利用改进的SUSAN算法提取航空影像中孤立的特征点

Extracting isolated feature points in aerial images by using improved SUSAN algorithm

本文针对地形起伏较大、无明显建筑物的航空影像,分析了SUSAN算法角点检测理论,提出一种提取孤立特征点的方法。该方法先对图像进行梯度幅值运算,然后对梯度幅值进行Otsu法阈值分割,设计模板并对孤立特征点进行套合,最后利用SUSAN算法计算原始影像的角点初始响应,经过非极大值抑制提取孤立特征点。经实验证明,与传统的Harris角点、Forstner角点相比,该特征点受地形起伏、太阳高度角、视角变换等外界条件干扰较小,为下一步影像匹配做了较好的准备。

The paper analyzed the theory of SUSAN algorithms for the aerial images with obvious terrain changes and almost without architectural buildings, developed a method of extracting isolated feature points. Firstly, the method computed the gradient magnitude, made a threshold segmentation using Otsu, then designed a template to cover the segmentation result and gave the initiated values, finally computed the corner measurement on the original images using SUSAN algorithm, and got the isolated feature points after the non-maximum suppression. The experiment demonstrated that the external conditions including terrain changes, Solar elevation angle and shift of viewpoint had a little effect on the points, which made a full preparation for the next matching step.