InfraBase - Automatic Building Footprint Segmentation

The project deals with the fully automatic analysis of satellite imagery. The goal is to extract a pixel-accurate segmentation of building roofs to generate a rich meta-data layer.

The project 'infraBASE - An Infrastructure Baselayer generation service to enable Industry 4.0 operations and increase business growth' deals with the automatic analysis of satellite images.

The challenges for the project are the robust segmentation of objects in satellite images. In cooperation with the Kufstein University of Applied Sciences and the company GeoVille automatic analysis methods for satellite images are being developed.

Detection and Segmentation of Buildings

In the infraBASE project automatic methods for the precise segmentation of buildings in satellite images are being developed. In addition to buildings being detected the roof areas are segmented at pixel-level accuracy and converted to a polygon.

Methods: DeepLearning and Variational Level-Sets 

Methodically the project builds upon the parallel project "ImmBild". The Deep Learning methods for image analysis and segmentation that were developed in the course of the project as well as Variational Level-Set methods for the finding and refinement of object contours are being applied. Cooperation partner GeoVille  provides satellite image material including pixel-accurate annotations. The methodical implementation takes place in close cooperation with the Kufstein University of Applied Sciences.

Neural networks architectures optimized for object segmentations are trained from a large set of satellite images to automatically segment building footprints in images. Since neural networks for segmentation do not necessarily provide exact object boundaries we leverage approaches for precise contour finding to improve the segmentation. Contour models are being used in combination with optimisation methods (energy minimisation). Polygons, which delineate the building outlines, are then extracted from the extracted contours. The developed methods allow for the automatic detection of buildings in areas where no geo-information data is available. Building detection und segmentation play an important role in the automatic assessment of damages after natural disasters (e.g. floodings).

Economic Factor

In the future, the results of the infraBASE project will simplify the access to and work with satellite images and support the assessment and analysis of the local infrastructure. InfraBASE will enable the generation and quantitative analysis of automatic metadata for areas where no geoinformation data are available and so save time and money.

  • ​GeoVille Information Systems and Data Processing GmbH
  • Fachhochschule (FH) Kufstein Tirol
04/01/2018 – 09/30/2019
Involved Institutes, Groups and Centers
Institute of Creative\Media/Technologies
Research Group Media Computing