城市建筑爆炸事故后果的 VR 沉浸式可视化

To tackle the challenges associated with visualizing and characterizing the consequences of explosions, this study introduces an immersive visualization and damage detection framework utilizing 3D Gaussian Splatting (3DGS) and Virtual Reality (VR) technologies. This framework reconstructs accident scenes with high fidelity from multi-view images, providing a robust foundation for multi-scale damage assessment and interactive visualization. The system begins by employing Structure from Motion (SfM) to process unordered image sets, estimate camera poses, and generate sparse point clouds. A 3DGS model is subsequently constructed and optimized using differentiable rendering and adaptive density control, ensuring an accurate representation of both geometric structures and surface textures. The pre- and post-explosion models are converted into point clouds for comparative analysis. At the structural level, bidirectional nearest-neighbor matching utilizing Euclidean distance quantifies structural deformation, while at the texture level, color feature distances are employed to identify surface damage. The results of this analysis are then mapped back onto the 3DGS model, producing a three-dimensional visualization that highlights explicit damage features. For implementation, a VR interaction platform was developed using the Unity engine, compatible with PICO 4 headsets. GPU-accelerated compute shaders significantly enhance the rendering efficiency of large-scale Gaussian distributions, achieving real-time performance at 286 frames per second. In the VR environment, users can freely navigate scenes, toggle between pre- and post-explosion states, and conduct model alignment, area selection, and real-time quantitative analysis using handheld controllers. Experimental validation was performed on a single-story reinforced concrete test building, where a GoPro 11 camera captured 313 pre-explosion images and 498 post-explosion images. The reconstructed models effectively captured both the intact structural morphology and the post-explosion damage patterns. Quantitative comparisons revealed that the system accurately identified structural deformation and surface damage, with damage quantification errors remaining below 10. 5% . These results address the limitations of conventional two-dimensional visualization, significantly enhancing the precision and immersion of explosion consequence analysis. Moreover, this approach offers an innovative methodology for accident investigation and safety assessment.