基于城市遥感卫星影像对的立体匹配

Stereo matching based on urban satellite remote sensing image pair

为了解决城市卫星遥感影像中存在大量阴影以及大量的视差阶跃区域导致立体匹配效果不佳的问题,提出一种适应于城市遥感影像对的立体匹配算法。对算法所使用的匹配代价函数、代价聚合方法和视差优化方法等进行研究,改进了匹配代价函数,利用多阶加权Census算法减小噪声等因素的影响。在代价聚合中加入建筑边缘信息的约束条件,在视差细化部分充分考虑城市建筑的形态特点,对视差图进行优化。实验结果表明:在MiddleBury数据集上,本文算法的准确率比经典SGM算法提升4.54%;在城市区域WorldView-2立体影像对上,建筑屋顶的高程方差为0.71,满足基于城市卫星遥感影像对获取高精度视差图的要求,能够为城市三维重建提供良好的条件。

To solve the problem of poor stereo matching effect owing to numerous shadows and disparity step regions in urban satellite remote sensing images,a stereo matching algorithm suitable for urban remote sensing image pairs was proposed. The matching cost function,cost aggregation method,disparity,and optimization method used by the algorithm were investigated. First,the matching cost function was improved and the multi-order weighted census algorithm was used to reduce the influence of noise and other factors. Subsequently,the constraints of the building edge information were added to the cost aggregation. Finally,regarding disparity refinement,the disparity map was optimized by fully considering the characteristics of urban building morphology. The experimental results show that on the Middlebury dataset,the accuracy of this algorithm is 4. 54% higher than that of the classic SGM algorithm. On the WorldView-2 stereo image pair in the urban area,the variance of the building roof elevation is 0. 71. The requirements to obtain high-precision disparity maps are met based on urban satellite remote sensing images and good conditions for urban three-dimensional reconstruction are provided.