The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLIII-B2-2020
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLIII-B2-2020, 1309–1313, 2020
https://doi.org/10.5194/isprs-archives-XLIII-B2-2020-1309-2020
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLIII-B2-2020, 1309–1313, 2020
https://doi.org/10.5194/isprs-archives-XLIII-B2-2020-1309-2020

  14 Aug 2020

14 Aug 2020

EFFICIENT BUILDING CATEGORY CLASSIFICATION WITH FAÇADE INFORMATION FROM OBLIQUE AERIAL IMAGES

C. Xiao1,3, X. Xie2,3, L. Zhang4, and B. Xue2,3 C. Xiao et al.
  • 1Artificial Intelligence and Earth Perception Research Center, School of Automation Engineering, University of Electronic Science and Technology of China, China
  • 2Key Lab of Pollution Ecology and Environmental Engineering, Institute of Applied Ecology, Chinese Academy of Sciences, Shenyang 110016, China
  • 3Key Lab for Environmental Computation and Sustainability of Liaoning Province, Shenyang 110016, China
  • 4Department of compute science and engineering, Southern University of Science and Technology, China

Keywords: Building Category, Façade, Oblique Aerial Images, Remote Sensing, Classification

Abstract. Building category refereed to categorizing structures based on their usage is useful for urban design and management and can provide indexes of population, resource and environment related problems. Currently, the statistics are mainly collected by manual from street data or roughly extracted from remote sensing data which are either laborious or too coarse. With remote sensing data (e.g. satellite and aerial images), buildings can be automatically identified from the top-view, but the detailed categories of single buildings are not recognized. Façade from oblique-view image can greatly help us to identify the categories of buildings, for example, balcony usually exist in resident buildings. Hence, in this paper, we propose an efficient way to classify building categories with the façade information. Firstly, following the texture mapping procedure, each building’s façade textures are cropped from oblique images via a perspective transformation. Then, the average colour, the stander deviation in R, G, B channel, and the rectangle Haar-like features are extracted and feed to a further random forest classifier for their category identifications. In the experiment, we manually selected 262 building façades that can be classified into four functional types as: 1) regular residence ; 2) educational building; 3) office ; 4) condominium. The results shows that, with 30% data as training samples, the classification accuracy can reach 0.6 which is promising in real applications and we believe with more sophisticated feature descriptors and classifiers, e.g., neuronal networks, the accuracy can be much higher.