Comparing classification methods for the building plan components

Daniel Wilfing, Oliver Krauss

Research output: Chapter in Book/Report/Conference proceedingsConference contributionpeer-review

Abstract

The classification methods Histogram of Oriented Gradients, Bag of Features, Support Vector Machines and Neural Networks are evaluated to find a fitting solution for the automatic classification of building plan components. These components feature shapes with little features and minor differences. After processing the building plans for the classification, feature analysis methods, as well as machine learning based approaches, are tested. The results of the classification methods are compared and the behaviors of the classification methods are analyzed. First results have shown, that neural network classification using line data extracted via Hough transformation and additional calculations surpass other classification methods tested in this work. It was found that the basic structure of building plan components can be detected with neural networks, but further improvements have to be made, if only a single classification process is to be relied on. In the future this work will be used to create 3D building models from 2D plans and enable agent based simulation in the models.

Original languageEnglish
Title of host publication32nd European Modeling and Simulation Symposium, EMSS 2020
EditorsMichael Affenzeller, Agostino G. Bruzzone, Francesco Longo, Antonella Petrillo
PublisherDIME UNIVERSITY OF GENOA
Pages154-160
Number of pages7
ISBN (Electronic)9788885741454
DOIs
Publication statusPublished - 2020
Event32nd European Modeling and Simulation Symposium, EMSS 2020 - Virtual, Online
Duration: 16 Sept 202018 Sept 2020

Publication series

Name32nd European Modeling and Simulation Symposium, EMSS 2020

Conference

Conference32nd European Modeling and Simulation Symposium, EMSS 2020
CityVirtual, Online
Period16.09.202018.09.2020

Keywords

  • Building plan
  • Classification
  • Feature analysis
  • Machine learning
  • Modeling & Simulation

Fingerprint

Dive into the research topics of 'Comparing classification methods for the building plan components'. Together they form a unique fingerprint.

Cite this