面向三维重建的无人机影像并行处理技术  

UAV image parallel processing 3D reconstruction

作　　者：庞巧遇 邓宝松 桂健钧 鹿迎 PANG Qiao-yu;DENG Bao-song;GUI Jian-jun;LU Ying(National Innovation Institute of Defense Technology,Academy of Military Science,Beijing 100071,China)

机构地区：[1]军事科学院国防科技创新研究院,北京100071

出　　处：《计算机工程与设计》2023年第2期526-534,共9页Computer Engineering and Design

基　　金：国家自然科学基金项目(61902423)。

摘　　要：为提升三维重建任务的执行速度,解决行业现实应用对时效性的需求,提出一种无人机影像并行处理与特征提取算法,基于CPU与GPU两种计算架构在三维重建的两个阶段并行加速处理。一是基于CPU的并行处理策略,针对多核处理器采用OpenMP多线程机制,对无人机影像进行并行加载,为后续处理提供高效数据源;二是基于GPU的并行处理策略,通过改进SIFTGPU算法在GPU上以并行方式对图像进行特征提取,为快速重建提供特征输入。真实数据的实验结果表明,与现有算法相比,在图像处理速度上提升了2倍,特征点数量提升了4倍的同时,提取速度提升了11倍。To improve the execution speed of 3D reconstruction task and meet the requirement of timeliness in practical application of industry, a parallel processing and feature extraction algorithm of UAV image was proposed. CPU and GPU were used to accelerate parallel processing in the two stages of 3D reconstruction. The first was a CPU-based parallel loading processing stra-tegy, in which OpenMP multi-thread processing mechanism was used for multi-core processors to process UAV images in parallel. The second was a GPU-based parallel processing strategy, in which the SIFTGPU algorithm was used to extract the features of the image in parallel on GPU. Experimental results of real data show that compared with the existing algorithms, the image processing speed is increased by 2 times. While the number of feature points is increased by 4 times, the extraction speed is increased by 11 times.

关 键 词：室外大场景 无人机影像 三维重建 并行图像处理 并行特征提取 并行编程技术 基于图形处理器的尺度不变特征变换 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]