Assessing the quality of building height extraction from ZiYuan-3 multi-view imagery

Abstract

ABSTRACTThe ZiYuan-3 (ZY-3) satellite, equipped with three-line array scanners, is China’s first civilian stereo mapping satellite. In this paper, the first assessment of the stereoscopic capacities of the ZY-3 imagery for building height estimation in heterogeneous urban areas is performed. Digital surface models (DSMs) are generated and optimized by different stereo pairs of a ZY-3 triplet for the city of Wuhan, China. The normalized DSMs (nDSMs) are computed as the height of the off-terrain objects by using morphological top-hat by reconstruction of the DSMs. We adopt random forest classification and object-based building segmentation to detect the buildings from the orthographic image, and estimate the building height by assigning the maximum nDSM value within the building regions. The actual heights of 400 buildings are used as reference data. Comparisons between the building heights extracted from the different stereo pairs are presented, indicating a higher accuracy by the nadir-forward stereo pair. The performance of the ZY-3 triplet is also compared with that of a WorldView-2 (WV-2) stereo pair, better result is achieved by ZY-3 nadir-forward image pair.