标题: Indoor interior segmentation with curved surfaces via global energy optimization

作者: Su, F (Su, Fei); Zhu, HH (Zhu, Haihong); Li, L (Li, Lin); Zhou, G (Zhou, Gang); Rong, W (Rong, Wei); Zuo, XK (Zuo, Xinkai); Li, WD (Li, Wende); Wu, XM (Wu, Xinmei); Wang, WL (Wang, Weilin); Yang, F (Yang, Fan); Hu, HJ (Hu, Huanjun); Ying, S (Ying, Shen)

来源出版物: AUTOMATION IN CONSTRUCTION   卷:131 DOI: 10.1016/j.autcon.2021.103886 出版年: NOV 2021  

摘要: Most existing indoor interior segmentation methods typically focus on planar structures rather than curved structures. Random sample consensus-based methods perform curved surface segmentation via regularity model fitting but suffer from spurious model generation when noise and outliers are present due to the uncertainty in the random sampling. This paper formulates indoor interior segmentation by fitting representative models and matching each primary cell with corresponding model simultaneously. The models are fitted by a combination of points and supervoxels with adaptive resolutions instead of just points, guaranteeing the correctness of sampling on the same surface and avoiding spurious models. Cell-to-model matching is achieved by iterative refinement/ clustering under the global energy optimization method, which ensures optimal overall segmentations. Experimental tests demonstrate the effectiveness of our method in dealing with both planar and nonplanar surfaces, resulting in performance metrics of approximately 0.75 for the structure F1-score and over 0.9 for edge precision and recall.

作者关键词: Point cloud; Indoor segmentation; Curved structure; Supervoxel; Global energy optimization

地址: [Su, Fei; Zhu, Haihong; Li, Lin; Zhou, Gang; Rong, Wei; Zuo, Xinkai; Li, Wende; Wu, Xinmei; Wang, Weilin; Yang, Fan; Hu, Huanjun; Ying, Shen] Wuhan Univ, Sch Resource & Environm Sci, 129 Luoyu Rd, Wuhan 430079, Peoples R China. [Su, Fei] Shandong Jianzhu Univ, Sch Surveying & Geo Informat, Jinan 250101, Peoples R China.

通讯作者地址: Zhu, HH; Li, L; Ying, S (通讯作者)，Wuhan Univ, Sch Resource & Environm Sci, 129 Luoyu Rd, Wuhan 430079, Peoples R China.

电子邮件地址: sftx016@whu.edu.cn; hhzhu@whu.edu.cn; lilin@whu.edu.cn; zhou_gang@whu.edu.cn;rongw@whu.edu.cn;zuoxinkai2012@whu.edu.cn;wendeli@whu.edu.cn; xinmwu@whu.edu.cn; wangweilin@whu.edu.cn; yhlx125@whu.edu.cn; huhuanjun@whu.edu.cn; shy@whu.edu.cn

影响因子：7.7