【摘要】 为解决地震期间缺乏建筑结构振动实测数据的问题，本文提出一种基于虚拟现实与计算机视觉的建筑地震响应可视化监测与验证方法。首先，利用虚幻引擎构建室内建筑模型，并通过蓝图系统导入OpenSees计算的地震响应时程，实现结构构件的非线性协调变形，从而获得逼真的地震作用场景。其次，采用深度学习模型实现对墙、梁、楼板等构件的实例分割，并提取边缘线段特征。再基于摄影测量原理建立图像边缘点的空间几何关系，计算构件的三维位移与层间位移角时程。考虑实际监控摄像头的微小振动，进一步分析了平动、光轴位移与转角扰动对层间位移角识别结果的影响规律。结果表明，本文方法识别的层间位移角与虚幻引擎的输入高度一致，最大百分误差仅为0.28%；摄像头平移抖动在层间差分中可完全抵消，光轴微动和转角扰动引起的误差影响较小。研究表明，本文方法能够在虚拟环境中以低成本实现视觉监测算法的验证，并为实际建筑地震响应的视觉识别提供可行技术路径。更多还原

【Abstract】 To address the lack of measured structural vibration data during earthquakes, this study proposed a visual monitoring and verification method for building seismic responses based on virtual reality and computer vision. First, an indoor building model was constructed using Unreal Engine, and the seismic response time histories calculated by OpenSees were imported through the blueprint system to achieve nonlinear and coordinated deformation of structural components, thereby creating realistic seismic scenarios. Second, a deep learning model was employed to perform instance segmentation of walls, beams, slabs, and other components, followed by the extraction of edge line features. Then, the spatial geometric relationships of image edge points were established based on photogrammetric principles, and the three-dimensional displacements and inter-story drift angle time histories of the components were calculated. Considering the minor vibrations of actual surveillance cameras, the effects of translational motion, optical-axis displacement, and angular perturbation on the identified inter-story drift angle were further analyzed. The results showed that the inter-story drift angles identified by the proposed method were highly consistent with the inputs of Unreal Engine, with a maximum percentage error of only 0.28%. The translational jitter of the camera could be completely canceled out through inter-story differencing, while the effects of optical-axis motion and angular perturbation remained minimal. This study demonstrates that the proposed method can achieve low-cost validation of visual monitoring algorithms in a virtual environment and provide a feasible technical pathway for visual identification of seismic responses in real buildings.更多还原