GPU并行计算加速的实时可视外壳三维重建及其虚实交互

发布时间：2018-03-20 03:07

本文选题：可视外壳　切入点：三维重建　出处：《计算机辅助设计与图形学学报》2017年01期 　论文类型：期刊论文

【摘要】：针对现有的基于图像的三维重建方法难以实现真实物体的快速三维重建,无法满足虚实交互等应用需求的问题,提出一种基于GPU并行计算的实时三维重建及其虚实交互方法.首先把物体所在空间剖分成具有数据独立性的体素集合,结合可视外壳重建算法和精确行进立方体算法并行遍历每个体素得到体素状态序列;然后并行压缩体素状态序列得到非空体素集合,对非空体素进行并行三角形网格化,并利用图形硬件的多重纹理映射和可编程功能进行基于像素的纹理映射;最后假定虚拟物体的粒子为运动受限的拉格朗日流体粒子,重建物体网格顶点为流体边界,通过流体动力学方程的并行光滑粒子动力学方法求解来计算虚实交互.实验结果表明,该方法在GPU上进行完全并行求解,在32×32×32的空间剖分精度下,实现了实时三维重建和20帧/s左右的虚实交互计算,适用于计算机图形学和虚拟现实等领域中的虚实交互应用.
[Abstract]:To solve the problem that the existing 3D reconstruction methods based on images are difficult to realize the fast 3D reconstruction of real objects and can not meet the needs of virtual real interaction and other applications. A real-time 3D reconstruction method based on GPU parallel computing and its virtual real interaction method are proposed. Firstly, the object space is divided into a set of voxels with data independence. Combining the visible shell reconstruction algorithm and the accurate marching cube algorithm, each voxel state sequence is traversed in parallel, then the voxel state sequence is compressed in parallel to obtain the non-empty voxel set, and the non-empty voxel is triangulated. Using the multi-texture mapping and programmable function of graphics hardware, the texture mapping based on pixels is carried out. Finally, the particles of virtual objects are assumed to be Lagrangian fluid particles with limited motion, and the vertices of the mesh are reconstructed to be fluid boundaries. The virtual real interaction is calculated by solving the parallel smooth particle dynamics method of the hydrodynamic equation. The experimental results show that the method is solved completely in parallel on GPU, and with 32 脳 32 脳 32 脳 32 space partition accuracy. The real time 3D reconstruction and the virtual / real interaction computation of about 20 frames / s are realized, which is suitable for the virtual reality interaction applications in the fields of computer graphics and virtual reality.
【作者单位】：华北电力大学控制与计算机工程学院;NVIDIA半导体科技(上海)有限公司;
【基金】：国家自然科学基金项目(61502168,61472020) 国家“八六三”高技术研究发展计划(2015AA016403) 河北省自然科学基金(F2016502069)
【分类号】：TP391.41;TP391.9

【相似文献】