Radial View-Based Culling Using Dynamic Clustering for Continuous Self-Collision Detection of Closed Deformable Models.

• Main Authors: Hung, Chun-Hung, 洪駿宏
• Other Authors: Wong, Sai-Keung
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/73099528055540391374

碩士 === 國立交通大學 === 多媒體工程研究所 === 102 === In this thesis, we use the concept of Radial view based culling (RVBC) to perform self-collision culling and improve the RVBC method. Experimental results of RVBC show the more the number of clusters is, the lower the cost of inter-cluster check is. However, the more the number of clusters is, the higher the cost of intra-cluster check is. RVBC determines the number of clusters at the preprocessing stage and then the number of clusters is fixed. When an object deforms, the cluster distribution may result in more negatively oriented triangles. Thus, we propose a dynamic clustering to dynamically adjust the number of clusters so that the cost of inter-cluster check and intra-cluster check is balanced to improve culling performance. We also find the result of cluster decomposition in RVBC can be improved. % There are more negatively oriented or uncertain triangles. There are negatively oriented or uncertain triangles should be assigned to the other atomic clusters. Moreover, cluster decomposition in RVBC spends a lot of time. Thus, we propose a new cluster decomposition method to improve clustering. Besides, RVBC uses skeleton motion to update positions of observer primitives at the runtime stage. % In order to reduce the limitation of using our method, We compute barycentric coordinates of observer primitives at the preprocessing stage and use it to update positions of observer primitives at the runtime stage.