TY - CHAP A1 - Chizhova, Maria A1 - Gurianov, Andrey A1 - Korovin, Dmitrii A1 - Brunn, Ansgar A1 - Stilla, Uwe T1 - Automatically Extraction and Reconstruction of Cupola Geometries of Orthodox Churches from Precision Point Clouds T2 - 37. Wissenschaftlich-Technische Jahrestagung der DGPF in Würzburg – Publikationen der DGPF N2 - Complex geometry extraction from point clouds is an actual problem in reverse engineering. Simple geometrical models (like parallelepipeds, prisms, pyramids, cones, spheres) were already applied in construction and machine-building modeling, but are not sufficient for high quality BIM now. This work, which is carried out in the context of virtual reconstruction of destroyed orthodox churches, presents a robust and efficient method of cupola (domes) and tambour geometry extraction from precise point clouds. The rich diversity of architectural forms, which are defined by many parameters, does not allow to consider this problem as a trivial duty, because usual geometry extraction methods fail for these object types. The new developed algorithm is presented and realized Y1 - 2017 UR - http://www.dgpf.de/src/tagung/jt2017/proceedings/proceedings/papers/42_Pos_DGPF2017_Chizhova_et_al.pdf IS - 26 SP - 434 EP - 443 ER - TY - JOUR A1 - Brodovskii, Maksim A1 - Korovin, Dmitrii A1 - Chizhova, Maria A1 - Brunn, Ansgar A1 - Stilla, Uwe T1 - Recognizing Separate Structural Elements of Churches Using Neural Network JF - Dreiländertagung der DGPF, der OVG und der SGPF in Bern, Schweiz – Publikationen der DGPF N2 - In this paper we develop a new approach to recognize structural elements of orthodox churches. We will work with 3D point clouds, received as a result of 3D point cloud acquisitions of churches, e.g. from laser scanning. Because of the large amount of points in such clouds, we have to use a projection (elevation) to decrease the calculation effort. To get meaningful images from the projection of the point cloud we do some prior segmentation of the 3D cloud. Images binary, with a predefined resolution that depends on the resolution of the 3D point cloud. To recognize elements we decide to use neural networks (Perceptron and Counter propagation neural networks) as they allow the automation of the process and have a broad range of methods to recognize images. For the subsequent 3D modeling we use analytic expressions, that describe each of the structural church elements. A further step is the deduction of those expressions that describe each of the recognized sectional views Y1 - 2016 UR - http://www.dgpf.de/src/tagung/jt2016/proceedings/papers/15_DLT2016_Brodovskii_et_al.pdf ER -