基于CUDA架构的有向距离函数三维重建  

Design of 3D-reconstruction Using SDF Based on CUDA

作　　者：杨科 张之江[1] 

机构地区：[1]上海大学特种光纤与光接入网重点实验室

出　　处：《工业控制计算机》2019年第7期90-91,94,共3页Industrial Control Computer

摘　　要：为了解决三维重建中运算复杂、计算量大导致的重建实时性较低的问题,通过对有向距离函数三维重建算法进行并行性研究,设计出一种基于CUDA框架的并行处理算法。采用“共享内存”、“两次归约”求和等策略在GPU上进行并行加速。实验结果表明,并行的算法相比于串行算法具有明显的加速效果,位姿估计部分最高加速比为17.88倍,可以稳定、实时地进行物体或场景的三维重建。This paper designs a parallel processing algorithm based on CUDA framework by studying the parallelism of the three-dimensional reconstruction algorithm of sdf.It uses the shared memory,twice reductions summation strategy to parallel acceleration on the GPU.The experimental results show that the proposed algorithm has obvious acceleration effect compared with the serial algorithm.The highest acceleration ratio of the pose estimation part is 17.88 times,which can stably and real-time three-dimensional reconstruction of objects or scenes.

关 键 词：CUDA 有向距离函数(SDF) 共享内存 两次归约 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]