CAJ | 학술논문

基于人的理解过程和方法，文中提出了工程视图理解的新算法框架.它允许用户参与到理解过程中，在出现歧义时，根据人的视觉认知规律进行“猜测”和“推理”，可得到与人的理解结果更为一致的解，并允许加入剖视图的处理.该文着重介绍该方法的算法思想，以及基元的划分与生成、剖视图处理、视觉推理等关键算法步骤.
기우인적리해과정화방법，문중제출료공정시도리해적신산법광가.타윤허용호삼여도리해과정중，재출현기의시，근거인적시각인지규률진행“시측”화“추리”，가득도여인적리해결과경위일치적해，병윤허가입부시도적처리.해문착중개소해방법적산법사상，이급기원적화분여생성、부시도처리、시각추리등관건산법보취.
Interpretation of engineering drawings has long been considered as a challenging task. The main object of understanding engineering drawings is to allow the computer to acquire the type of knowledge humans use to understand projections, so that the computer can, in a sense, understand these projections as humans do. In this paper a new algorithm is proposed to interpret the engineering draws based on human understanding approaches of multiple projections. The multi-view projections are globally interpreted by volume-oriented approach, which is similar to CSG modelling. In order to partition the goal solid into primitives and represent it as a CSG tree, we restrict the primitives as only one special type, namely, the cell primitive whose 3D solid could be generated by general translational sweeping operation. So the primitive recognition is converted into the partition of primitives. This abstract primitive covers the common primitives such as cube, cylinder, ball etc. The wire-frame approach is then applied locally to create the concrete 3D model of the abstract primitive description generated by primitive partition. It is efficient to construct the 3D model of primitive since the model is usually simple, and few pathological edges or vertices are generated. When multiple interpretations exist, the visual reasoning engine is activated to pick out the most acceptable interpretation, and make the interpretation be more consistent with that from human engineers. The sectional views could be involved to eliminate the unsuitable interpretations of projection views. Moreover, the users could be allowed to be involved in interpretation process in different granularity levels, such as primitive level or baseface level.Comparing with the existing wireframe-oriented algorithms, its solid model is reconstructed by tanslational sweeping. The vertices, edges and faces of shape are generated simultaneously, and pathological elements are much reduced. Comparing with the volume-oriented methods, there is only one type of primitive involved in it. Its cell-primitives are recognized automatically in most cases, and much less interactions in primitive recognition are required than that in volume-oriented approach.