一种基于GPU的标量场驱动物理变形算法

An Algorithm of Physically-based Scalar-fields Guided Deformation on GPU

基于标量场的变形技术是计算机图形学中的研究热点之一,其时效性问题一直未得到很好的解决.从自适应采样距离场的表示方法和基于物理的建模技术的优点出发,提出了一种基于GPU的标量场驱动的物理变形算法.在GPU上构造基于八叉树的自适应采样距离场(adaptively sampled distancefields,ADFs)对模型进行表示,将质点弹簧物理模型与ADFs相结合,依据物理动力学原理直接对ADFs进行控制产生模型的变形,在变形过程中对ADFs进行动态的自适应调整.为了避免由非规则结构引起物理上的非同质性,根据ADFs局部的空间分辨率大小来调整非均匀弹簧的刚度大小.实验结果表明,该算法具有较高的时间和空间效率,比CPU上的算法在时间上快一个数量级,可有效用于基于物理的交互式雕刻等动态应用中.

Scalar fields guided deformation is one of the hot research issues in computer graphics.However,the problem of time efficiency of SFD is yet not to be solved.In this paper,a GPU-based shape deformation algorithm is proposed,integrating the advantage of both the representations of ADFs（adaptively sampled distance fields） and the physically-based modeling techniques.The octree-based ADFs are constructed and represented on the GPU.Then the deformation of the ADFs is governed by the principle of physical dynamics and achieved by manipulating the scalar fields directly.The algorithm constructs the ADFs and represents the octree structure on the GPU,which processes all the octree nodes at the same depth in parallel and provides fast access to the ADFs nodes with the look up tables.A dynamic adaptive resampling strategy for ADFs is employed during the deformation.Meanwhile,the physical properties are integrated into irregular fields for deformation and the stiffness of the non-uniform springs in the system is adaptive to spatial resolution of ADFs to avoid non-homogeneity in physics caused by the irregular structure.The results show that the time and space efficiency of our physically-based deformation algorithm based on the representation of ADFs are relatively high,which is over one order of magnitude faster than CPU algorithm.It also implies the algorithm has potential applications in physically-based interactive sculpting.