Automatic Simplification Approach For The LoDs Of Complex3D Building Models

A complex3D building model is a type of building model that contains a detailed description of both its appearance and its internal structure with authentic architectural components. Because of their high complexity and huge data volumes, Levels of Detail (LoDs) should be created for the real-time visual application of such models, on the one hand, to alleviate the pressure on the visualization system and, on the other hand, to reduce the information density to raise the efficiency of spatial cognition.Most of the existing mesh simplification algorithms focus on the simplification of continuous surfaces and are not suitable for maintaining the characteristics of buildings, and3D generalization methods extended from2D cartographic generalization are mainly proposed for regular prismatic building models rather than models with internal structures and smooth surfaces. As a result, LoDs of complex3D building models are often created interactively, which is time consuming and error-prone. It has also been one of the bottlenecks that prohibit the wide application of such models.The main goal of this thesis is to propose automatic simplification methods that could effectively reduce complexities of geometry, structure and cognition for complex3D building models while ensuring the consistency of perception and semantics between LoDs. All these are based on the quantitative study of the perceptibility of appearance, the topological relationship and semantics of components and internal structure.First, the existing methods of mesh simplification as well as the generalization of3D building are detailed reviewed in a chronological order following the evolution of methods. Future trends and the capability of each method for simplifying complex3D building models are analyzed which provides guidelines for related researches.Second, the traditional method neglects the perceptual consistency between successive LoDs which may results in the perceptual redundancy or popping effect during visualization. To solve this problem, the perceptual metric is defined as the quantitative indicator of the visual perceptibility of model details at a given viewing distance. A perceptual metric function is then defined, based on the perceptual metric values extracted at equal interval viewing distances. In the end, the minimum detail redundancy model is proposed for the optimal design of discrete LoDs. Based on the quantitative instructions provided by the previous steps, a ray-casting like method is proposed to precisely map the perception information from the image onto the geometric primitives that contribute the information. Finally, simplification that reduces the imperceptible details of the model is conducted to produce LoDs according to the design.Existing3D generalization methods cannot consistently express geometry, topology and semantics of complex3D building model in multiple LoDs. To overcome this problem, a mathematical morphology-based generalization algorithm is proposed.(1) semantic topologies between components are extracted and reflect structural connectivity in the building at a certain scale;(2) connected components are merged and trivial geometric features of the components are eliminated simultaneously, with semantics associated with components then updated according to the generalization; and (3) mesh simplification is carried out to further reduce the redundancy of triangular facets. The proposed method generalizes the components of complex3D building model at multiple scales in a unified manner, and their general shapes and structures are retained.Finally, a generalization method using typification for masonry structure of complex3D building models is proposed. Firstly, the global and local features of such components that present masonry structure are analyzed. Then, a parametric-based approach of feature extraction is employed, which is composed of the recognition of tiles, the extraction of rows and columns of tiled roof and the approximation of roof surface using spline surface. Based on the parametric representation, the generalization using typification is conducted. The original tiles are replaced by less and larger newly produced tiles and the roof surface is simplified towards a plane. As a result, not only the number of components is drastically reduced but also the general structure of the roof is preserved.Experiments on a typical complex3D building model-the timber-framed classical Chinese building model are conducted. The results prove the effectiveness of the proposed method.