Image Masking
Image Masking for Building Density Analysis
-
Years: 2021
Team: Peter Evhert, Carolina Leite
Tutors: Zuardin Akbar
Examiner: Thomas Wortmann
Course: Computing in Architecture
University: University of Stuttgart
Institutes: Institute for Computational Design and Construction (ICD)
This project is an exploration of generative deep learning that uses Image segmentation to identify buildings in satellite images and then evaluate building density on a large scale. A pre-trained Convolutional Neural Network model was used for image segmentation. The training dataset consists of aerial satellite images containing buildings, mostly suburban homes, and ground truth image masks of the building footprints. For the final application, image segmentation was performed on a composite aerial image of a suburban area of Stuttgart, Germany. Finally, the building density was evaluated from the pixel values of the resulting masks.
