To promote the model reuse efficiency of 3D Computer Aided Design (CAD), aiming at the flaw of semantic expression in previous 3D model retrieval systems, a similarity measure method based on model semantic representation of Web Ontology Language (OWL) was presented. Firstly, 3D CAD master model was transformed into structuralized representation model with class-property feature as its basic semantic object. And then, the feature semantic information for matching two models was extracted from OWL representation model to be a quantitative similarity unit. And a method of total weight combining subgraph isomorphism and Tversky algorithm was proposed for similarity measure. Finally, the experiment verified the feasibility and effectiveness of the proposed method. The comprehensive quantitative assessments of the experiment show that the proposed method makes the evaluation benchmark switch from the object itself to the set of semantic description of two object properties, and can objectively reflect the similar degree of the two contrast models.

The directional relations between spatial objects cannot be quantitatively analyzed with directional feature in existing models, which caused regional query to be involved in only single-direction open region. To resolve this problem, a space vector model algorithm combined with vector operation and MB-tree was proposed to deal with directional relation query in continuous and open regions. The proposed algorithm mainly included two steps. One was filtration, and the other was purification. In the process of filtration, the relations between query region and the vertexes of space object's Minimum Bounding Rectangles (MBRs) were analyzed quantitatively and the corresponding judging method was given. Effective pruning rule for nodes of MB-tree was given to reduce I/O cost when directional query was executed by the order of MB-tree. In the process of purification, in order to find actual target objects, MBRs which were saved in the filtration step were traversed. The experiments show that the proposed algorithm can not only solve the query of single open region, but also solve the query of continuous and multidirectional region. In addition, the algorithm can be used in two- and three-dimensional spaces.

This dissertation proposed a new motion planning algorithm for human arm based on depth buffer(z-buffer).This method expanded the A* algorithm to three-dimensional space and detected collision taking advantage of a special computer graphics hardware-depth buffer.Thus it was addressed the problem of confined space motion planning for human arm.The simulation results show that this algorithm can realize the motion planning for a human arm with 3 links and 7 DOFs rapidly and effectively.

In many effective algorithms for cluster, DBSCAN algorithm is outstanding for its good performance in spatial data。Relying on a density-based notion of clusters, DBSCAN can discover clusters of arbitrary shape. But it cant support non-spatial attributes, in some application of cluster, the non-spatial attributes play important role. Based on the DBSCAN, referencing some notion of DBRS and considering data type of non-spatial attribute, the paper proposed a method of extension of DBSCAN and gave main algorithm。The algorithm can operate spatial and non-spatial attribute.

A video coder without motion compensation based on 3D-SPECK was proposed. In order to carry out the three dimensional wavelet transform, the video sequences were partitioned into groups of frames (GOP), and the wavelet coefficients were organized as many wavelet cubes (WCs). The WCs were coded by the 3D-SPECK efficiently. The wavelet transform was carried out in the temporal axis instead of the traditional motion estimation/compensation, thus much computation was saved compared with the traditional hybrid coding model. Experiment results show that the proposed coding model achieves both measurable and subjective performances comparable with 3D-SPIHT.

CDN is a fictitious network, covered in Internet and made up of nodal servers in different areas. Through caching, duplicating, load balancing and user request redirecting, CDN pushes information resources to network edges to provide end users with the best and most available servers to access various contents. A brief view of CDN technology was presented, including basic working principles of CDN platform, design principles and mechanism compare of the content routing, and the nodal content engine was compared. The establishment of information sources in broadband networks was also mentioned.

Support vector machine is a new classification algorithm based on statistical learning theory. After the discussion of the current multiclass SVMs, a novel multiclass SVM classifier based on geometric distance was proposed. The Posterior probability output of binary SVM was generalized to multiclass SVM. Without iteration computing, this method improved prediction accuracy with fast computing. The numeric experiments prove that above two methods have good generalization, which can increase prediction accuracy to unknown examples.

OLAP is one of important techniques in DSS.The computation of CUBE is the basis of ad hoc query in OLAP.In order to advance the corresponding to querying,we must pre aggregate data in tables or multidimensional databases.Typically,the computation of CUBE consists of materializing strategy,the choice of materialized paths and its optimization and CUBE algorithms.Based on them,this paper analyzes and expatiates concepts,strategies,steps,methods and performances about the computation of CUBE in details.