基于SVD-SURF的宽基线鲁棒景象匹配算法

SVD-SURF-based Wide-baseline Robust Scene Matching Algorithm

提出一种基于SVD-SURF的宽基线鲁棒景象匹配算法。首先,在实时图与基准图奇异值分解的基础上构建SURF尺度空间,运用快速Hessian矩阵定位极值点;然后,计算出图像的64维SURF描述子;最后,通过Hessian矩阵迹进行特征点匹配,并利用RANSAC参数估计方法剔除出格点,从而实现位置参数的精确估计。实测航空图像序列位置估计实验表明了该景象匹配算法对图像的旋转、尺度变换及噪声不敏感,具有较强的实时性、精确性和鲁棒性。

A SVD-SURF-based wide-baseline robust scene matching algorithm is presented. SURF scale space is firstly set up on basis of singular value decomposition（SVD） of real-time image and reference image, and extreme point is positioned by using fast Hessian matrix; and then, 64-dimensional SURF feature descriptors of image were calculated; and finally feature points was matched by using trace of Hessian matrix, the outliers for accurate posi- tion estimation is eliminated by using RANSAC parameter estimation algorithm so as to implement accurate position parameters estimation. Experiments on actual measured aviation image sequence position estimation show that this scene matching algorithm is insensitive to rotation of image, scale transformation and noise, and it has intensive re- al-time performance, accuracy and robustness.