Extract

Digitising historical planning documents

The challenge

Traditional planning processes are significantly slowed by decades of planning information trapped in various formats. This includes paper maps, PDFs, handwritten notes, and blurry scanned documents. Planners currently take 1-2 hours to manually process each document. This prevents councils from making faster planning decisions.

The solution

Extract is an AI tool that combines multiple artificial intelligence technologies to digitise historical planning documents automatically. The system uses advanced language models including Google DeepMind's Gemini to read and extract text information, while computer vision technology including Meta's Segment Anything Model analyses images and maps. The tool works through a two-stage process: first extracting key details like dates, locations, and planning decisions from text, then using intelligent image analysis to trace property boundaries and create accurate digital maps.

The results

Extract has demonstrated significant performance improvements for planning document processing in early testing:

  • 100% text extraction accuracy for all expected text fields from planning documents
  • 94% accuracy in correct identification of dates and key information
  • 90% boundary tracing accuracy matching human-drawn boundaries (0.8 Intersection over Union)
  • 82% location accuracy for boundary center placement within ground truth parameters
  • Takes just 2 minutes per document to extract key information
  • 10 pence cost per document for processing historical planning information

Details

Organisation name
Ministry of Housing, Communities and Local Government (MHCLG) / Incubator for Artificial Intelligence (i.AI)
Government body
UK Government
User group
Wider public sector
Use case type
Specific
Type of technology
Generative AI
Phase
Alpha
Impact
Time savings / Improved efficiency

Links

Get in touch

Email ai-knowledge-hub@dsit.gov.uk to:

  • find out how to use or collaborate on this tool
  • talk to us about your own tool

Content created: 15 September 2025 | Last updated: 27 January 2026