CDS&E: Enabling Time-critical Decision-support for Disaster Response and Structural Engineering through Automated Visual Data Analytics

点击翻译按钮获取中文摘要

After a disaster, teams of trained engineers are charged with the task of collecting perishable data. These building reconnaissance teams collect data and information from the buildings that experienced the disaster, including photographs and measurements in the region. This information is collected to better understand the consequences of these events, and to improve the design of future structures. An enormous amount of images and videos is generated in just a few days, and to gather the most critical information in the time allowed, the engineers on these teams must quickly make daily decisions on where and what data to collect to achieve their mission. This research, which harnesses powerful computer vision methods to address real world civil engineering problems, aims to develop efficient methods to analyze and organize the collected images in the field, thereby enabling teams to collect the most useful data for building resilient communities worldwide. The project will leverage decades of experience in field missions from project researchers and domestic and international collaborators. A diverse set of students will be engaged in interdisciplinary research with international opportunities.The application of computer vision methods to address disaster response and structural engineering problems is not simple or straightforward. This project will systematically build the knowledge needed for their successful implementation in time-critical situations. Engineers with significant field-mission experience will annotate images. These records will provide the basis for determining the visual contents needed to make decisions in the field and how the contents are spatially interconnected in the images. This forms the foundation for determining the prior knowledge that can and must be included in the deep neural network structures to facilitate rapid decision-making in the field. To quantitatively evaluate the approach, a reconnaissance testbed will be established using a diverse set of images from past data collection missions. The computational time and accuracy will be measured and documented to establish a detailed profile of the classification results. This knowledge will enable the team collecting data during a reconnaissance mission to maximize the value of the data they collect by ensuring that they can successfully perform a given task, in a certain amount of time, applied to a suite of images. This capability will provide the evidence on which to base recommendations for further investigations and/or changes to design guidelines.