基于SIFT和三角网格的VHR建筑物检测技术

Building detection from VHR based on SIFT and Delaunay triangulation

传统尺度不变特征变换(scale-invariant feature transform,SIFT)算法的误配准问题导致基于该算法的建筑物检测率较低,因此提出一种改进的高分辨率(very high resolution,VHR)遥感卫星图像中建筑物的检测方法。首先通过改进传统的SIFT配准方法,使其能够更加准确地描述VHR遥感卫星图像中的建筑物信息,之后通过欧氏距离获得2幅图像的初始匹配点集,然后将配准后的一幅图像中所得到的配准点作为Delaunay三角剖分的初始点集,根据Delaunay三角剖分特性,剔除SIFT算法中误匹配的特征点,得到最终的结果。实验结果表明,研究所提出的算法可以有效地检测出一幅VHR遥感卫星图像中的建筑物信息,在时间效率、配准精度、建筑物的检测普遍性上,都能得到很好的预期效果。

The mis-registration problem of traditional scale-invariant feature transform （SIFT） algorithm results in a low detection rate of buildings based on this algorithm. Therefore, this paper proposed an improved detection method of buildings in very high resolution （VHR） satellite imagery. First, the traditional SIFT registration method was improved to characterize the building information in VHR remote sensing satellite images. Then, the set of initial matching points of two images could be obtained with Euclidean distance, followed by taking matched registration points in an image as the initial point set of the Delaunay triangulation. Finally, the result was got by removing mismatched feature points according to Delaunay triangulation properties. Experimental results show that the algorithm proposed in this paper can effectively detect the building information in a VHR remote sensing satellite image. Moreover, the proposed algorithm performs well in terms of time efficiency, registration accuracy and universality in detection of buildings.