Automatic 3-D Building Model Reconstruction from Very High Resolution Stereo Satellite Imagery

Abstract

Recent advances in the availability of very high-resolution (VHR) satellite data together withefficient data acquisition and large area coverage have led to an upward trend in their applicationsfor automatic 3-D building model reconstruction which require large-scale and frequent updates,such as disaster monitoring and urban management. Digital Surface Models (DSMs) generatedfrom stereo satellite imagery suffer from mismatches, missing values, or blunders, resulting inrough building shape representations. To handle 3-D building model reconstruction using suchlow-quality DSMs, we propose a novel automatic multistage hybrid method using DSMs togetherwith orthorectified panchromatic (PAN) and pansharpened data (PS) of multispectral (MS) satelliteimagery. The algorithm consists of multiple steps including building boundary extraction anddecomposition, image-based roof type classification, and initial roof parameter computation whichare prior knowledge for the 3-D model fitting step. To fit 3-D models to the normalized DSM(nDSM) and to select the best one, a parameter optimization method based on exhaustive searchis used sequentially in 2-D and 3-D. Finally, the neighboring building models in a building blockare intersected to reconstruct the 3-D model of connecting roofs. All corresponding experimentsare conducted on a dataset including four different areas of Munich city containing 208 buildingswith different degrees of complexity. The results are evaluated both qualitatively and quantitatively.According to the results, the proposed approach can reliably reconstruct 3-D building models, eventhe complex ones with several inner yards and multiple orientations. Furthermore, the proposedapproach provides a high level of automation by limiting the number of primitive roof types and byperforming automatic parameter initialization.

A Robust Performance Evaluation Metric for Extracted Building Boundaries From Remote Sensing Data

“…The NMAD is considered as a robust metric to evaluate the extracted building shapes in the presence of outliers with a non-normal distribution [22,50]. The MAE is another frequently used metric, similar to the Chamfer distance used by several authors to evaluate the extracted 3D buildings [21].…”

“…The NMAD ν and MAEμ are defined in Eqs. 12 and 13, respectively, where h is the set of all minimum distances from the extracted points (S e ) to the reference points (S r ) and c ν = 1.4826 is a constant to normalise the distance [22]. However, none of these satisfies the properties of a mathematical metric discussed earlier.…”

“…The existing metrics [12,14] for evaluating the performance of building boundary extraction techniques can be categorised based on matched rates (e.g., completeness, correctness and quality rates) [6,15], shape similarity (e.g., turning angle, Fourier descriptors, dominant angle rotation) [16,17], positional accuracy (e.g., the Root Mean Square Error (RMSE), the Normalised Median Absolute Deviation (NMAD) and the Mean Absolute Error (MAE)) [18][19][20][21][22]. The matched rates are mainly used by the area-based, also known as the pixel-based, approach [12].…”

“…It is based on the distance of each corner of an extracted building polygon to a corresponding reference corner. The NMAD and the MAE are used to investigate the accuracy of the reconstructed 3D buildings from the satellite imagery [21,22]. Yet, there is no standard metric for evaluation of the extracted building boundaries in the remote sensing community [27].…”

Digital Modelling and Accuracy Verification of a Complex Architectural Object Based on Photogrammetric Reconstruction

“…A 3D building reconstruction can be derived from different types of data-satellite data [31], stereoscopic data [32] and LiDAR data [33][34][35][36] are used for this purpose. Due to the widespread use of unmanned aerial vehicles (UAV) technology, there is an increasing number of studies demonstrating the potential for reconstruction both based on data acquired only with UAVs and their combination with terrestrial data [15,[37][38][39][40][41].…”