Since the 20th century, heritage preservation and protection regarding forms both tangible and intangible have received growing and enthusiastic international attention [1
]. Several important principles were promulgated concomitantly during this time period, where the most significant guideline was the International Charter for the Conservation and Restoration of Monuments and Sites, commonly referred to as the Venice Charter 1964, which set a remarkable benchmark for principles that govern architectural conservation and restoration [9
]. Currently, a unique focus is placed on surrogating fragile objects, such as ancient architecture [10
]. Digital modeling is one of the most effective and available measures toward achieving this goal.
Ancient Chinese architecture, as an art, is the essence of Chinese cultural heritage [11
]. It has an indigenous and unique system of construction that has retained its principal characteristic from prehistoric times to the present and has spread its influence to other countries, such as Korea and Japan [12
], as shown in Figure 1
a–c. Unfortunately, a mass of the artistic architectures disappeared due to the brittle timber structures and natural or human-made disasters. 3D digital modeling technology hence has become increasingly necessary to understand and restore them. Additionally, curved roofs act as the peculiarly characteristic and ornate parts and are the most complicated and difficult to construct, although they are aesthetically important. That is also why current Chinese antique buildings (Figure 1
d) are composed of a traditional roof and a modern body. To this end, this paper places the research emphasis on 3D digital modeling for ancient Chinese-style architectural roofs.
Modeling buildings has been performed by scholars adopting various processing methods that can be classified into two fundamental categories: data-driven and model-driven methods [13
]. Data-driven methods utilize a LiDAR point cloud dataset for a primary data source, which utilizes several segmentation algorithms, such as region growing (RG) [14
], the 3D Hough transform [15
], feature clustering [16
], model fitting [17
] and random sample consensus (RANSAC) [18
]. Polyhedral building models are generated from the planar segmented patches through intersection and step-edge generation and are then regulated by certain construction rules to improve the shape. Although building models, even those that incorporate complex roof shapes, can be reconstructed using the above approaches, the modeling process and corresponding object structures may be hampered and deformed when certain features lack original data [20
]. To avoid this drawback, aerial imagery was introduced as supplemental data to improve the accuracy because of its high resolution. Therefore, certain studies have used both LiDAR data and aerial imagery for building reconstruction during the last decade [21
]. Because of this improved process, building models can incorporate acceptable shape effects, although they are always geometric volumes that disregarded important semantic information about the structures.
To obtain visually-realistic and topologically-accurate building models that include semantic descriptions, a model-driven strategy was proposed. Certain scholars have applied constructive solid geometry (CSG)-based methods, which combine simple basic primitive structures (cube, cylinder, cone, sphere, etc.) and use Boolean operators (intersection, union, subtraction, etc.) for building modeling [26
]. Thus, a complex building, or a structure that includes a complex roof, could be reconstructed by assembling basic primitive structures that were designed in a library. The models grouped by CSG primitive structures that include few control parameters possess semantic information, and their processes are relatively easy to apply. However, properly decomposing complex buildings into CSG primitives is open and sophisticated, and the CSG modeling process generally requires manual adjustments.
The validity of assembling primitives into the correct building style and structure is another challenge, which may be addressed by incorporating semantic constraints [30
]. CityGML [31
] is the standard for the representation and exchange of 3D city models and contains rich information in terms of geometry, semantics and topology. With respect to semantic topology, CityGML adopts a simple XLinkconcept to represent the interrelations between different geometric aggregates or thematic features [32
]. Gröger and Plümer extended the existing axiomatic characterization of 3D surfaces to guarantee the semantic topological consistency of semantic objects, such as bridges and tunnels, in 3D city models [33
]. A semantics-constrained profiling approach was proposed that ensured the consistency of the geometrical, topological and semantic relationships when profiling complex 3D city models [34
]. Nevertheless, topological relationships between the semantic features were not improved or detailed when these approaches were used. To develop a more comprehensive topology in CityGML, a two-level topological model (semantic level and geometric level) is proposed to represent 3D topological relationships [35
In alignment with the above-mentioned studies and considering that the structure of ancient Chinese-style architectural roofs is fairly complicated and includes a variety of detailed components, including a uniquely-distinguished Chinese-style and time-honored cultural spirit, an improved flexible semantic 3D model approach is proposed to adapt to our research theme more accurately. The contributions of this study are as follows: (1) a two-level semantic decomposition of roofs in terms of structure and decoration is proposed according to the characteristics of ancient Chinese-style architecture; (2) the corresponding topological constraints and derived transformations are determined; (3) an extension model, ACRoofADE, based on CityGML, is developed using the Application Domain Extension (ADE) mechanism; and (4) several LODs (levels of detail) are precisely described for a roof structural model.
6. Experimental Results
The Beijing courtyard has a time-honored history in China and was probably first found at the courtyard site of West Zhou Dynasty (1046 BC–771 BC). During the Ming (1368–1644) and Qing (1636–1912) dynasties, distinctive four-sided courtyards were built in many regions throughout the country [47
]. Here, we present the experimental results for modeling the principal buildings of the Palace Museum, which is the largest and most complete ancient architectural group currently in China [48
] and includes a variety of roof styles.
The model data used for the CityGML software, named FZK Viewer, with the GML file are illustrated in the following figures. The left side of the viewer includes the hierarchical tree of ancient architectural semantic components, and the right view displays the 3D model related to the selected items on the left. Figure 9
a presents an overview of the models, which include thirteen surrounding buildings and one common base ground. The Tai-He Palace model without slope surfaces is illustrated in Figure 9
b by filtering the components on the left to reveal the more detailed structure of the multi-eave hip roof.
To more clearly demonstrate the model process, the bottom-up generation states are presented in Figure 10
, from which we can simultaneously note the components mentioned in Section 3
in a correctly-assembled ruler. Figure 11
a–e illustrates the separate models in LOD2 of five palaces, therein representing a multi-eave hip, a rectangular pyramidal pavilion, a multi-eave flush gable-hip, a single-eave flush and a gable-hip roof. The corresponding LOD3 models are illustrated in Figure 11
7. Conclusions and Perspectives
In this study, we utilized a semantic modeling approach for ancient Chinese-style architectural roof styles by extracting the exterior roof components into two functional categories: structural and decorative. Semantic modeling may be more understandable and controllable for the unified description of the components. Relative topological rules were used to control the combination of the above-mentioned components and to validate the correctness of the model. Several transformative forms were listed to enrich the diversity of the building group. An application model termed ACRoofADE that extended CityGML was developed to express and ensure the consistency of the geometrical, semantic and topological relationships when rebuilding the roof models. To improve the display efficiency and ensure visual fidelity, three detail levels (LOD1, LOD2 and LOD3) were used according to the former semantic decompositions.
The proposed method was tested on the principal buildings of the Palace Museum, which demonstrated the feasibility of the model. One advantage of this study is that it adopted a two-level semantic decomposition and topological reconstruction of complex 3D building models to allow the operator to focus on the structure design rather than cumbersome geometric drawings. In addition, the introduction of CityGML makes the previous thought feasible in practice, and the LOD concept further ensures accurate model representation.
After the primary exploration of 3D modeling for ancient Chinese-style architectural roofs, we envision several possible directions for future studies. The research scope could be broadened to the entire building structure, including the areas below the roof, to develop a complete research system. Moreover, interrelations, such as the degree of similarity between ancient Chinese-style architecture styles, could be analyzed to increase the efficiency of digitizing architectural heritages or guiding antique constructions.