Computer-Assisted Imaging for Structural Damage

在地震等灾难发生后，检查人员的任务是评估受影响建筑物和结构的完整性。根据灾难的规模，所需的检查次数可能高达数千次。既要考虑公共安全，也要考虑经济压力，因此对建筑物和构筑物进行快速准确的评估至关重要。本研究项目的目标是开发一种利用数字图像分析和人工智能来评估自然灾害后土木结构完整性的方法。完全实现后，这项技术将使灾后检查人员仅使用一台数码相机和便携式计算机就可以快速准确地估计结构损坏。将继续研究一种基于计算机视觉的结构评估的综合方法，这种方法足够灵活，可以在高度多样化和具有挑战性的现场环境中运行。为了真正全面，这种方法必须是分层的，首先识别在图像中观察到的结构组件的系统级上下文，然后利用该信息来增强从分段例程中提取的损伤的本地化描述。在这个研究项目中，将挖掘和分析NEES实验的可视化和仪器记录，以验证算法的有效性。该研究计划将促进对视觉可见损伤如何与结构性能相关的理解，并将为分层学习技术在基于计算机视觉的结构评估领域中使用的适用性提供见解。更广泛地说，该项目将产生一种可扩展的可视、非接触结构评估方法，为未来的结构监测系统奠定基础。

After a disaster, such as an earthquake, inspectors are tasked with assessing the integrity of affected buildings and structures. Depending on the scale of the disaster, the number of required inspections can range into the thousands. There are both public safety and economic pressures to consider, and so rapid and accurate assessments of buildings and structures are vital. The goal of this research project is to develop a method using digital image analysis and artificial intelligence to assess the integrity of civil structures after a natural disaster. Fully realized, this technology will enable post-disaster inspectors to rapidly and accurately estimate structural damage using only a digital camera and portable computer. Research in a comprehensive method of computer vision-based structural assessment will be pursued, one that is flexible enough to operate in highly varied and challenging field environments. To be truly comprehensive, such a methodology must be hierarchical, first recognizing the system-level context of structural components observed in an image and then leveraging that information to augment localized descriptions of damage extracted from segmentation routines. In this research program, the visual and instrumentation records of NEES experiments, available through the NEESHub, will be mined and analyzed to validate the algorithm. The research program will advance understanding as to how visually observable damage correlates to structural performance, and will provide insights into the suitability of hierarchical learning techniques for use in the field of computer vision-based structural assessment. More broadly, the project will result in an extensible method of visual, non-contact structural assessment that will provide a foundation for future structural monitoring system.