Building Recognition using AI at Large-Scale

What is BRAILS

The software framework Building Recognition using AI at Large-Scale, BRAILS, is provided by SimCenter to assist in creating the assest inventories needed to perform regional scale simulations to study the effects of a hazard on a region. The BRAILS framework provides a set of modules that utilize machine learning (ML), deep learning (DL), and computer vision (CV) techniques to extract information from satellite and street view images. Each of the modules performs a specific task. For example, for building assets inventories, seperate modules provide information on occupancy class, roof type, foundation elevation, roof elevation, and year built among others. The BRAILS framework will also provide workflows that allow the users to put the individual modules together to determine multiple attributes in a single pass. One such workflow application, CityBuilder, is an application released with BRAILS to create regional-scale building inventories by utilizing multiple sources of data, including OpenStreetMap, Microsoft Footprint Data, Google Maps, and extracting information from them using the BRAILS modules.

SimCenter, which provides applications that are used in the assessment of the impact of natural hazard at the regional scale, is currently developing BRAILS in order to generate inventories for use in its testbeds and will in the future include BRAILS applications in it’s R2D application. BRAILS, as a consequence, is under constant development with new modules and workflows being constantly added.

SimCenter also provides trained models so that users can immediately use of many of the modules within BRAILS. These trained models, which are released on Zenodo, are the product of SimCenter ML development/validation efforts and are being constantly updated as the SimCenter training sets are expanded upon. While use of these training sets are encouraged, it should be noted that the models are trained using data obtained and labelled for SimCenter testbeds and using the sparse data available from OpenStreetMap. How these modules generalize to new and unseen data from different geographical locations depends on how similar they are to the training data. Generalization of machine learning models remains an active research area. Users should exercise cautions when the modules are used beyond their intended purposes and trained model ability.

User Manual

• 1. Acknowledgments
• 2. About
• 3. Installation
• 4. User Guide
• 5. Troubleshooting
• 6. Examples
• 7. Bugs & Feauture Requests
• 8. Copyright and License

Technical Manual

• 1. Theory and Implementation
• 2. Validation
• 3. Generalization

Developer Manual

• 1. How to Extend
• 2. Coding Style

How to cite

Charles Wang, Sascha Hornauer, Barbaros Cetiner, Yunhui Guo, Frank McKenna, Qian Yu, Stella X. Yu, Ertugrul Taciroglu, & Kincho H. Law. (2021, March 1). NHERI-SimCenter/BRAILS: Release v2.0.0 (Version v2.0.0). Zenodo. http://doi.org/10.5281/zenodo.4570554

Contact

NHERI-SimCenter nheri-simcenter@berkeley.edu

References

BHF+19

Benjamin Bischke, Patrick Helber, Joachim Folz, Damian Borth, and Andreas Dengel. Multi-task learning for segmentation of building footprints with deep neural networks. 2019 IEEE International Conference on Image Processing (ICIP), pages 1480–1484, 2019.

Goo97

Pierre Goovaerts. Geostatistics for natural resources evaluation. Oxford University Press on Demand, 1997.

HZRS16

Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770–778. 2016.

HW68

David H Hubel and Torsten N Wiesel. Receptive fields and functional architecture of monkey striate cortex. The Journal of physiology, 195(1):215–243, 1968.

LHF+19

Weijia Li, Conghui He, Jiarui Fang, Juepeng Zheng, Haohuan Fu, and Le Yu. Semantic segmentation-based building footprint extraction using very high-resolution satellite images and multi-source gis data. Remote Sensing, 11(4):403, 2019.

Mic

Microsoft. US Building Footprints. URL: https://github.com/microsoft/USBuildingFootprints.

SVI+16

Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. Rethinking the inception architecture for computer vision. Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2818–2826, 2016.

Van10

Erik Vanmarcke. Random fields: analysis and synthesis. World Scientific, 2010.

WC18

C Wang and Q Chen. A hybrid geotechnical and geological data-based framework for multiscale regional liquefaction hazard mapping. Géotechnique, 68(7):614–625, 2018.

Wan20

Chaofeng Wang. Occupancy test. December 2020. URL: https://doi.org/10.5281/zenodo.4386991, doi:10.5281/zenodo.4386991.

WCSJ17

Chaofeng Wang, Qiushi Chen, Mengfen Shen, and C Hsein Juang. On the spatial variability of cpt-based geotechnical parameters for regional liquefaction evaluation. Soil Dynamics and Earthquake Engineering, 95:153–166, 2017.

Wan19

Charles Wang. Random satellite images of buildings. October 2019. URL: https://doi.org/10.5281/zenodo.3521067, doi:10.5281/zenodo.3521067.

WYM+19

Charles Wang, Qian Yu, Frank McKenna, Barbaros Cetiner, Stella X. Yu, Ertugrul Taciroglu, and Kincho H. Law. Nheri-simcenter/brails: v1.0.1. October 2019. URL: https://doi.org/10.5281/zenodo.3483208, doi:10.5281/zenodo.3483208.

ZKJS18

Kang Zhao, Jungwon Kang, Jaewook Jung, and Gunho Sohn. Building extraction from satellite images using mask r-cnn with building boundary regularization. CVPR Workshops, pages 247–251, 2018.