedule permanent repairs when conditions allow.
Using survey history
Repeat surveys show where defects return. Locations that fail again and again point to drainage or base problems, which a larger repair may solve for good.
Who Does What in a Detection-Driven Program
| Role | Responsibility | Output |
|---|---|---|
| Vehicle operators | Keep cameras clean, powered, and aimed correctly | Usable images on regular routes |
| Data reviewer | Confirm detections and set severity | Verified defect list |
| Maintenance planner | Rank, group, and schedule repairs | Prioritized work orders |
| Crew leader | Complete repairs and record materials and time | Closed orders with photos |
| Program manager | Track results, budget, and public communication | Quarterly performance report |
Common Mistakes to Avoid
- Buying detection before the city has repair capacity, which only grows the backlog.
- Skipping a validation sample, so nobody knows the real false positive rate.
- Treating every detection as equal instead of ranking by risk and road class.
- Leaving results in a vendor portal that maintenance crews never open.
- Failing to record the completion photo, which weakens the repair record if a claim arises.
- Surveying once and never repeating, which removes the trend value.
Questions to Ask Any Detection Vendor
Demos look polished. These questions expose what happens on your roads.
- How was detection accuracy measured, on which road types, and can we test a sample of our own streets?
- What are the typical false positive and missed defect rates, and how are they reported?
- Which camera models and mounting positions are supported on our vehicles?
- How are faces, license plates, and raw images handled, stored, and deleted?
- Can detections be exported with photo, coordinates, defect type, severity, and timestamp?
- Is there an open interface for sending findings to our maintenance system?
- What support exists for model tuning when local pavements differ from training data?
- What happens to our data if the contract ends?
Closing the Loop With Residents
Detection should not replace resident reports; it should work alongside them. Each resident request can be matched to a detected defect or a new inspection.
A Sample Twelve-Month Rollout
Beyond Potholes: Other Uses of the Same Survey Data
Once vehicles capture the road regularly, the same images can support other asset work, depending on what the platform detects.
- Pavement condition trends that inform resurfacing lists and capital planning.
- Faded markings, missing or damaged signs, and blocked drainage structures.
- Evidence of road condition on a given date for claims and public records requests.
- Before and after comparisons that show whether completed repairs are holding.
Keep scope realistic
Start with potholes and cracks, prove the repair workflow, then add other asset types one at a time so staff are not overloaded.
Frequently Asked Questions
Turn road data into repairs your crews can finish
Organize detected defects as prioritized work orders and keep a complete repair record with Oxmaint.
