Specialized Solution

Floor Plan OCR

Purpose-built optical character recognition designed specifically for noisy, rotated, and hand-annotated text in architectural drawings.

Key Extraction Targets

Rotated Room Labels
Door & Window Tags (e.g. D101)
Dimension Callouts
Title Block Revision Data
Tabular Door Schedules
Keynotes & Spec Notes

The Problem

Standard OCR fails catastrophically on architectural plans.

Generic OCR tools expect clean, horizontal paragraphs. Floor plans feature text rotated at 90°, 45°, or arbitrary angles, overlapping line work, abbreviated tags, and tabular schedules.

System Workflow

How the AI System Works

01

Orientation & Text Angle Detection

Scan drawing for text bounding boxes and estimate rotation angles (0°, 90°, 180°, 270°, or arbitrary).

02

Bespoke Text Recognition

Run custom-trained OCR tuned on architectural fonts, shorthand, and blueprint symbols.

03

Spatial Text-to-Element Association

Link text callouts to nearby physical drawing elements (e.g. tag 'D-04' to nearest door arc).

Capabilities

Engineered Features

360° Rotated Text Extraction
Overlapping Line-Text Separation
Architectural Tag Parser (D-101, W-02)
Schedule Table Structural Extraction
Dimension Callout Parsing & Conversion
Title Block & Revision History OCR

Architecture

Technical Stack

CRAFT Text Detection ModelREADY
Fine-Tuned PaddleOCR / TrOCRREADY
OpenCV Image Pre-Processing FilterREADY
Spatial KD-Tree Text LinkingREADY
FastAPI Service ContainerREADY

Relevant Case Study

Door & Window Schedule Extraction

Achieved 98.4% accuracy on rotated door tag OCR across complex commercial architectural plan sets.

Read Case Study

Frequently Asked Questions

Our text detection pipeline predicts 4-point polygon bounding boxes and applies affine transformations to rotate text upright before recognition.

Tell us what you want AI to do with your drawings

Fill out the details below and an AI engineer will evaluate your requirements.

Upload Sample Plan (Optional)

PDF, DWG, PNG, or ZIP up to 50MB

Strict confidentiality. Your sample drawings and project details are never shared or used to train public models.