结合深度学习与规则特征的立面窗户检测方法  

Combining deep learning and regular features for facade windows detection

作　　者：钱建国[1] 张宇[1,2] 王伟玺 谢林甫 李晓明 汤圣君 QIAN Jianguo;ZHANG Yu;WANG Wei;XIE Linfu;LI Xiaoming;TANG Shengjun(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China;Smart City Research Institute,School of Architecture and Urban Planning,Shenzhen University&Amp,Key Laboratory of Urban Natural Resources Monitoring and Simulation,Ministry of Natural Resources&Amp,Shenzhen Key Laboratory of Spatial Information Intelligent Perception and Service&Amp,Guangdong Key Laboratory of Urban Spatial Information Engineering,Shenzhen,Guangdong 518061,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000 [2]深圳大学建筑与城市规划学院智慧城市研究院,自然资源部城市自然资源监测与仿真重点实验室,深圳市空间信息智能感知与服务重点实验室,广东省城市空间信息工程重点实验室,广东深圳518061

出　　处：《测绘科学》2023年第8期94-101,共8页Science of Surveying and Mapping

基　　金：国家自然科学基金项目(42001407,41971341,41971354)。

摘　　要：针对现有通过检测窗户角点实现窗户检测方法中存在窗户误检的问题,该文在窗角点分组阶段,以建筑物立面窗户的分布规律及其自身的几何结构特征为依据,提出一种参数自适应的窗角点分组方法。该方法是在使用深度学习方法获取窗户4个角点坐标的基础上,结合窗户角点及其连线的空间位置关系、平行垂直关系,建立窗角点分组判别依据,实现对窗角点检测结果的准确划分,进而得到有效窗户检测结果。为验证该方法的有效性,选用4个公开数据集进行窗户检测实验,结果表明:该方法可有效支持多类图像数据、实现全自动化运行,且与现有方法相比,具有更高的检测精度。Aiming at the problem of window misdetection in the existing window detection method by detecting window corner points,this paper proposed a parameter adaptive window corner point grouping method in the window corner point grouping stage,based on the distribution pattern of windows on the building facade and its geometric structure characteristics.The method was based on the use of deep learning methods to obtain the coordinates of the four corner points of windows and combines the spatial position relationship and parallel-perpendicular relationship of window corner points and their connecting lines to establish the basis for window corner point grouping discrimination,to achieve accurate classification of window corner point detection results,and then obtained effective window detection results.To verify the effectiveness of the method,four publicly available datasets were selected for window detection experiments.The results showed that the method could effectively support multiple types of image data,achieved fully automated operation,and had higher detection accuracy compared with existing methods.

关 键 词：深度学习 建筑物立面 窗户检测 规则性特征 

分 类 号：P237[天文地球—摄影测量与遥感]