Automatic building height estimation with shadow correction over heterogeneous compact cities using stereo Gaofen-7 data at sub-meter resolution

Abstract

Fine-grained building heights are indispensable in various aspects of urban energy consumption, environment, and climate. The highly accurate and refined building height product is expected to be produced by high-resolution remote sensing images in this paper. To obtain this product painlessly, we aim to design a framework that automatically extracts building heights within cities. Specifically, high-resolution remote sensing images are shadow corrected, on which the algorithm is applied to generate the digital surface model. Then the model is masked using OSM's building data. Finally, the height of each building is obtained by zonal statistics. The abovementioned data in Hong Kong are collected. The experimental results show that the highest accuracy is achieved in the medium building density area, outperforming the low and high-density areas by 1.42 and 4.06 m in RMSE. At the same time, the extracted building heights with shadow correction are closer to the reference heights. It can be concluded that extracting building heights using high-resolution remote sensing satellite imagery is feasible; the accuracy varies for buildings of different heights and densities; and correcting the shadows can significantly improve the accuracy of building height extraction. The innovations are the automatic process and the shadow correction. It allows for dynamic building height monitoring when long time series of remote sensing images are available and proves that shadow may harm height estimation performance. Our research sheds light on the design of building height estimation algorithms and provides height validation data in developed cities and height estimation solutions for developing cities.