Reconstruction of Existing Buildings for Evacuation Assessment under Emergency Situations Using 3D Gaussian Splatting and Machine Learning

Xiaofeng Liao, Bingyang Zhou, Yiqiao Liu, Liwen Zhang, Zhaoyin Zhou, Heap-Yih Chong,
Reconstruction of Existing Buildings for Evacuation Assessment under Emergency Situations Using 3D Gaussian Splatting and Machine Learning,
Reliability Engineering & System Safety,
2025,
111887,
ISSN 0951-8320,
https://doi.org/10.1016/j.ress.2025.111887.
(https://www.sciencedirect.com/science/article/pii/S0951832025010877)
Abstract: Existing buildings characterized by aging and structurally complex layouts pose significant challenges for emergency evacuation, necessitating advanced and dynamic risk assessment methods that go beyond conventional static evaluations. The primary scientific objective of this study is to develop and validate an integrated computational framework that combines high-fidelity 3D modeling with machine learning to quantitatively assess and predict evacuation risks in such structures. Using laser-scanned structural data, we reconstructed detailed building information models (BIM) via anisotropic 3D Gaussian Splatting, enabling accurate spatial representation. These models were coupled with agent-based Pathfinder simulations to examine evacuation dynamics under diverse crowd densities and movement speeds. Simulated egress capacity informed a risk stratification mechanism. Further, we developed four machine learning models to predict evacuation time, among which the Improved Particle Swarm Optimization-integrated Convolutional Neural Network–Long Short-Term Memory (IPSO-CNN-LSTM) model achieved superior predictive accuracy. The results confirm the efficacy of the proposed framework in identifying critical evacuation risks and supporting real-time emergency decision-making. This research contributes a scientifically grounded, data-driven paradigm for dynamic risk assessment in built environments and offers scalable solutions for enhancing safety in aging urban infrastructure under various emergency scenarios.
Keywords: Building, 3D Gaussian Splatting;Machine Learning;Evacuation Assessment; Emergency Situations; Fire; Digitalization