Collaborative Research: Machine Vision Enhanced Post Earthquake Inspection and Rapid Loss Estimation

该项目的目标是开发和验证用于钢筋混凝土框架建筑震后检查的计算框架，该框架将能够快速、自动地评估结构的损坏状态以及修复结构所需的成本和时间。拟议的自动化程序将从使用安装在检查员S安全帽上的高分辨率摄像机收集视频帧开始。然后，将使用最先进的检测和提取算法来检测RC柱，并识别和表征柱的损伤。构件损伤将使用基于经验的模型进行分类，并将使用构件损伤来确定建筑物的损坏状态。建筑物的损坏状态、结构和类型将被用来查询一组脆弱性曲线，这些曲线定义了建筑物在余震期间倒塌的可能性，从而提高了对风险的了解。可交付成果包括基本可见损伤特征目录、基于模型的识别和分析工具、通过硬件进行的演示和验证、研究结果文档、工程专业学生教育以及面向建筑评估师的外展研讨会。如果成功，该研究结果将在结构构件和视频损伤识别领域提供首个可靠的方法。这一科学突破将使研究人员能够将这项工作整合到竣工建筑信息建模、项目监控、虚拟和增强现实以及其他对工程界重要的应用程序中。此外，这将是已知的第一项研究，利用稳健的概率方法将建筑构件(柱或墙)的视觉损害与建筑倒塌的可能性定量联系起来。预计该项目所寻求的发现将作为建立新的损害评估知识库的基础，并促进计算机视觉和结构工程领域之间的智能交叉授粉。结果将被传播，以允许创建商业软件，这些软件提高了精度，降低了成本，并与减轻重量的设备一起工作。工程专业的研究生和本科生以及K-12年级的学生将从课堂教学和参与研究中受益。

The objective of this project is to develop and validate a computational framework for post-earthquake inspection of reinforced concrete (RC) frame buildings that will enable rapid, automated assessment of the damage state of the structure and of the cost and time required to repair the structure. The proposed automated procedure will start with collection of video frames using a high-resolution video camera mounted on an inspector?s hardhat. Then, state-of-the-art detection and extraction algorithms will be employed to detect RC columns and identify and characterize column damage. Component damage will be classified using empirically based models, and component damage will be used to determine the damage state of the building. Building damage state, configuration and type will be used to query a set of fragility curves defining the likelihood of building collapse during an aftershock and, thereby, provide an improved understanding of risk. Deliverables include a catalog of fundamental visible damage characteristics, model-based recognition and analysis tools, demonstration and validation via hardware, documentation of research results, engineering student education, and outreach seminars to building evaluators.If successful, the results of this research will provide the first robust method in the area of structural member and damage recognition from video. This scientific breakthrough will allow researchers to integrate this work in as-built building information modeling, project monitoring, virtual and augmented reality and other applications of importance to the engineering community. Also, this will be the first known study to quantitatively link visual damage in a building component (column or wall), to the likelihood of building collapse using robust probabilistic methods. The discoveries sought in this project are expected to serve as a foundation for a new knowledge base in damage assessment and to promote intellectual cross-pollination among the fields of computer vision and structural engineering. The results will be disseminated to allow the creation of commercial software that have increased precision, reduced cost, and work with reduced weight devices. Graduate and undergraduate engineering students and K-12 students will benefit through classroom instruction and involvement in the research.