城区场景建筑物单体分割与结构重建耦合研究
Coupled study on instance segmentation and structural reconstruction of buildings in 3D urban scenes
董梦成 1谢科 1黄惠1
作者信息
- 1. 深圳大学计算机与软件学院可视计算研究中心,深圳 518060
- 折叠
摘要
随着无人机以及激光雷达在测绘领域的大规模应用,大量的城市场景稠密三维网格模型被生产出来,对其进行建筑物单体分割与结构化重建是计算机图形学领域极具挑战性的研究问题.本文对此提出了一种城区场景建筑物单体分割与结构化重建耦合处理的方法.首先,本文基于马尔可夫(Markov)随机场从场景中提取建筑物轮廓;然后,基于提取出的建筑物轮廓以及人造建筑竖直方向的规则性,引入了 一种新的建筑物结构化模型——多层轮廓模型;最后,设计了一个自顶向下的城区场景建筑物单体分割与结构化重建耦合处理的并行系统,该系统能够高效地处理大规模城区场景.实验结果表明,本文方法生成的结构化模型最高实现了厘米级的精度.
Abstract
With the widespread application of drones and LiDAR in the field of surveying and mapping,a vast number of dense three-dimensional mesh models of urban scenes have been generated.Segmenting these models into individual buildings and performing structural reconstruction are challenging research problems in the field of computer graphics.We propose a method for the coupled processing of instance segmentation and structural reconstruction of buildings in urban scenes.Firstly,we utilize a Markov random field to extract building outlines from the scene.Then,based on the extracted building outlines and the regularity in the vertical direction of man-made buildings,a new structural multi-layer contour model for buildings is constructed.Lastly,a top-down parallel system is developed for the coupled processing of instance segmentation and structural reconstruction of buildings.This system can efficiently process large-scale urban scenes.Experimental results demonstrate that the structural models generated by our method can achieve centimeter-level accuracy at best.
关键词
建筑物单体分割/结构重建/耦合处理/城市场景理解/马尔可夫随机场Key words
building instance segmentation/structural reconstruction/coupled processing/urban scene understanding/Markov random field引用本文复制引用
基金项目
国家自然科学基金(U2001206)
国家自然科学基金(U21B2023)
国家自然科学基金(62161146005)
广东省高等学校城市三维创新团队项目(2022KCXTD025)
深圳市科技创新项目(KQTD20210811090044003)
深圳市科技创新项目(RCJC20200714114435012)
深圳市科技创新项目(JCYJ20210324120213036)
人工智能与数字经济广东省实验室(深圳)资助项目()
出版年
2024
NETL
NSTL
万方数据