Authors: Shairoz Sohail*, Esri, Omar Maher, Esri - Manager, Amin Tayyebi, Esri - Collegue, Daniel Wilson, Esri - Collegue, Rohit Singh, Esri - Collegue
Topics: Hazards, Risks, and Disasters, Quantitative Methods, Spatial Analysis & Modeling
Keywords: Computer vision, object detection, building damage, building footprints, disasters, first responders, deep learning
Session Type: Paper
Start / End Time: 9:55 AM / 11:35 AM
Room: Washington 6, Marriott, Exhibition Level
Presentation File: No File Uploaded
In post-disaster scenarios, the identification of buildings that have undergone damage is a key pre-requisite in allocating potentially life-saving resources. Unfortunately, the manual identification of such buildings takes significant manpower and time, which is critical in these scenarios. With the increased accessibility of drone imagery and computer vision models, a complete automation of damaged structure detection is now possible. We propose and demonstrate the usage of a combination of Mask RCNN and CNN deep learning models (trained on over a million building footprints and hundreds of damaged building examples) to automatically identify building footprints from aerial imagery and provide damage level assessment for each building. We also provide an integration of these models with ArcGIS Pro to provide an online heat map of damaged structures for easy consumption by first responders.