摘要
为了使机器人在陌生环境中能够通过传感器对环境进行探测,并在对自身定位的同时得到周围环境的三维重建地图,视觉SLAM(Simultaneous Localization and Mapping),即时定位与地图构建应运而生。文章对视觉SLAM方法与特征匹配SIFT算法进行了论述与提取效率分析。SIFT算法包含大量单指令多数据流模式的密集型计算,使用CPU+GPU混合异构加速平台在理论上能有效提高其执行性能。通过对国内外学者在GPU并行加速领域研究成果的分析,针对GPU加速SIFT算法展开了分析与展望。
In order to enable the robot to detect the environment through sensors in an unfamiliar environment and obtain a threedimensional reconstructed map of the surrounding environment,while positioning itself,visual SLAM(Simultaneous Localization and Mapping)came into being.This paper discusses the visual SLAM method and the feature matching SIFT algorithm,and analyzes the extraction efficiency.SIFT algorithm contains a large number of intensive calculations in Single Instruction Multiple Data mode.The use of a CPU+GPU hybrid heterogeneous acceleration platform can theoretically effectively improve its execution performance.Through the analysis of domestic and foreign scholars'research results in the field of GPU parallel acceleration,the GPU accelerated SIFT algorithm is analyzed and prospected.
作者
郑泰皓
方子彬
Zheng Taihao;Fang Zibin(School ofComputer and Information Engineering,Henan University,Kaifeng,Henan 475000,China)
出处
《计算机时代》
2020年第7期16-21,共6页
Computer Era
