Building Detective For Disaster Preparedness

Building Recognition using AI at Large-Scale for Natural Hazard Engineering

The Building Detective For Disaster Preparedness project is a part of the Building Recognition using AI at Large-Scale (BRAILS) project, which intends to learn the vulnerabilities of buildings to natural hazards using machine learning and deep learning. The project is hosted by the Computational Modeling and Simulation Center (SimCenter), which is funded by National Science Foundation and provides next-generation computational modeling and simulation software tools, user support, and educational materials to the natural hazards engineering research community with the goal of advancing the nation’s capability to simulate the impact of natural hazards on structures, lifelines, and communities. In addition, the Center will enable leaders to make more informed decisions about the need for and effectiveness of potential mitigation strategies.

Natural disasters pose significant destructive impacts to society, from damaging buildings, endangering lives to economic loss. For example, earthquake hazards can cause damage to buildings and infrastructures because of crumbling the above-ground structure and warping the underground foundation. Hurricanes and tornados can have detrimental effects on buildings due to the differential pressures on roofs and walls, and wind-borne debris on windows and building facades. Damages to the building envelope also lead to damages to the building's interior from rain and storm.

Deep Learning for City-scale Building Information Modeling

While the occurrence of natural hazards cannot be precisely predicted, their impacts on buildings and infrastructures are fairly well understood. Damages and losses can be minimized with effective management and hazard mitigation planning. Mitigation measures can be identified, prioritized, and implemented through comprehensive risk assessment studies. Buildings represent a major portion of the built environment and are vulnerable to a broad variety of natural hazards. For regional planning, buildings are of major consideration for response planning and disaster management. The first step of regional hazard risk analysis is therefore to acquire information about the buildings. To this end, we initiated the BRAILS project, in which we use deep learning techniques to extract building information from satellite or street view images. The extracted information will help to better understand the vulnerabilities of the buildings and to reduce the uncertainties in numerical simulations.

Your Contribution

The project's phase one goal is to extract building information from images using deep learning. To achieve that goal, citizen scientists can help to create annotations on street view and satellite images, which are needed to train a deep learning system to identify and classify building attributes.

References:
G. G. Deierlein, & A. Zsarnóczay, eds. (2021, February 23). State of the Art in Computational Simulation for Natural Hazards Engineering (Version v2). Zenodo. http://doi.org/10.5281/zenodo.4558106
Q. Yu, C. Wang*, B. Cetiner, S. X. Yu, F. McKenna, E. Taciroglu, K. H. Law, Building Information Modeling and Classification by Visual Learning at A City Scale, 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada, 2019. https://arxiv.org/abs/1910.06391