Studies On Solid Reconstruction From Engineering Drawings

Solid reconstruction from engineering drawings is one of the efficient technologiesto product solid models, which has become one of the important research topics incomputer graphics and artificial intelligence. This dissertation focuses on several keytechnologies of the solid reconstruction to solve the existing problems in the coverageof input views, the domain of reconstructed objects, and the process efficiency. Themain contributions are summarized as follows.1. A method is presented for automatically labelling a variety of views from a multi-view drawing based on the knowledge of section expressions. In this approach,view relations of the drawing are formalized as a view relational graph with re-spect to the 2D drawing space. Furthermore, a reasoning technique based onevidence theory is employed for constructing a view tree that consists of themost valuable relations for determining view adjacency in the 3D object space.By the method, geometric entities and symbolic entities in the drawings are han-dled within a uniform framework for identifying spatial position of views andsections and supplying them to the downstream reconstruction processes. Thismethod can handle engineering drawings with full sections, offset sections, andaligned sections, allowing that neither the types of sections nor the layout of var-ious views is known a priori.2. An efficient method for generating 3D curvilinear wire-frames from three ortho-graphic views is presented, which is the pivotal stage of solid reconstruction bywire-frame-oriented approaches. A projection matching algorithm based on deci-sion trees is proposed to efficiently reveal all matching projections. Furthermore,a four-point algorithm is proposed to uniformly generate various types of conicsin arbitrary positions by employing the rational Be′zier form of conics. Takingfull advantage of the projective properties of spatial edges, this method can im-prove the efficiency and reliability of curvilinear wire-frame reconstruction byreducing the search space and simplifying the reconstructing process. 3. A new method is proposed to reconstruct curved objects from three orthographicviews. The Link-Relation Graph (LRG) is introduced as a multi-graph repre-sentation of orthographic views, and quadric surface features (QSFs) are definedby special basic patterns of LRG as well as aggregation rules. By hint-basedpattern matching in the LRGs of three orthographic views in an order of pri-ority, the corresponding QSFs are recognized, and the geometry and topologyof quadric surfaces are recovered simultaneously. The method covers a broaderobject domain to accommodate interacting quadric surfaces with higher orderintersections. Furthermore, by employing a hybrid wire-frame in place of an in-termediate wire-frame, the processing time can be drastically reduced due to theprocedure simplification without the combinatorial search and backtracking usedby the existing algorithms.