The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Articles | Volume XL-2/W1
13 May 2013
 | 13 May 2013


M. Mohamed, T. Landes, P. Grussenmeyer, and W. Zhang

Keywords: Optical imagery, LiDAR, 3D building modeling, Quality assessment, RMSE, Surface indices, Volume indices

Abstract. The production of 3D city models requires the reconstruction of individual 3D building models. As the performance of data acquisition methods improves, the quality evaluation of building models in 3D has become an important issue. The main objective of the presented work is to introduce a multi-dimensional approach for assessing the quality of 3D building vector models. This approach performs assessments in 1D, 2D and 3D by comparing calculated building models to their reference. For 1D assessment, homologous points in two buildings to be compared are analyzed. For 2D assessment, homologous planes enter in the evaluation process. Quality of the planes under study is assessed by calculating a set of indices in vector format. For 3D assessment, building models are considered as one object. Quality of the buildings is assessed by calculation of vector volumetric quality factors. These factors require the not trivial calculation of vector intersection volumes which calculation is presented in the paper. Intersection volume is defined by superimposing the building model to be tested with the reference one. The multi-dimensional vector assessment approach has been applied to evaluate the building models produced with three different reconstruction processes created from different types of datasets. The datasets are obtained by photogrammetry (UltraCam-X and Zeiss LMK cameras), by LiDAR, and also by integration of photogrammetric and LiDAR datasets. The 1D, 2D or 3D assessment approach allows highlighting the source of deviations in the tested buildings. The error budget affecting the final product is not only composed of errors due to the reconstruction algorithm. Also errors due to the quality of the raw data, the processing of LiDAR data, of aerial data and the shape of the produced buildings should be considered.