All Projects
Production

AI-Powered Road Infrastructure Monitoring System

Computer vision system mounted on waste trucks for automated city-wide road defect detection, enabling proactive infrastructure maintenance.

PyTorch Computer Vision Object Detection GCP Docker
AI-Powered Road Infrastructure Monitoring System

Problem

Road infrastructure deterioration poses significant safety risks to drivers and commuters. Traditional manual inspection methods are costly, time-consuming, and unable to provide consistent city-wide coverage. Councils often discover road defects reactively — after they’ve already caused damage or complaints — rather than proactively identifying and prioritizing repairs.

Solution

I designed and built an AI-powered road monitoring system that leverages the regular coverage patterns of waste collection trucks. By mounting cameras on these vehicles, the system captures imagery of every road in a council’s service area during routine collection rounds, then processes the data through a deep learning pipeline:

  • Object Detection Pipeline: Built with PyTorch, trained to identify and classify multiple types of road defects including potholes, cracking, edge deterioration, and surface deformation
  • Cloud Processing: Deployed on Google Cloud Platform with containerized inference services (Docker) for scalable batch processing of daily image captures
  • Geospatial Mapping: Each detected defect is geo-tagged and mapped, creating a comprehensive, continuously updated view of road conditions across the entire service area
  • Priority Scoring: Defects are ranked by severity and location to help councils allocate maintenance budgets effectively

Impact

  • Enabled automated, city-wide road condition monitoring using existing waste collection infrastructure — no additional vehicle deployments required
  • Transformed road maintenance from reactive to proactive, allowing councils to identify emerging issues before they become safety hazards
  • Provided councils with data-driven prioritization tools to optimize limited maintenance budgets

Technical Highlights

  • End-to-end ML pipeline: data ingestion, preprocessing, model inference, post-processing, and result delivery
  • Custom-trained object detection models optimized for varied lighting conditions, weather, and road surfaces
  • Scalable cloud architecture handling large volumes of daily image data
  • Integration with council GIS systems for seamless defect reporting