Abstract:

In order to improve the accuracy and completeness of the three-dimensional reconstruction model, it is necessary to increase the amount of data of three-dimensional reconstruction, which will increase the computation and running time of the reconstruction. In order to solve this problem, a point cloud reconstruction parallel process is designed to reduce the time-consuming and improve the efficiency of three-dimensional reconstruction. A parallel algorithm for point cloud reconstruction in multi-core CPU, GPU architecture and CPU/GPU heterogeneous environment is proposed, and Kermit and hallFeng data sets are reconstructed on different experimental platforms. The experimental results show that, compared with the serial point cloud reconstruction algorithm, the parallel algorithm of point cloud reconstruction achieves a better speedup under the condition of guaranteeing the reconstruction accuracy, and the parallel algorithm has the scalability of the experimental platform and data scale.

Key words: point cloud reconstruction, parallel algorithm, heterogeneous computing, Graphics Processing Unit（GPU）, multi-core CPU

摘要：

在三维重建问题中，为了提高重建模型的精确度和完整性，需要增大三维重建的数据量，由此会增加重建的计算量和运行时间。针对该问题，对点云重建过程进行并行设计，降低耗时、提高三维重建的效率，提出在多核CPU、GPU架构和CPU/GPU异构环境下点云重建的并行算法，并在不同实验平台上对Kermit和hallFeng数据集进行了点云重建的并行实验。实验结果表明，相比于串行的点云重建算法，点云重建并行算法在保证重建精度的条件下，取得了较好的加速比，并且并行算法具有实验平台和数据规模的可扩展性。

关键词: 点云重建, 并行算法, 异构计算, 图形处理器（GPU）, 多核CPU