作者:刘明珠，陈瑞，陈俊羽，孙晓明

Authors:LIU Ming-zhu,CHEN Rui,CHEN Jun-yu,SUN Xiao-ming
摘要:摘要:针对传统ORB算法尺度不变性较差的问题，提出了基于图像金子塔的ORB算法。首先研究了高斯图像金字塔和B-spline图像金字塔，通过实验对两种图像金字塔进行了探讨。实验表明，3阶B-spline图像金字塔的图像所包含的图像信息要高于相同层上高斯图像金子塔的图像所包含的信息。其次，在图像尺度发生变化的前提下，通过实验对传统ORB算法、基于高斯图像金字塔的ORB算法和基于B-spline图像金字塔的ORB算法进行了探讨。实验表明，基于3阶B-spline图像金字塔的ORB算法相比较传统ORB算法所提取的特征点配准率提高了40%。同时，相较基于高斯图像金字塔的ORB算法，本文提出的改进算法在匹配准确率和运行速度上也有所提升。

Abstract:Abstract:Aiming at the problem of poor scale invariance of traditional ORB algorithms, this paper proposes an ORB algorithm based on image golden tower. Firstly, Gauss image pyramid and B-Spline image pyramid are studied, and the two image pyramids are discussed through experiments. Experimental results show that the image information of the third order B-Spline image pyramid is higher than that of the Golden Tower of Gauss image on the same layer. Secondly, the traditional ORB algorithm, the ORB algorithm based on Gaussian image pyramid and the ORB algorithm based on B-Spline image pyramid are discussed through experiments under the premise of changing image scale. Experimental results show that the registration rate of feature points extracted by ORB algorithm based on third-order B-Spline image pyramid is 40% higher than that of traditional ORB algorithm. At the same time, compared with the ORB algorithm based on gaussian image pyramid, the improved algorithm proposed in this paper has improved the matching accuracy and running speed.

PDF全文下载地址:

可免费Download/下载PDF全文