基于视差的高分辨率遥感影像建筑物变化检测

针对城区高分辨率遥感影像建筑物变化检测中视差引起误检测问题，提出一种基于视差的建筑物变化检测方法。首先，利用改进的双峰分裂法实现对建筑物阴影的检测；然后，借助阴影估计屋顶的位置，并自动选取种子区域，结合区域生长法进行建筑屋顶提取；最后，以视差、视差方位角以及面积等作为判定条件对两时相影像建筑物的变化情况进行分析。通过对WorldView-2影像和IKONOS影像进行变化检测，平均正确率达到了89.6%。实验结果表明:在屋顶光谱均匀且有明显建筑阴影的稀疏建筑区，该方法能够有效解决建筑屋顶种子区域自动选取及高层建筑视差引起误检测等难点问题。

Building Change Detection Method Considering the Parallax for High Resolution Remote Sensing Image

Considering the problem of false detection caused by the parallax in building change detection of urban high resolution remote sensing images, a new method of building change detection is proposed. Firstly, the classical method of bimodal splitting is used to extract the shadows of buildings with small improvements, and then the seed area is selected automatically by estimating the possible roof position with the help of shadows, and the roof is extracted by the region growing method. Finally, the change of building in two-phase image is analyzed by using parallax, parallax azimuth, area, etc.. Through the change detection of WorldView-2 images and Ikonos images, the average accuracy rate reached 89.6%. Experimental results show that in sparse building areas with uniform roof spectrum and obvious building shadows, the proposed method can effectively solve the difficult problems of automatic selection of roof seed areas and false detection caused by the parallax in building change detection.