基于SIFT特征描述子的立体匹配算法

A Stereo Matching Algorithm based on SIFT Feature Descriptor

目前,立体匹配是计算机视觉领域最活跃的研究课题之一。为了克服传统的局部特征匹配算法对噪声和图像灰度的非线性变换敏感的缺点,本文提出了一种新的基于SIFT(Scale Invariant Feature Transform)特征描述子的立体匹配算法。该算法利用图像梯度信息,构造基于三维梯度方向直方图的SIFT特征描述子作为区域特征描述符,通过立体视觉理论中的极线约束将匹配特征的搜索空间从二维降到一维,最后以基于特征描述子欧氏距离的最近邻匹配得到匹配结果。实验结果表明,该方法匹配精度高,对图像灰度的非线性变换比较鲁棒,可以应用于对匹配算法鲁棒性要求比较高的立体视觉系统中。

Stereo matching is one of the most active research subjects in the field of computer vision at present. A new stereo matching algorithm based on SIFT feature descriptor is proposed in this paper, in order to restrain the high sensitivity to image noise and non-linear intensity transformation as for conventional local matching algorithms. This algorithm first constructs SIFT local feature descriptor based on 3D histogram of gradient location and orientation by exploiting image gradients information, and then reduces the search space of matching features from 2D to 1D in the light of epipolar constraints in stereo vision theory, the matching result is finally obtained according to the nearest neighbor matching based on Euclidean distance of SIFT descriptors. The experimental results prove that this algorithm wins high matching accuracy and robustness against non-linear image intensity transformation. So it can be used in stereo vision system which is in demand of high robustness for stereo matching algorithm.