Automatic Road Extraction and Alert Generation for New Roads

Category: AI/ML + GIS (Smart City & Remote Sensing)
Difficulty: Advanced
Time to Build: 6–8 weeks
Prerequisites: Python, Jupyter, Docker, basic deep learning and GIS concepts
Deliverables: SRS, architecture diagram, training notebooks, Docker setup, test report, working prototype

Get Project Kit (Free) | Request Instructor Pack | Book 15-min Consult

Problem Statement & Expected Outcome

Problem Statement
Urban development and disaster response depend on accurate, up-to-date maps. Manual updates are slow and expensive. There is a need for an automated system that can detect newly constructed or modified roads from satellite imagery and instantly alert planners.

Expected Outcome
A deep-learning-based pipeline that:

  1. Extracts road networks from satellite images,
  2. Detects newly built or modified roads,
  3. Sends automated alerts and provides ready-to-use GIS outputs.

Abstract

This project deploys a computer-vision pipeline based on the open-source CRESI framework to detect and map roads from high-resolution satellite images. It compares current and historical imagery to identify new road segments and triggers real-time alerts through email, SMS, or webhooks. The entire workflow is containerized with Docker for reproducible training and deployment.

Details: User Stories & Acceptance Criteria
  • As a city planner, I can receive automated notifications when new roads appear in my jurisdiction.

  • As a disaster-response team member, I can quickly identify temporary access routes after floods or earthquakes.

  • Acceptance Criteria: New roads are detected with >90% precision and alerts are dispatched within one hour of imagery ingestion.

Scope & Modules

Module 1: Data Preparation

Pre-process satellite images, create masks, and generate folds for training.

Module 2: Model Training & Inference

Train deep-learning models (e.g., UNet/ResNet) inside Docker containers using prepared configs.

Module 3: Change Detection

Compare outputs from current and previous imagery to detect new or altered roads.

Module 4: Alert Generation

Send email/SMS/webhook alerts when road changes cross a configurable threshold.

Module 5: Visualization Dashboard

Display new and existing roads on a live map with export options to GeoJSON or Shapefile.

Proposed Architecture & Tech Stack
Recommended Stack
  • Core: Python, Jupyter Notebooks

  • Deep Learning: PyTorch / TensorFlow (as configured in CRESI)

  • Containerization: Docker (CPU or GPU mode)

  • Database: PostGIS or spatially enabled PostgreSQL

  • Visualization: Leaflet or Mapbox for interactive maps

  • Alerting: FastAPI with email/SMS/webhook integrations

Key Features
  • Fully containerized training and inference for reproducibility

  • High-precision road mask generation using semantic segmentation

  • Automatic stitching and skeletonization of extracted roads

  • Real-time change detection and alert dispatch

  • Export of detected networks to standard GIS formats

Milestones & Timeline
  • Week 1: Finalize SRS and collect initial satellite imagery

  • Week 2: Set up Docker environment and run baseline inference

  • Week 3: Prepare training data and run first training cycle

  • Week 4: Implement change detection and road skeletonization

  • Week 5: Develop alert generation module and integrate email/SMS/webhooks

  • Week 6: Build and test visualization dashboard

  • Week 7: Final testing with live imagery and accuracy tuning

  • Week 8: Documentation and final presentation

Who It’s For
  • Students & Capstone Teams: An impactful AI+GIS project for B.Tech/M.Tech theses.

  • Smart City Planners: Automated, low-latency road network updates.

  • Disaster Response Agencies: Rapid route detection in emergencies.

Progress Checklist

  • Docker container built and running
  • Training data prepared and masks generated
  • Model trained and validated
  • Change detection pipeline tested
  • Alerts sent and verified
  • Visualization dashboard deployed

Resources & Links

Download Project Kit

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Read More