基于改进SIFT算法的单目SLAM图像匹配方法  

A Monocular SLAM Image Matching Method Based on Improved SIFT Algorithm

作　　者：胡为[1] 刘兴雨 HU Wei;LIU Xing-yu(Department of Aerospace Engineering,Shenyang Aerospace University,Shenyang 110136,China;School of Automation,Shenyang Aerospace University,Shenyang 110136,China)

机构地区：[1]沈阳航空航天大学航空航天工程学部,沈阳110136 [2]沈阳航空航天大学自动化学院,沈阳110136

出　　处：《电光与控制》2019年第5期7-13,共7页Electronics Optics & Control

基　　金：辽宁省教育厅基金(L201749)

摘　　要：基于单目视觉的机器人同步定位与地图创建(SLAM)方法是实现机器人自主行走的重要研究方向之一,而图像特征匹配技术是该方法中的关键技术。基于尺度不变特征变换(SIFT)算法的SLAM匹配方法具有提取特征点数量丰富、稳定等优点,然而在速度以及正确率方面仍存在一些不足。因此针对SIFT算法描述子维数高、匹配时间长等问题,提出了一种改进的SIFT算法,将原128维的特征描述子降至内部矩形外部圆形的24维特征描述子,匹配过程中应用了三线性插值、RANSAC算法等对匹配结果去除误匹配。实验结果最终表明,改进后的SIFT算法不仅对角度变化、光照变化等情况均具有良好的鲁棒性,匹配速度和正确率也有显著提升,可满足实时SLAM同步地图构建的需要。The Simultaneous Localization and Mapping(SLAM) method based on monocular vision is one of the most significant research directions for the robot autonomous walking,and the image feature matching technology is the key technology of this method.The SLAM matching method based on Scale Invariant Feature Transform(SIFT) algorithm has the advantage of abundant and stable feature points,but it also has some disadvantages in speed and accuracy.Therefore,considering that the SIFT descriptor has high dimensions and long matching time,an improved SIFT algorithm is proposed.The original 128-dimensional feature descriptor is reduced to a 24-dimensional feature descriptor with a rectangular inner side and round outer side.In the matching process,trilinear interpolation and RANSAC algorithm,etc.are applied to eliminate the mismatching of matching results.The experimental results finally show that the improved SIFT algorithm not only has good robustness to the changes of angle and illumination conditions and improves the matching speed and accuracy,but also meets the needs of SLAM synchronous map construction.

关 键 词：图像匹配 单目视觉 尺度不变特征变换 特征描述子 RANSAC算法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]