基于分块三维高斯泼溅的城市低空风险地图构建方法

张建平; 王致远; 张光远; 罗创; 陈运翔

doi:10.19818/j.cnki.1671-1637.2026.155

基于分块三维高斯泼溅的城市低空风险地图构建方法

doi: 10.19818/j.cnki.1671-1637.2026.155

张建平^{1, 2,},
王致远^{1, 2},
张光远^{1, 2, ,},
罗创^{1, 2},
陈运翔^{1, 2}

1.
西南交通大学交通运输与物流学院, 四川成都 610031
2.
低空交通智能管控四川省重点试验室, 四川成都 610031

基金项目:

国家重点研发计划 2022YFB4300903

国家自然科学基金民航联合研究基金重点项目 U2433217

国家自然科学基金项目 52472332

四川省重大科技专项 2024ZDZX0044

四川省自然科学基金项目 2025ZNSFSCO394

详细信息

作者简介:
张建平(1976-)，男，安徽芜湖人，研究员，博士生导师，工学博士，E-mail：zhangjp@swjtu.edu.cn

通讯作者:
张光远(1979-)，男，辽宁庄河人，正高级实验师，博士生导师，工学博士，E-mail：gyzhang@swjtu.edu.cn

中图分类号: U8
计量
- 文章访问数: 20
- HTML全文浏览量: 13
- PDF下载量: 4
- 被引次数: 0
出版历程
- 收稿日期: 2025-08-26
- 录用日期: 2026-01-23
- 修回日期: 2025-11-27
- 刊出日期: 2026-03-28

Construction method of urban low-altitude risk map based on block 3D Gaussian splatting

ZHANG Jian-ping^{1, 2
,},
WANG Zhi-yuan^{1, 2},
ZHANG Guang-yuan^{1, 2
, ,},
LUO Chuang^{1, 2},
CHEN Yun-xiang^{1, 2}

1.
School of Transportation and Logistics, Southwest Jiaotong University, Chengdu 610031, Sichuan, China
2.
Low-altitude Traffic Intelligent Control Key Laboratory of Sichuan Province, Chengdu 610031, Sichuan, China

Funds:

National Key R&D Program of China 2022YFB4300903

Key Project of the Civil Aviation Joint Research Fund of National Natural Science Foundation of China U2433217

National Natural Science Foundation of China 52472332

Sichuan Provincial Major Science and Technology Project 2024ZDZX0044

Natural Science Foundation of Sichuan Province 2025ZNSFSCO394

More Information

Corresponding author: ZHANG Guang-yuan, senior experimentalist, PhD, E-mail: gyzhang@swjtu.edu.cn

Article Text (Baidu Translation)

摘要

摘要: 提出了城市低空三维风险地图构建方法，实现地形、人文、气象多个风险因子统一建模与表达；研究了基于分块三维高斯泼溅的无人机影像大区域重建，确保几何与光度连续，并派生数字表面模型、数字高程模型及形态指标，将人口密度和分层风速、风切变数据与地形因子在统一体素网格对齐；建立了对数线性池化与质量图加权融合的三维风险场，采用分位阈值形成风险等级；输出了二维热力、三维体渲染效果，并在成都市某新区约15 km²区域开展了试验。研究结果表明：分块一致性重建策略有效减少重建接缝与几何台阶，使峰值信噪比由27.8 dB提升至28.6 dB，派生地形指标分布更平滑；地形、人文、气象因子互补作用明显，融合后的风险分布合理，核心区风险超过可接受阈值的概率为7.4%，显著高于外围区域的1.5%；二维切片与三维体渲染均展现出空间连通性，城市核心区在30、80、120、200、300 m的综合风险分别为0.68、0.61、0.53、0.56、0.57；消融试验进一步验证了块间一致性、曲率特征、场景掩模、风切变与概率校准等模块的重要性。
- 低空交通 /
- 空域规划 /
- 城市低空风险地图 /
- 分块三维重建 /
- 三维高斯泼溅 /
- 地形因子 /
- 人文因子 /
- 气象因子
Abstract: An urban low-altitude three-dimensional (3D) risk mapping method was proposed to achieve unified modelling and representation of multiple risk factors, including terrain, human, and meteorological factors. A large-area unmanned aerial vehicle (UAV) image reconstruction approach based on block 3D Gaussian Splatting (3DGS) was investigated to ensure geometric and photometric continuity. A digital surface model (DSM), a digital elevation model (DEM), and morphometric indices were derived. Population density, layered wind speed, wind shear, and terrain factors were then aligned within a unified voxel grid. A 3D risk field was established using logarithmic linear pooling and quality-map weighted fusion, and risk levels were formed using quantile-based thresholds. Two-dimensional (2D) heat maps and 3D volume-rendering visualizations were generated. Experiments were conducted in a 15 km² area of a new district in Chengdu. Research results indicate that the block-consistency reconstruction strategy effectively reduces reconstruction seams and geometric steps, increases the peak signal-to-noise ratio (PSNR) from 27.8 dB to 28.6 dB, and produces smoother distributions of the derived terrain indices. The complementary effects of terrain, human, and meteorological factors are evident, with a reasonable fused risk distribution. The probability that the core area risk exceeds the acceptable threshold is 7.4%, which is significantly higher than 1.5% in the peripheral area. Both 2D slices and 3D volume rendering show spatial connectivity. The comprehensive risks of the urban core area at 30, 80, 120, 200, and 300 m are 0.68, 0.61, 0.53, 0.56, and 0.57, respectively. Ablation experiments further confirm the importance of modules such as inter-block consistency, curvature features, scene masks, wind shear, and probability calibration.
- low-altitude traffic /
- airspace planning /
- urban low-altitude risk map /
- block-based 3D reconstruction /
- 3D Gaussian splatting /
- terrain factor /
- human factor /
- meteorological factor

HTML全文

图 1 场景表示与可微渲染模型示意

Figure 1. Schematic of scene representation and differentiable rendering model

下载: 全尺寸图片幻灯片

图 2 自适应分块策略示意

Figure 2. Schematic of adaptive blocking strategy

下载: 全尺寸图片幻灯片

图 3 场景局部地形模型

Figure 3. Scene local terrain model

下载: 全尺寸图片幻灯片

图 4 城市新区无人机俯拍切片

Figure 4. UAV overhead photographys of the new urban area

下载: 全尺寸图片幻灯片

图 5 人口密度栅格裁剪研究区域

Figure 5. Study area of population density raster clipping

下载: 全尺寸图片幻灯片

图 6 分块3DGS与其余2种算法效果的对比

Figure 6. Results comparison of 3DGS with the other two algorithms

下载: 全尺寸图片幻灯片

图 7 地形迫降风险项数据可视化图

Figure 7. Visualization diagrams of terrain emergency landing risk item data

下载: 全尺寸图片幻灯片

图 8 二维风险热力图与三维渲染地图

Figure 8. 2D risk heatmap and 3D volume rendering map

下载: 全尺寸图片幻灯片

图 9 不同高度风险切片图

Figure 9. Slice maps of different risk levels

下载: 全尺寸图片幻灯片

表 1 近地面风数据

Table 1. Near-surface wind data

高度层/m	平均风速/ (m·s^-1)	风向/(°)	垂直风切变/ [m·s^-1/(100 m)]
30	2.8	15	2.4
80	4.1	25	2.6
120	5.0	30	2.3
200	6.0	38	1.3
300	6.9	45	0.9

下载: 导出CSV

表 2 试验设置

Table 2. Experimental setup

类别	项目	设置/取值	说明
采集与场景	研究区域	成都市某新区住商混合片区，约15 km²	同时覆盖高层建筑群、学校、医院、滨河绿地与交通走廊
	航线与时段	弓字形覆盖	地面站航线规划规划
	作业条件	选择能见度良好、近地面风力较小时段	降低气象噪声对重建与风险估计的干扰
空间组织	坐标与网格	统一到通用坐标与高程基准，建立统一三维网格(水平到街区尺度，垂向覆盖常见飞行高度带)	为因子对齐、融合与多视图表达提供公共底板
三维重建与质量	重建方式	端到端分块三维重建全域训练	兼顾大范围覆盖与块间连续性
三维重建与质量	地形派生	由重建结果派生DSM/DTM，计算坡度、曲率、形状指数等	概括起伏、遮挡、通道，用于地形因子建模
因子构建与对齐	地形因子	基于DSM/DTM的形态指标计算三维地形迫降风险项G(x, y, z)	反映遮挡与地表复杂度
	人文因子	年度人口栅格，裁剪至研究区并限制在三维场景柱向重叠域U(x, y)	避免场景外的伪热点与阈值漂移
	气象因子	气象平台地区风速与风切变，插值到统一高度带	国家气象台公开数据，体现高度相关的扰动差异
风险融合与分级	融合计算	在统一网格按本文总公式融合与质量图Q得到R(x, y, z)∈[0, 1]	保持单调与可解释，可做概率刻度校准
风险融合与分级	分级与阈值	以全域分位为初值，结合管理需求小幅调整	兼顾跨区域可比与本区处理目标
地图表达与合规	二维、三维视图	定高热力(等级控色、强度控透明度)、三维体渲染	保持多视图语法一致，方便审读
评估与复现	开放材料	提供示例切片、因子层与图件示例	便于同专业学者在相近场景复现

下载: 导出CSV

表 3 三维重建试验定量指标

Table 3. Quantitative indicators of 3D reconstruction experiments

指标类别	指标	带约束的分块3DGS	逐块独立的3DGS	SfM+MVS
重建层	PSNR/dB	28.6	27.8
	SSIM	0.918	0.902	0.673
	LPIPS	0.152	0.198	0.244
	几何漂移Δz/m	0.042	0.091	0.094
派生层	高程全域偏差/m	0.18	0.29	0.35
	高程标准差/m	0.21	0.37	0.39
	坡度分布一致性	0.946	0.887	0.802
	曲率分布一致性	0.931	0.865	0.829

下载: 导出CSV

表 4 二维形态风险数值

Table 4. 2D morphological risk values

区域	地形风险	顶面风险	二维形态风险	风险值大于可接受范围概率P(R>0.8)/%
1	0.64	0.53	0.61	10.2
2	0.61	0.46	0.52	7.8
3	0.45	0.39	0.43	4.8
4	0.21	0.14	0.19	1.9
5	0.16	0.12	0.14	1.6

下载: 导出CSV

表 5 人口暴露风险数值

Table 5. Population exposure risk value

区域	人口暴露风险	风险值大于可接受范围概率P(R>0.8)/%
1	0.67	14.1
2	0.63	12.4
3	0.45	4.8
4	0.22	0.5
5	0.18	0.3

下载: 导出CSV

表 6 高度相关风场风险数值

Table 6. Risk values of height dependent wind field

高度/m	风速强度	垂直风切变强度	气象风险	风险值大于可接受范围概率P(R>0.8)/%
30	0.22	0.60	0.38	1.2
80	0.48	0.68	0.52	3.5
120	0.61	0.55	0.58	5.8
200	0.72	0.41	0.62	6.9
300	0.81	0.30	0.65	7.6

下载: 导出CSV

表 7 不同高度层下的综合风险数值

Table 7. Comprehensive risk values at different elevation levels

区域	30 m风险	80 m风险	120 m风险	200 m风险	300 m风险	风险值大于可接受范围概率P(R>0.8)/%
1	0.68	0.61	0.53	0.56	0.57	7.4
2	0.61	0.56	0.50	0.53	0.54	5.8
3	0.54	0.48	0.44	0.46	0.47	3.9
4	0.47	0.42	0.39	0.40	0.40	2.4
5	0.41	0.37	0.34	0.35	0.35	1.5

下载: 导出CSV

表 8 消融试验设置与结果

Table 8. Ablation study setup and results

编号	模块	操作内容	IoU	L1
A0	完整模型	全模块启用(块间一致性、曲率与形状指数、人口场景掩模、风速与切变、几何池化)	0.642	0.124
A1	块间一致性	将3DGS的块间一致性损失置零，各块独立训练	0.497	0.189
A2	地形形态	去曲率(仅保留坡度)，移除曲率强度与形状指数	0.562	0.158
A3	场景掩模	人口栅格不再裁剪到场景柱向范围内	0.545	0.171
A4	气象建模	去切变(仅保留风速)	0.571	0.149
A5	融合策略	将几何池化改为线性加权	0.553	0.163

下载: 导出CSV

参考文献(28)

[1]	张洪海, 吕文颖, 万俊强, 等. 扇区空中交通风险态势网络建模与演化特征[J]. 交通运输工程学报, 2023, 23(1): 222-241. doi: 10.19818/j.cnki.1671-1637.2023.01.017 ZHANG Hong-hai, LYU Wen-ying, WAN Jun-qiang, et al. Network modeling and evolution characteristics for air traffic risk situation in sectors[J]. Journal of Traffic and Transportation Engineering, 2023, 23(1): 222-241. doi: 10.19818/j.cnki.1671-1637.2023.01.017
[2]	谢华, 韩斯特, 尹嘉男, 等. 城市低空无人机飞行计划协同推演与优化调配方法[J]. 航空学报, 2024, 45(19): 263-285. XIE Hua, HAN Si-te, YIN Jia-nan, et al. Cooperative deduction and optimal allocation method for urban low-altitude UAV flight plan[J]. Acta Aeronautica et Astronautica Sinica, 2024, 45(19): 263-285.
[3]	钟罡, 华骏鸣, 杜森, 等. 基于复杂网络的城市低空飞行计划优化调度[J]. 航空学报, 2025, 46(11): 246-268. ZHONG Gang, HUA Jun-ming, DU Sen, et al. Urban low-altitude flight plan optimal scheduling based on complex network[J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(11): 246-268.
[4]	HAN P, YANG X Y, ZHAO Y F, et al. Quantitative ground risk assessment for urban logistical unmanned aerial vehicle (UAV) based on Bayesian network[J]. Sustainability, 2022, 14(9): 5733. doi: 10.3390/su14095733
[5]	杨艺苑, 杨存建. 基于GIS的东川区生态环境敏感性分析[J]. 测绘通报, 2022(3): 7-12. YANG Yi-yuan, YANG Cun-jian. Sensitivity analysis of ecological environment in Dongchuan district based on GIS[J]. Bulletin of Surveying and Mapping, 2022(3): 7-12.
[6]	陈运翔, 张建平, 王致远, 等. 基于机动避撞策略的低空多旋翼无人机安全间隔计算模型[J]. 航空学报, 2025, 46(11): 349-365. CHEN Yun-xiang, ZHANG Jian-ping, WANG Zhi-yuan, et al. Safety separation calculation model for multi-rotor drones in low-altitude airspace based on avoidance strategy[J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(11): 349-365.
[7]	GRASER A, STRAUB M, DRAGASCHNIG M. Towards an open source analysis toolbox for street network comparison: Indicators, tools and results of a comparison of OSM and the official Austrian reference graph[J]. Transactions in GIS, 2014, 18(4): 510-526. doi: 10.1111/tgis.12061
[8]	王俊潼, 包丹文, 周佳怡, 等. 低空空域规划研究现状与展望[J]. 航空学报, 2025, 46(11): 82-107. WANG Jun-tong, BAO Dan-wen, ZHOU Jia-yi, et al. Low-altitude airspace planning: A review and prospect[J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(11): 82-107.
[9]	李诚龙, 屈文秋, 李彦冬, 等. 面向eVTOL航空器的城市空中运输交通管理综述[J]. 交通运输工程学报, 2020, 20(4): 35-54. doi: 10.19818/j.cnki.1671-1637.2020.04.003 LI Cheng-long, QU Wen-qiu, LI Yan-dong, et al. Overview of traffic management of urban air mobility (UAM)with eVTOL aircraft[J]. Journal of Traffic and Transportation Engineering, 2020, 20(4): 35-54. doi: 10.19818/j.cnki.1671-1637.2020.04.003
[10]	MILDENHALL B, SRINIVASAN P P, TANCIK M, et al. NeRF: Representing scenes as neural radiance fields for view synthesis[M]. Berlin: Springer International Publishing, 2020.
[11]	MVLLER T, EVANS A, SCHIED C, et al. Instant neural graphics primitives with a multiresolution hash encoding[J]. ACM Transactions on Graphics, 2022, 41(4): 1-15.
[12]	TANCIK M, CASSER V, YAN X C, et al. Block-NeRF: Scalable large scene neural view synthesis[C]//IEEE. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2022: 8238-8248.
[13]	TURKI H, RAMANAN D, SATYANARAYANAN M. Mega-NeRF: Scalable construction of large-scale NeRFs for virtual fly-throughs[C]//IEEE. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2022: 12912-12921.
[14]	KERBL B, KOPANAS G, LEIMKUEHLER T, et al. 3D Gaussian splatting for real-time radiance field rendering[J]. ACM Transactions on Graphics, 2023, 42(4): 1-14.
[15]	YU A, LI R L, TANCIK M, et al. PlenOctrees for real-time rendering of neural radiance fields[C]//IEEE. 2021 IEEE/CVF International Conference on Computer Vision (ICCV). New York: IEEE, 2021: 5732-5741.
[16]	KULHANEK J, RAKOTOSAONA M J, MANHARDT F, et al. LODGE: Level-of-detail large-scale Gaussian splatting with efficient rendering[EB/OL]. (2025-10-29). https://arxiv.org/abs/2505.23158.
[17]	MELNYK R, SCHRAGE D, VOLOVOI V, et al. A third-party casualty risk model for unmanned aircraft system operations[J]. Reliability Engineering & System Safety, 2014, 124: 105-116.
[18]	YANG Z Y. Large-eddy simulation: Past, present and the future[J]. Chinese Journal of Aeronautics, 2015, 28(1): 11-24. doi: 10.1016/j.cja.2014.12.007
[19]	ZHAO W J, ZHANG N, SUN J N, et al. Evaluation and parameter-sensitivity study of a single-layer urban canopy model (SLUCM) with measurements in Nanjing, China[J]. Journal of Hydrometeorology, 2014, 15(3): 1078-1090. doi: 10.1175/JHM-D-13-0129.1
[20]	MARONGA B, GRYSCHKA M, HEINZE R, et al. The Parallelized Large-eddy Simulation Model (PALM) version 4.0 for atmospheric and oceanic flows: Model formulation, recent developments and future perspectives[J]. Geoscientific Model Development, 2015, 8(8): 2515-2551. doi: 10.5194/gmd-8-2515-2015
[21]	YANG Y, LIU D W, LU L S. A study of wind shear influences on the aerodynamic performances of a UAV airfoil[J]. Applied Sciences, 2023, 13(6): 3764. doi: 10.3390/app13063764
[22]	陈艺君, 余莎莎, 张学军. 城市低空场景下无人机运行对地风险量化评估[J]. 北京航空航天大学学报, 2025, 51(3): 806-815. CHEN Yi-jun, YU Sha-sha, ZHANG Xue-jun. Ground risk quantitative assessment for UAV operations in urban low-altitude scenarios[J]. Journal of Beijing University of Aeronautics and Astronautics, 2025, 51(3): 806-815.
[23]	羊钊, 齐洪彪, 于阳阳, 等. 风险规避与组合策略融合的多无人机协同路径规划方法[J]. 交通运输工程学报, 2026, 26(3): 140-158. YANG Zhao, QI Hong-biao, YU Yang-yang, et al. Multi-UAV cooperative path planning method integrating risk avoidance and combined strategy[J]. Journal of Traffic and Transportation Engineering, 2026, 26(3): 140-158.
[24]	屈若锟, 王致远, 刘晔璐, 等. 面向城市空中交通的无人机视觉定位技术[J]. 航空学报, 2025, 46(11): 379-394. QU Ruo-kun, WANG Zhi-yuan, LIU Ye-lu, et al. UAV visual positioning technology for urban air mobility[J]. Acta Aeronautica et Astronautica Sinica, 2025, 46(11): 379-394.
[25]	SCHÖNBERGER J L, FRAHM J M. Structure-from-motion revisited[C]//IEEE. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2016: 4104-4113.
[26]	SEITZ S M, CURLESS B, DIEBEL J, et al. A comparison and evaluation of multi-view stereo reconstruction algorithms[C]// IEEE. 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2006: 519-528.
[27]	REZATOFIGHI H, TSOI N, GWAK J, et al. Generalized intersection over union: A metric and a loss for bounding box regression[C]//IEEE. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2019: 658-666.
[28]	SONG Z, WOODRUFF D P, ZHONG P L. Low rank approximation with entrywise L1-norm error[C]//ACM. Proceedings of the 49th Annual ACM SIGACT Symposium on Theory of Computing. New York: ACM, 2017: 688-701.