Study Of The 3D CAD Model Clustering And Retrieving Technology

With the rapid improvements of acquiring and rendering 3D CAD models, thedemands for retrieving models from large repositories of 3D CAD models have gainedincreasing prominence in the computer graphics and computer visions. To meet thesedemands, various retrieval methods have been proposed that enable the e?cient query-ing of a desired 3D model from the given repositories. Therefore the research of 3DCAD model retrieval methods has been the focus of both academy and industry.To solve this problem, this paper pays special attention to 3D model clusteringand retrieving methods. The main researches and contributions are as follows:Firstly, quasi random sequence is employed to choose the sampled points on themodel surface. In the sampling process, more uniformly distributed random sequencesin zone [0,1] are generated based on Halton sequence, and transformed into the propersequences in the sampling space. The retrieval results give us a clear view that theretrieve e?ciency is improved by using the quasi random sequence when sampling.Secondly, the Radius Angle Histogram (RAH) method is introduced to retrievemechanical parts in the engineering field. RAH method employs the angle betweenthe radius and the normal of the point on the model surface as the shape descriptor.This shape descriptor is translation and rotation invariant, and is robust to noise. Theexperiment shows that the RAH has better retrieval performance for some mechanicalmodel than Shape Distribution method.Thirdly, to improve the retrieval performance we combine RAH with the K-meansclustering method (KMM) and fuzzy c-means clustering method (FCM) in the me-chanical domain. The experiments show that by using these methods, better retrievalperformance can be obtained for Simple Pipes and Intersecting-Pipes datasets thanthat of Shape Distribution, while no single method performs best for all shape cate-gories.Lastly, a 3D model retrieval prototype system is designed based on UG. In theprototype, new model can be designed in UG, and then played as a query model toretrieve the similar model in the database. The returned models are ranked in thespecial order according to the similarity between them . The retrieval prototype issuitable for industrial use, especially for pipe-like components design.The steps discussed above can be used to integrate the design process, retrievalprocess, and the post-processes, etc. and to accelerate the manufacturing process andimprove the response time to the market.This paper is sponsored by the National Natural Science Foundation of China(Grant No. 50465001) to Prof. Meifa Huang and Guangxi Postgraduate Student Innovation Program (Grant No. 2007105950802M07) to Mr. Hui Jing.