EN

陈登峰

副教授   硕士生导师

个人信息 更多+
  • 教师拼音名称: Chen Dengfeng
  • 所在单位: 建筑设备科学与工程学院

其他联系方式

邮编:

通讯/办公地址:

邮箱:

论文成果

当前位置: 中文主页 - 科学研究 - 论文成果

An integrated damage-fidelity and geometry-enhancing Gaussian Splatting framework for post-earthquake building reconstruction

发布时间:2026-09-15
点击次数:
DOI码:
10.1016/j.jobe.2026.116976
发表刊物:
Journal of Building Engineering
关键字:
Post-earthquake building assessment; UAV building inspection; Gaussian splatting; Crack-aware reconstruction; Structural damage assessment
摘要:
Post-earthquake environments require rapid and reliable building safety assessment and damage documentation. However, conventional close-range inspections are often labor-intensive, time-consuming, and potentially hazardous in damaged structures. Although UAV imaging and neural rendering techniques have recently enabled efficient three-dimensional reconstruction, existing methods still struggle to preserve fine-scale structural defects such as cracks and spalling, particularly under limited viewpoints and complex visual conditions.To address these challenges, this study proposes a Damage-Fidelity and Geometry-Enhancing Gaussian Splatting (DGE-GS) framework for defect-aware reconstruction of earthquake-damaged buildings. Starting from coarse geometry obtained via structure-from-motion (SfM), the proposed method introduces curvature- and boundary-rate-driven adaptive Gaussian densification to enhance sampling around structural defects. A hierarchical color-clustering initialization strategy is further designed to improve the representation of crack and spalling regions. In addition, a gated residual and view-attention mechanism is employed to align and fuse multi-view features with Gaussian representations. Within a differentiable rendering framework, end-to-end optimization is achieved using a combined photometric–perceptual–geometric loss. Experiments on two representative real post-earthquake masonry building cases and the ETH3D benchmark demonstrate that the proposed method improves reconstruction quality in the evaluated scenarios.The results indicate notable improvements in both appearance fidelity and geometric consistency, indicating its potential for post-earthquake damage documentation and defect-aware 3D reconstruction in representative damaged masonry scenes.
合写作者:
刘世鹏,李嘉豪,李皓明,赵亮
第一作者:
张中霖
论文类型:
期刊论文
通讯作者:
陈登峰
卷号:
130
是否译文:
否
发表时间:
2026-01-01
收录刊物:
SCI