← Back to project catalogue
GP-CV-0YOUGV4CivilOpen for request

Pavement distress assessment system

An image-based assessment system for locating, classifying, and measuring selected visible pavement distresses from prepared road imagery.

  • Nuxt
  • Vue
  • FastAPI
  • PostgreSQL
  • PostGIS
  • Python
  • PyTorch
  • OpenCV
  • Docker

Project definition

Problem statement

Manual pavement surveys require consistent definitions and can vary between inspectors, while image models are affected by shadows, markings, repairs, debris, camera angle, and missing scale.

The engineering problem is to define a limited distress catalogue, detect and measure visible defects, estimate severity under explicit calibration assumptions, and keep uncertain cases available for engineer review.

Project objectives

  • Prepare and annotate road images for selected crack, pothole, patch, and surface-defect classes.
  • Check image blur, exposure, viewing angle, location, and scale metadata before assessment.
  • Compare a visual baseline with a selected detection or segmentation model.
  • Estimate distress position, area or length, confidence, and documented severity category.
  • Aggregate reviewed observations by road section without hiding uncertain results.

System design

System modules

01

Survey data manager

Imports permitted road images and records route, chainage or location, camera, scale, direction, and quality.

02

Annotation workspace

Supports polygons, masks, classes, severity labels, reviewer agreement, and dataset splits.

03

Distress model

Detects or segments selected visible distress classes and preserves confidence and model version.

04

Measurement module

Uses provided calibration or reference geometry to estimate visible length, width, and area where supported.

05

Condition dashboard

Maps reviewed observations and shows class, severity, confidence, image evidence, and section summaries.

Methodology

System workflow

  1. 01
    Prepare survey

    Images are checked for permission, location, quality, overlap, and available scale information.

  2. 02
    Annotate

    Reviewers label selected distress classes under a written guide and resolve disagreements.

  3. 03
    Train and test

    The model uses road-section-separated splits to reduce location leakage.

  4. 04
    Measure and review

    Predictions are calibrated where possible and uncertain or low-quality observations are reviewed.

  5. 05
    Summarise section

    Accepted observations are mapped and aggregated using the documented condition method.

Demonstration scenario

A prepared road section is uploaded with calibrated images. The model marks cracks, a pothole, and an old patch. One shadow is rejected during review, while the pothole dimensions and severity are accepted. The map and section summary update with the retained evidence.

Engineering

Technical architecture

Web application
Nuxt and Vue for image review, annotation, map navigation, measurements, and section summaries.
Vision API
FastAPI for datasets, annotation, model jobs, predictions, measurements, reviews, and exports.
Spatial data layer
PostgreSQL and PostGIS for road sections, image locations, observations, geometries, models, and reviews.
Vision pipeline
Python, PyTorch, and OpenCV for quality checks, augmentation, detection or segmentation, and calibration.
Evaluation design
Road-section-separated tests, per-class metrics, measurement tolerances, and human-review agreement.

Testing

Evaluation

Evaluation measures

  • Per-class precision, recall, F1 score, and mean average precision or intersection over union
  • Performance on road sections excluded from training
  • Length, width, and area error where reference calibration is available
  • Severity agreement with prepared engineer-reviewed labels
  • Error rates under shadows, markings, repairs, blur, and different camera angles
  • Processing and review time per kilometre or prepared image set

System boundaries

  • The system assesses only visible surface conditions represented in the prepared dataset.
  • Image-based severity does not replace structural testing, deflection measurement, coring, or engineer inspection.
  • Dimensions require valid camera or reference calibration and are withheld when scale is unavailable.
  • The system does not automatically assign repair contracts, safety closures, or legal responsibility.

Included

  1. 01Road-image annotation and inspection interface
  2. 02Distress detection, measurement, severity, and mapping modules
  3. 03Location, confidence, error, and condition-summary dashboard
  4. 04Prepared road imagery, labels, experiments, and results
  5. 05Complete source code in a private GitHub repository
  6. 06Complete project documentation with synopsis, abstract, methodology, pavement diagrams, model results, screenshots, and conclusion
  7. 07Setup and usage guide

Project record

No buyer information is collected on this page.

Permanent project ID
GP-CV-0YOUGV4
Catalogued
21 Aug 2026
Completed
Pending
Verified
Pending
Demonstration
Added when ready

Handover

After purchase

  1. 01
    Payment is confirmed

    The project is marked unavailable and cannot be purchased again.

  2. 02
    Repository access is granted

    The buyer's submitted GitHub account receives access to the private repository.

  3. 03
    The purchase record is delivered

    The certification sheet is prepared from the reviewed buyer details and sent privately by email.