Improving CityGML LOD-1 building models using terrestrial point cloud data

The elevation of the generated OGC CityGML LOD-1 models is usually determined as the height of the roof from ground. However, this criterion might result in significant discrepancies between the generated models and some real buildings, such as arcades and open spaces on first floors. This study develops a procedure to improve the OGC CityGML LOD-1 building models based on terrestrial point clouds derived from close-range images. The point clouds are generated from multiple photos acquired using non-metric cameras (including cell phones), and analyzed by the CMVS (Clustering Views for Multi-view Stereo) algorithm which is one of the dense matching methods for point cloud generation. Experimental results presented in this paper demonstrate the effectiveness of the developed procedure, and show that the improved LOD-1 building models are more close to real buildings in terms of shapes and appearances. In addition, this study also briefly compares the developed method with other conventional approaches (e.g., laser scanning and total station) for improving OGC CityGML LOD-1 building models.