The aim of this AI service is to extract the information of building elements contained in a 2D floor plan. The focus is on general component properties that are relevant for most component classes. A central feature is the recognition of the displayed geometries as polygons.
Component recognition on 2D plans is a fundamental task for automating the modeling process. For this purpose, building element information is extracted from floor plans using a deep learning-based instance segmentation. The service provides a list of components with properties that can be derived from a single plan. Basic properties of the building elements such as component class and geometry as well as complementary properties such as opening direction are recognized. The starting point of the recognition are pixel-based plans.
The deep-learning-based approach can incorporate semantic information into the recognition and provides the flexibility to process a wide variety of representations of the plans. The geometries of the building elements are recognized directly in a vector representation and are thus immediately usable for modeling.
Data
- Quality
- 2D raster plans with a clear representation of building elements
- Input data format
- PNG or JPEG of a floor plan
- Output data format
- JSON with recognized building model of the floor in “BIMKIT building construction format
- PNG with visualization as intermediate result
Extracted information
List of building components with properties:
- Component class (wall, door, window, room)
- Geometry on plan as polygon
- Other properties
- Door: direction of opening
- Wall: inner/outer wall
- Door, window: relation to wall
Contact person:
Heinrich Fröml, Hottgenroth Software AG
Julius Freiny, Hottgenroth Software AG