Study On Acceleration Algorithm Of COONS Surface Using GPU

As one of the most important patterns for the three dimensional reconstruction, Coons surface can be employed to create graphics in combination with the NURBS and Bézier surfaces. In the area of surface modeling, usually Coons surfaces are used to generate preliminary models, based on which the NURBS or Bézier surfaces are then constructed.With the rapid enhancement on power of the Graphics Processing Unit (GPU), the computational speed of computer graphics raised greatly, which boosted the development of relevant applications. Consequently, general purpose computation on GPU has been a focus in recent yeas.In order to meet the requirement of general purpose GPU computation, we transformed the equations of traditional Coons surfaces. The transformed equations are decomposable and can be used for the parallel construction of Coons surface.The main work of this thesis is as follows:(1) Based on the computational power of GPU, we establish and decompose the bilinear Coons surface according to the input-data requirements of GPU.(2) Based on the computational power of GPU, we establish and decompose the fractal Coons surface according to the input-data requirements of GPU.(3) Based on the general GPU computing framework, algorithms are proposed to achieve the calculation of Coons surfaces on GPU. Our algorithms are implemented using the Cg language under a combined environment of Microsoft VC++ 6.0 and the Open GL library.(4) Through experiment we draw the conclusion that with the interpolation points increasing, execution time of fragment program still maintains stable to certain degree, without significant rise; the efficiency of GPU parallel computing is higher than CPU; In the process of operation based on GPU, there is much bulk data exchange between GPU and CPU, which will affect the efficiency of GPU; The time which a fragment program of GPU run in single-circulation of single-graph-pipeline is less than that in multi-circulation of multi-graph-pipeline.