Rotated Rectangles for Symbolized Building Footprint Extraction

Matt Dickenson, Lionel Gueguen; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2018, pp. 225-228

Abstract


Building footprints (BFP) provide useful visual context for users of digital maps when navigating in space. This paper proposes a method for extracting and symbolizing building footprints from satellite imagery using a convolu- tional neural network (CNN). The CNN architecture out- puts rotated rectangles, providing a symbolized approxi- mation that works well for small buildings. Experiments are conducted on the four cities in the DeepGlobe Chal- lenge dataset (Las Vegas, Paris, Shanghai, Khartoum). Our method performs best on suburbs consisting of individual houses. These experiments show that either large buildings or buildings without clear delineation produce weaker re- sults in terms of precision and recall.

Related Material


[pdf]
[bibtex]
@InProceedings{Dickenson_2018_CVPR_Workshops,
author = {Dickenson, Matt and Gueguen, Lionel},
title = {Rotated Rectangles for Symbolized Building Footprint Extraction},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2018}
}