基于倾斜三维数据的建筑物分层方法
A Method of Building Layering Based on Oblique Photogrammetry 3D Data
张祖宇 1孙时钟 1温久民 1张冠 1高云龙2
作者信息
- 1. 广西壮族自治区地理信息测绘院,广西壮族自治区 柳州,545000
- 2. 武汉大学测绘遥感信息工程国家重点实验室,湖北 武汉,430079
- 折叠
摘要
以单体建筑为对象的属性挂接方法已无法满足城市建筑立体空间多层差异的属性信息挂接需求.因此,提出一种基于倾斜三维数据的建筑物智能楼层提取方法,该方法利用现有二维建筑物矢量边界提取建筑物立面纹理,并在Mask R-CNN模型上增加一个特征增强结构——反向的特征金字塔(feature pyramid network,FPN),充分利用高低层特征信息,提升窗户识别检测率;同时,根据窗户排列规则进行规则化补全,然后进行层高计算、分割楼层.实验证明所提方法容错性较好,即使在窗户识别不全或有遮挡的情况下,经过简单的后处理也能实现楼层的分层.
Abstract
The attribute linking method that takes the single building as the object can no longer meet the demand of attri-bute information linking for the multi-layer difference in 3D space of urban buildings.Therefore,we propose an intelli-gent method to extract building floors based on oblique photo-grammetry 3D data.This method uses the existing 2D vector boundary data of the buildings to extract the facade texture of the buildings,and adds a feature enhancement structure,re-verse feature pyramid network(FPN)to Mask R-CNN model to make full use of the feature information of high and low floors to improve the window recognition and detection rate.According to the rules of window arrangement,regular completion is performed,and the floor height is calculated and the floors are divided.The experiment verifies the pro-posed method has better fault tolerance.Even when the win-dows are not fully identified or blocked,the floor layering can be achieved through simple post-processing.
关键词
建筑物分层/属性挂接/深度学习/窗户提取/规则补全Key words
building layering/attribute linking/deep learning/window extraction/rule completion引用本文复制引用
基金项目
国家重点研发计划(2019YFC1520105)
应急管理部消防救援局科技项目(2020xfzd15)
出版年
2024
NETL
NSTL
万方数据