A road image can reveal a possible pothole, crack, patch or deformation. By itself, it does not reveal the cause of the defect, its depth, damage to lower pavement layers or the appropriate intervention. That distinction defines a responsible use of artificial intelligence in road engineering.
When configured for a municipal project, computer vision helps locate and organize observations made along a route. The result is a queue of records for triage and review, not an asphalt diagnosis. Inspection, measurement, technical reports and maintenance decisions remain with the qualified professionals and public authorities.
What is AI-assisted road defect detection?
It is the use of images and location context to flag visual patterns that may match categories defined for a project. These categories may include apparent cavities, cracks, patches, surface wear and visible deformation. Availability of each category depends on the camera, capture position, road condition and acceptance testing.
A record can associate the image, location, route reference, suggested category and review status. These elements reduce manual searching and allow different passes to be compared. The image nevertheless remains a surface observation. Determining the origin of a failure or sizing repair work may require an on-site inspection, instruments and specific tests.
The route defines survey coverage
Mobile enforcement vehicles already travel through areas defined by their operation. When the pavement-observation module is contracted, configured and authorized, the same route may serve as a capture channel. This makes use of a pass that would already occur, but it does not make data collection free. Cameras, processing, connectivity, storage, review, integration and support remain project costs.
Coverage corresponds to the segments actually driven and to usable images obtained during those passes. A route does not represent the entire road network. Streets outside the route, obstructed lanes, segments with poor visibility and unusable frames must appear as gaps, not as areas with no defects.
Light, shadow, rain, speed, occlusion, pavement texture and previous repairs affect the reading. Pass frequency and minimum capture conditions must therefore be defined before deployment.
From capture to technical triage
A municipal workflow may be organized into five stages:
- Route capture: the vehicle records images along the segments driven for the authorized purpose.
- Context: the image, location, route reference and suggested category become part of the same record.
- Organization: observations may be grouped to reduce duplicates and highlight frames that require review.
- Human review: the team confirms or rejects the suggestion, adjusts the category and decides whether an additional inspection is needed.
- Routing: after review, the record may be exported or sent to a compatible municipal system.
This workflow does not create a work order by default. Creation, priority, ownership and closure depend on the authority's rules, available interfaces and agreed responsibilities.
How DNIT references fit into a project
Brazilian standard DNIT 005/2003-TER organizes terminology for defects in flexible and semi-rigid pavements [1]. It is a useful reference for building a category dictionary and aligning communication between technology and engineering teams.
DNIT 006/2003-PRO and DNIT 008/2003-PRO describe objective surface evaluation and continuous visual survey procedures, each with its own scope, method and requirements [2][3]. Using terminology from those documents does not turn a visual software classification into a DNIT-compliant survey.
If a municipality needs to use a normative procedure in a report, contract or engineering decision, the project must demonstrate how sampling, personnel, instruments, measurements and documentation meet the applicable method. One tool alone cannot guarantee compliance.
Quality must be measured on the actual route
No single accuracy percentage is true for every city, camera and category. Acceptance must use samples representative of the routes and conditions in which the system will operate.
The evaluation protocol should define at least:
- included categories and examples that are outside scope;
- who prepares and reviews the reference sample;
- lighting, weather, speed and image-quality conditions;
- criteria for correct findings, false alerts, missed occurrences and duplicates;
- accepted location error and treatment of records without a reliable position;
- limits by category and the procedure for revalidation during operations.
Results should be published with their method, sample and boundaries. A single average may hide the fact that one category performs adequately while another still requires adjustment.
Integration with pavement management
A reviewed record can feed maps, reports or management systems when compatible formats and interfaces are available. Integration must define fields, credentials, duplicate handling, status updates and ownership for each stage. Generic compatibility with every engineering package should not be assumed.
Indexes such as IGG, PCI or IRI have their own methods and inputs. A visual observation may contribute part of the data, but it does not calculate those indexes by itself or replace required measurements. The same applies to material volume, repair costs and technical solutions: a surface image is not a works budget.
Within a municipal maintenance workflow, integration may route a reviewed record to triage or a work order. The municipality defines which stages require human validation, which systems participate and how status updates will be tracked.
What to specify in procurement
A sound procurement process starts with the decision the data needs to support. Before discussing models or dashboards, the requirements should clarify:
- routes, segments and minimum capture frequency;
- defect categories and required evidence;
- minimum image and location quality;
- sample, metrics and acceptance limits;
- human review, responsibilities and need for inspection;
- exports, integrations and status return;
- access, security, retention and disposal of records.
These criteria allow proposals to be compared and performance to be checked in the municipal context. Generic performance labels cannot replace an acceptance contract.
Where each Areatec solution fits
Olho Vivo Patrol is the complete mobile capture and enforcement vehicle. The pothole detection page details specialist pavement observation. The Urban Maintenance platform receives, reviews, routes and tracks different categories of municipal occurrences.
Separating these responsibilities prevents the wrong promise: the vehicle observes, the module organizes indications and municipal management decides what to do with each record.
Frequently asked questions
Does artificial intelligence replace an engineering inspection?
No. It can expand observation and organize indications. Confirmation, measurement, diagnosis, technical reporting and definition of the intervention remain professional responsibilities.
Does the system create work orders by itself?
Not by default. A reviewed record may be routed when a compatible integration and municipal rules are in place. Creation and priority follow the authority's procedure.
Does the vehicle cover every street in a city?
No. Coverage corresponds to the areas actually driven by the routes and to the usable images obtained. Gaps must be recorded and considered in planning.
What is the detection accuracy?
It varies by category, camera, route and capture condition. The relevant result is the one measured in an acceptance test representative of the project, with the method and boundaries disclosed.
Does using DNIT categories guarantee compliance with its procedures?
No. The terminology may guide the class dictionary. Procedural compliance depends on meeting the applicable scope, method, sampling, measurements and responsibilities.
References
- BRAZILIAN NATIONAL DEPARTMENT OF TRANSPORT INFRASTRUCTURE. DNIT 005/2003-TER: Defects in flexible and semi-rigid pavements, terminology. Available at: DNIT 005/2003-TER.
- BRAZILIAN NATIONAL DEPARTMENT OF TRANSPORT INFRASTRUCTURE. DNIT 006/2003-PRO: Objective evaluation of flexible and semi-rigid pavement surfaces. Available at: DNIT 006/2003-PRO.
- BRAZILIAN NATIONAL DEPARTMENT OF TRANSPORT INFRASTRUCTURE. DNIT 008/2003-PRO: Continuous visual survey for evaluating flexible and semi-rigid pavement surfaces. Available at: DNIT 008/2003-PRO.
