Semi-supervised Learning from Street-View Image and OpenStreetMap for Automatic Building Height Estimation

Introduction:

Accurate building height estimation is key to the automatic derivation of 3D city models from emerging big geospatial data, including Volunteered Geographical Information (VGI), where a low-cost and automatic solution for large-scale building height estimation is currently missing. More recently, the fast development of VGI data platforms, especially OpenStreetMap (OSM) and crowdsourced street-view image (SVI), offers a stimulating opportunity to fill this research gap. In this work, we propose a semi-supervised learning (SSL) method to automatically estimate building height from Mapillary SVI and OSM data, which is able to create low-cost and open-source 3D city models in LoD1.

Specifically, the proposed method consists of three parts: first, we propose an SSL schema with the option of setting different ratio of "pseudo label" during the supervised regression; second, we extract multi-level morphometric features from OSM data (i.e., buildings and streets) for the purposed of inferring building height; last, we design a building floor estimation workflow with a pre-trained facade object detection network to generate "pseudo label" from SVI and assign it to the corresponding OSM building footprint.

Complete feature list:

Abbr. name	Definition	Range/ Unit
Building*
area	Area of the building	meter
perimeter	Perimeter of the building	meter
circularcompactness	The ratio between the area of the building footprint and the area of the circumscribed circle.	[0, 1]
longestaxislength	Length of the longest axis of the building footprint. Axis is defined as a diameter of minimal circumscribed circle around the convex hull.	meter
elongation	Elongation of the minimum bounding box around the building footprint.	[0, 1]
convexity	Area of the footprint divided by the area of the convex hull around the footprint.	[0, 1]
orientation	Orientation of the longest axis of bounding rectangle in range 0 – 45. It captures the deviation of orientation from cardinal directions	degree
corners	Calculates number of corners of the building.	count
sharedwall	Length of wall shared with other buildings.	meters
Block*
Features of buildings in blocks
blockcount	Number of buildings in the block that the building is part of.	count
avBlockFootprintArea	Average footprint area of buildings in the block	squared meter
stdBlockFootprintArea	Standard deviation of footprint areas of buildings in the block.	squared meter
blockTotalFootprintArea	Total building footprint of the block. Unit: squared meters.	squared meter
Features of block itself
BlockPerimeter	Total perimeter of the block.	meter
BlockLongestAxisLength	Length of the longest axis of whole block footprint.	meter
BlockElongation	Elongation of the minimum bounding box around the whole block footprint.	[0, 1]
BlockConvexity	Convexity of the whole block footprint.	[0, 1]
BlockOrientation	Orientation of the whole block footprint.	degree
BlockCorners	Number of corners of the whole block footprint.	count
Street & intersection*
closeness500	Local closeness centrality for the closest street to the building.	[0, 1]
betweenness	Betweenness centrality of the closest street to the building.	[0, 1]
global_closeness	Global closeness centrality of the closest street to the building.	[0, 1]
openness	Openness of the closest street to building. Proportion of the street where buildings are or not present on the sides of the street.	[0, 1]
width_deviations	Standard deviation of the width of the closest street to the building. Width is defined here as the average distance between buildings on both sides of the street.	meters
widths_street	Width of the closest street to the building.	meters
lengths_street	Length of the closest street to the building.	meters
distance_road	Distance between the building and the closest street.	meters
distance_intersection	Distance between the building and the closest intersection.	meters
Street-based block*
street_based_block_phi	Anisotropy index of the street-based block at the building location.	[0, 1]
street_based_block_area	Area of the street-based block at the building location.	squared meter
* 50,200,500m buffers applied and the mean and std values were calculated.

Contact

Dr. Hao Li
Email: [email protected]
Technische Universität München, Dartment Aerospace and Geodesy
Professur für Big Geospatial Data Management
Lise Meitner Str. 9, 85521 Ottobrunn