基于阴影的资源三号卫星数据城市建筑物高度估算

利用我国首颗民用高分辨率测绘卫星—ZY3影像,运用面向对象方法及形态学的思想,以形态学建筑物指数MBI和形态学阴影指数MSI,构建了基于知识规则的城市建筑物和阴影信息提取模型,很好地完成了北京方庄小区和厦门软件园的建筑物和阴影的提取,并实现了阴影长度的自动估算。利用太阳、卫星和建筑物成像几何关系,以阴影长度估算了城市建筑物高度,并以实测数据进行精度评价及误差分析。在以北京市方庄小区为例的试验中,建筑物高度提取的平均相对误差为7.08%,绝对误差小于6 m的建筑物占90%以上,80%以上的建筑物相对误差在10%以内,而误差较大的多为阴影较难识别的低矮建筑物和建筑密度较大使得阴影相互遮挡的高层建筑物。在厦门市软件园区实验中,94.3%的建筑物绝对误差在5 m以内,且计算高度与实际高度相关系数达到0.992,进一步验证了所用方法的可靠性。本文提出的建筑物信息提取方法,可很好地实现高分辨率影像城市建筑物高度估算,展示出资源三号卫星数据在城市建筑物信息提取方面的应用潜力。

Urban Building Heights Estimation from the Shadow Information on ZY-3 Images

The paper is based on the images of Chinese No. 3 Resource Satellite (ZY-3), the first civil high-resolution surveying and mapping satellite. A method combined with object-oriented approach and morphology ideas is developed in this paper, which not only extracted the information of buildings and shadows from the test sites (Fangzhuang Community in Beijing and the software park in Xiamen) successfully, but also estimated the shadow lengths automatically. Morphological Building Index(MBI) and Morphological Shadow Index(MSI) calculated from the merged images are used to extract information from city buildings and shadows. Subsequently, we estimate the building heights based on the lengths of shadows and their geometric relationship with the sun, the satellite and the buildings. Finally, we use field survey data to evaluate the accuracy of extraction and analyze the source of errors. In the example of Fangzhuang Community, the average relative error of the calculated building heights is 7.08%, and more than 80% of the building heights’ relative error is less than 10%. Moreover, 90% of the buildings’ absolute error is less than 6 meters. The relatively large errors mostly occurred on buildings that have low building heights with inconspicuous shadows, and high-rise buildings with overlapped shadows due to the high density of constructions. Meanwhile, the test site in Xiamen showed the reliability and accuracy of the method. 94.3% of the buildings’ absolute error is less than 5 meters, and the correlation coefficient between the actual heights and the estimated heights is around 0.992. Results demonstrate that the proposed method is able to accomplish building extraction and building height estimation effectively from high-resolution images. In addition, the experiment gives a preliminary indication of the Satellite ZY3, about its immense potential in applications on urban information extraction.