Generating Architectural Drawings from Images

This project examines the potential applications of structure from motion techniques to generating architectural floor plans. Using simple cameras, we can accurately capture the 2D structure of buildings, eliminating the need for expensive and laborious manual survey work that is still used to generate most record drawings.

RANSAC Wall Detection

By detecting the structure's walls, we can distill the model to its basic components and eliminate drift, distortion, and noise caused by accumulated errors.

Distortion Correction

The raw data capture, in white, suffers from distortion, resulting in curving walls and odd angles. By detecting the walls, we can correct and align the model, creating a more accurate plan.

Region Growing Segmentation

By segmenting the point cloud by distance and smoothness, we can eliminate objects, such as furniture, that should not factor into the architectural plan.

Wall Interpolation

Poor captures can result in data filled with holes. Wall interpolation fills in regions likely to be continuations of surfaces, cleaning the model.

Generated Floor Plan of the Yale University Art Gallery

A floor plan generated from a large and complex structure with many individual rooms.

Generating Architectural Drawings from Images: Work