基于InfiniBand集群并行全景图拼接系统设计

全景图拼接是通过对序列图像进行拼接,从而实现对场景全方位环视的技术,是计算机视觉和图像处理的研究热点.针对大规模高分辨率的序列图像拼接,单计算机的拼接效率已经难以满足实际应用需求的问题,提出了一种基于InfiniBand高速网络通信的多机并行架构实现全景图并行拼接的解决方案.首先,对全景图拼接的关键步骤进行并行性分析,将全景图拼接尽可能划分成细粒度和高并行度的子任务;然后,通过并行架构中的控制节点和计算节点进行任务的分配和执行;最后,针对随着图像数据量增大,采用传统的TCP/IP网络传输,计算节点间的数据传输速率低下的问题,设计了基于InfiniBand与TCP/IP结合的高速网络传输模块,有效地提高节点间的传输速率.实验结果表明,在此并行架构上实现的分布式计算,能够有效地提升全景图拼接效率,有明显的实际应用价值.

Design of parallel panoramic stitching system based on InfiniBand cluster

Panoramic stitching is a technology to achieve all-round look of the scene by stitching image sequence. It is a research hotspot of Computer Vision and Image Processing. For large-scale and high-resolution image sequence stitching, stitching efficiency of single computer is hard to meet the needs of the practical application. A new solution for panorama parallel stitching based on a multi-machine parallel architecture with high-speed network InfiniBand was proposed. At first, do the parallelism analysis for the key steps of panoramic stitching and divide into fine-grained and high degree of parallelism subtasks; Secondly assign and execute the tasks through the control node and the compute nodes in the parallel architecture; Finally, in order to solve the problem that the rate of the data transmission between the computer nodes using the traditional TCP/IP network is low with the amount of image data increasing, we designed a high-speed network transmission module based on InfiniBand and TCP/IP which can effectively improve the transmission rate. The experimental results proved that the distributed computing on this parallel architecture can effectively improve the panorama stitching efficiency with a significant practical value.