What problem does it solve?
A road network is surveyed by people driving or walking it, or by a specialist vehicle that visits every few years. Between surveys, the council learns about a pothole when a resident reports it or a crew happens to pass. That leaves the council reacting to complaints, with the defects that nobody reports still unrepaired.
The NSW Government describes the cost side: collecting data on asset condition is expensive and labor intensive, and the frequency of collection, which "can be every 1 to 5 years", means the data is limited and often out of date. Councils that already run vehicles on fixed routes, such as recycling trucks, can collect the footage as a by product of the round.
- The NSW Government says condition data on council assets is collected at a frequency that "can be every 1 to 5 years", which leaves the available data limited and often outdated.AssetAI: SmartNSW case study (2026)
How does it work?
- Capture. A camera fitted to a council vehicle records the road and footpath during its usual route. Moreton Bay pairs dashcams with AI on its recycling trucks, which cover the same routes every fortnight; Shellharbour fits a smart camera to the front of council vehicles.
- Protect privacy. The footage is processed so that people and number plates are not kept. The NSW Government's Asset AI pilot de identifies footage before it is analyzed; Moreton Bay says its system identifies defects "while protecting privacy".
- Detect and rate. AI finds potholes, cracks, uneven surfaces, faded line marking, damaged signs and overhanging vegetation. Shoalhaven ranks each defect with a score of zero to ten, and Shellharbour rates defects on a scale of severity with a GPS location and a photo.
- Create work. The defect enters the asset management system and becomes a work order. Shoalhaven sends the ranked data to an asset management team that dispatches field workers.
- Close the loop. Crews close the job with a photo of the completed work, and the next pass of the same route shows whether the defect is gone or has grown.
- Audience
- Back office
- Autonomy
- Copilot
- Adoption
- Early adopters
- Channels
- API and system to system
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Interactions handled | Not pooled | 2000 to 4547 | 2 | 2 organization |
Value drivers: Lower cost to serve, Speed and cycle time, Risk and loss reduction.
Indicative value
A council with 3,000 kilometers of local roads that fits cameras to 10 recycling trucks
USD 104,000 to USD 2.1 million
Avoided cost of late repairs per year
How this is calculated
Formula: defectsPerWeek * 52 * costPerLateRepairExtra. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Defects the AI detects per week that would otherwise be found late or not at all defectsPerWeek, defects per week | 100 | 500 | City of Moreton Bay repaired 4,547 AI identified potholes in six months, about 175 a week, and Shoalhaven reports about 2,000 defects a week across a 1,700 km network, counting potholes, branches, graffiti and signs. Only part of these would otherwise be found late, so the range sits near the Moreton Bay figure. Replace with your own network and defect types. |
| Extra cost of fixing a defect late rather than early costPerLateRepairExtra, USD per defect | 20 | 80 | Editorial assumption, replace with your own repair cost data. No source on this page gives it. |
What it leaves out: This is a worked illustration, not a saving any council has reported. It ignores the cost of the cameras, the software license and the extra repair crews the extra detections can require, and it treats every detection as a real defect that would have been missed.
Which organizations already do this?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
This library is not a list of Blits.ai customers. It documents public AI deployments from any vendor, each with its source.
Shellharbour City Council
Australia · Government and public sector · 2025
Shellharbour City Council, in New South Wales, front mounts a smart camera on council vehicles to photograph roads and pathways while they drive. An AI system detects potholes, cracks, uneven surfaces, litter, overhanging vegetation, faded line marking, damaged signage and graffiti, maps each defect with a GPS location and rates its severity so maintenance can be prioritized. The council plans a second camera to survey its 500 kilometer road network and 300 kilometers of pathways faster, and says early results show hazards identified faster. It publishes no count of defects or repairs.
No outcome disclosed.
City of Moreton Bay
Australia · Government and public sector · 2019
City of Moreton Bay, a council in Queensland, Australia, pairs dashcams on its kerbside collection trucks with AI that identifies potholes, cracking, line markings and signage. The council began the rollout in early 2019 and now uses it across all 14 recycling trucks, which drive the same routes every fortnight. In a February 2026 release it reports that 4,547 of the 12,896 potholes repaired in the previous six months were identified by the AI, and that the technology has detected 48,282 potholes since 2020. Detected defects enter the maintenance program and are prioritized by severity. The council also repaired 6,812 potholes found by its own inspections and 1,537 reported by customers in the same period.
- Interactions handled: 4547, potholes identified by AI among 12,896 repaired in the six months to February 2026
"From this total, 4,547 potholes were identified via artificial intelligence (AI) technology, which has helped to detect 48,282 potholes since 2020."
Claimed by: organization
Shoalhaven City Council
Australia · Government and public sector · 2023
Shoalhaven City Council, on the south coast of New South Wales, mounted cameras on six recycling trucks and used AI to scan the footage for cracks, potholes, graffiti, line mark deterioration, overhanging branches, debris and damaged street signs. Defects are ranked with a score of zero to ten and sent to the asset management team, which creates work orders for 76 field workers.
The council's Director of City Services says 87 percent of the 1,700 km road network can be assessed in two weeks and that the solution detected 2,000 defects a week in the first six months. The council turned to the trial after storms, floods and bushfires between 2019 and 2022 damaged its assets. TechnologyOne and Retina Visions supply the technology. The account comes from trade press quoting the council, not from a council report.
- Interactions handled: about 2000, defects detected per week in the first six months
"The solution detected 2,000 defects a week in the first six months"
Claimed by: organization
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- A road and footpath network register with segment identifiers
- GPS tagged video or images from vehicles on routine routes
- A defect classification and severity scale agreed with the maintenance team
Systems to integrate
- Asset management or work order system, such as the one the council already uses
- Vehicle fleet routes and schedules
- Privacy filtering of footage before storage
Complexity: Medium
The build needs cameras on vehicles that already drive the network, a place to process and store footage that meets privacy law, an asset management system that accepts defects as work orders, and a repair budget that can absorb the extra defects the survey finds.
- 1
Start with vehicles already on the network
Choose vehicles with regular routes, for example recycling trucks, because repeat passes show change over time. Moreton Bay and Shoalhaven both started with their recycling fleets.
- 2
Agree the defect list and severity scale before the model runs
Decide which defects create a work order and what each severity level means, so the AI ranks to the council's own standard and not the vendor's default.
- 3
Check detections against crew inspections on a sample
Have crews inspect a random sample of detections and of undetected stretches each month. None of the public sources on this page states a detection accuracy, so the council has to measure it.
- 4
Wire detections into work orders with a human triage step
A person confirms or dismisses each high severity item before a crew is sent, until the sample checks show the false positive rate is acceptable.
- 5
Plan the repair capacity first
A survey that finds more defects than the crews can fix creates a backlog and a liability. Set the repair budget and priorities before widening coverage.
Guardrails
- Footage is de identified before it is stored or shared, and retention is limited
- Every automated severity score can be overridden by an inspector
- Each detection keeps its photo and GPS location as evidence for later claims
- Coverage and detection checks are reported to the asset manager each month
KPIs to instrument
- Share of the network surveyed per month and the gap between passes
- Detections confirmed by a crew as a share of detections raised
- Days between detection and repair by severity
- Defects found by AI versus by inspection and by resident report
Human in the loop
An asset management team or inspector triages ranked defects and creates the work orders; Shoalhaven's account has the asset team creating work orders for its field workers from the ranked data. Residents can still report defects, and Moreton Bay encourages them to.
Common failure modes
- Detections outrun repair capacity
- A council finds thousands of defects and fixes a fraction, leaving a documented backlog. Match the survey rate to the repair budget.
- Model drift on new road types or lighting
- A model trained on one road surface or camera angle misses defects on another. Test coverage on each vehicle type and season.
- Privacy complaints stop the program
- Cameras on garbage trucks raise concerns about collecting images of people and property. Publish what is kept and for how long.
What are the risks and rules?
EU AI Act
Minimal risk
A system that finds road surface defects for maintenance planning is not listed in Annex III. Annex III point 2 covers safety components in the management and operation of road traffic, and a defect survey that feeds repair planning does not control traffic. Check this if the same system later steers traffic management or if its output is used to decide who is liable for a road injury.
Guidance
- Article 2, scope (European Union, Europe). Sets which AI systems the Act covers, as a first check on whether a road survey system is in scope.
Controls to put in place
- De identification of people and number plates before footage is stored
- Retention limits for raw footage and a documented lawful basis for processing
- A sample based accuracy check by crews each month
Frequently asked questions
- How many defects do councils find with AI?
City of Moreton Bay reports that 4,547 of 12,896 potholes repaired in six months were identified by AI, and 48,282 since 2020. Shoalhaven City Council says its system detected 2,000 defects a week in the first six months, counting more than potholes. These are the councils' own figures and say nothing about accuracy.
- Do the councils publish the accuracy of the detection?
Not in the sources reviewed. Moreton Bay, Shoalhaven and Shellharbour report volumes and qualitative benefits, and none states a precision, recall or false positive rate. A council adopting the approach should measure these against crew inspections.
- Does it need dedicated survey vehicles?
No. Moreton Bay and Shoalhaven use recycling trucks that already drive set routes, and Shellharbour fits a camera to council vehicles. Fixed routes give repeat passes, which show how a defect changes between visits.
- What about privacy?
Footage includes people and property. The NSW Government's Asset AI pilot de identifies footage before analysis, and Government News notes research on privacy issues of cameras on council garbage trucks. A council should set retention and masking rules before it starts.
How to cite this page
Blits.ai AI Use Case Library, "AI road surface defect detection from vehicle cameras", last verified 18 November 2026, https://www.blits.ai/ai-use-cases/road-surface-defect-detection. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 18 November 2026: First published