一种基于时空切片的SLAM关键帧提取方法  

Key Frame Extraction for SLAM Based on Spatio-temporal Slices

作　　者：张新良[1] 李阳 赵运基[1] Zhang Xinliang;Li Yang;Zhao Yunji(School of Electrical Engineering and Automation, Henan Polytechnic University, Jiaozuo 454000, Chin)

机构地区：[1]河南理工大学电气工程与自动化学院,河南焦作454000

出　　处：《系统仿真学报》2018年第5期1724-1729,共6页Journal of System Simulation

基　　金：国家自然科学基金(U1404612);教育部博士点基金-新教师类(20134116120003)

摘　　要：针对移动机器人三维视觉同步定位与地图构建SLAM(Simultaneous Localization And Mapping)过程存在的数据冗余和计算复杂度高的问题,提出了一种基于时空切片的关键帧提取技术。基于对时空切片纹理的特征分析,将图像采集设备的运动状态反映为时空切片纹理的倾斜度变化,进而通过度量相邻时空切片间像素的相近程度,利用最邻像素匹配法,确定能够准确描述图像采集设备运动状态的关键帧。实现了在定位精度要求下的关键帧选取,降低SLAM过程中的计算量。实验结果表明,该方法能有效减少冗余帧,在定位性能和处理速度上具有优势。To deal with the data redundancy and computational complexity of the mobile robot 3D visual SLAM （simultaneous localization and mapping）, a key frame extraction technique based on spatio-temporal slices is proposed. Based on the analysis of the texture characteristics of the spatio-temporal slices, it yields that the motion state of the image acquisition device can be represented by the slope change of the texture. The similarity degree between two adjacent slices is measured to find out the key frame by use of the nearest neighbor pixel-matching algorithm, which can accurately describe the motion state of the device with the reduction of the redundant frames. The experimental results show that the proposed method can reduce the data redundancy in frames effectively, and presents an advantage in the localization capability and processing speed in SLAM due to the exclusion of the unnecessary computation.

关 键 词：同步定位与地图构建 关键帧 时空切片 最邻像素匹配法 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]