In indoor evacuations, children may use three-dimensional (3D) motions (e.g., low crawling, climbing up/down) to navigate above or below indoor objects within constrained spaces or under higher urgency levels. Understanding how these motions influence children’s evacuation efficiency benefits the development of behavioural instructions. This study aims to investigate the influence of 3D motions on the efficiency of children’s indoor evacuations. We employ a simulation method that integrates a voxel-based 3D indoor model with an agent-based model. Detailed scenarios are elaborated to account for agent numbers, urgency levels, and physical attributes. These scenarios serve as exploratory demonstrations rather than validated simulations grounded in real-world evacuation experiments. Findings are: 1) The use of 3D motions is relatively low across all urgency levels and agent numbers. 2) The number of agents moving above or below indoor objects slightly increases with higher urgency and more agents. 3) 3D motions may not significantly influence children’s evacuation efficiency due to minor congestion, limited use of 3D motions, and lower speeds of 3D motions. We suggest further investigation into behavioural instructions, including 1) Limit the use of 3D motions in low-density scenarios. 2) Conditional use of 3D motions when local congestion significantly hinders direct evacuation paths. 3) Conduct evacuation exercises tailored to children’s physical and cognitive abilities under specific scenarios. This study fosters immediate evacuation response strategies, long-term preparedness, and educational measures for children.

# Abstract

Three-dimensional (3D) city models have evolved into essential tools for urban planning, simulation, and analysis. However, current formats like CityJSON and CityJSONSeq lack cloud-native optimizations for efficiently handling large-scale datasets. While cloud-optimized formats exist for 2D data, solutions for complex 3D city models remain limited. We introduce FlatCityBuf, a cloud-optimized encoding for CityJSON based on FlatBuffers, which combines binary serialization with spatial and attribute indexing. Our implementation employs a packed Hilbert R-tree for spatial queries and a bottom-up Static B+ tree for attribute indexing, enabling efficient bounding box and attribute filtering. By utilizing FlatBuffers' zero-copy deserialization and structured flattening of CityJSON's nested hierarchies, we achieve significant performance improvements. The format supports HTTP Range Requests, allowing clients to fetch only required portions of the file. Performance evaluation shows FlatCityBuf achieves up to 60× faster read times than CityJSONSeq for large datasets while reducing file sizes by 15-50%. Our web prototype demonstrates practical application by allowing interactive queries on a 4GB dataset covering a 20km×20km area without downloading the entire file. FlatCityBuf bridges the gap between CityJSONSeq and optimized binary formats, enabling scalable, real-time urban data processing while maintaining CityJSON's semantic richness.

3D point clouds used in geospatial applications typically contain billions of points. Processing 3D point clouds of this size as a whole with deep learning models requires computational resources (e.g., GPU memory) that are usually not available. To obtain 3D point clouds that can be processed by deep learning models, sampling methods that produce local subsets of large-scale 3D point clouds with a smaller extent or lower density are essential. Nonetheless, the impact of different input sampling methods on the semantic segmentation performance of deep learning models has received little attention so far. In this paper, we compare three widely used input sampling techniques (random sampling, farthest point sampling, and grid sampling) concerning the semantic segmentation performance of different deep learning architectures, using inputs of different spatial extents. We consider both indoor and outdoor scenarios, using the Stanford Large-Scale 3D Indoor Spaces and Paris-CARLA-3D datasets as reference datasets. We find that random and grid sampling outperform farthest point sampling in terms of segmentation performance, while random sampling displays the fastest execution time. For indoor scenarios, using input 3D point clouds with a small spatial extent and higher density yields the best results. For outdoor scenarios, similar performance is obtained for all tested input extents. In an additional experiment, we evaluate a curvature-weighted sampling approach to test whether geometric features derived from 3D point clouds can guide the selection of more informative input points for deep learning models. However, we find that using curvature as a sampling criterion decreases the segmentation performance, indicating a mismatch between the expected relevance of high-curvature points (e.g., points representing object borders) and the internally learned features of the deep learning models.

The CityGML Energy Application Domain Extension (ADE) was released in 2018 with the purpose to offer an open and standardised data model to facilitate multi-scale Urban Energy Modelling applications. The Energy ADE is based on and extends the international open standard CityGML 2.0. It has been used since its release in several national and international projects, mainly focusing on the simulation and computation of the building energy performance based on the integration of semantic 3D city models and several other sources of information. The technological innovations (e.g. the release of CityGML 3.0 in 2021) as well as experiences and feedback collected in the years since its release have contributed to forge several new ideas to improve, enhance and update the Energy ADE. Since 2024, work has been in progress to harmonise and implement such ideas, targeting towards a so-called Energy ADE 2.0. This paper provides an overview of the development process of the conceptual model so far, and presents a selection of the major changes and improvements that have been made to the original data model of Energy ADE.