ON THE ACCURACY ASSESSMENT OF LEAST-SQUARES MODEL-IMAGE FITTING FOR BUILDING EXTRACTION FROM AERIAL IMAGES

Model-based building extraction from aerial images has been an intensive research topic in the field of digital photogrammetry since the last decade. Based on our previous research work, the principle of Constructive Solid Geometry (CSG) is applied to model various buildings. Each building is represented by a combination of 3D primitives and each primitive is associated with a set of shape and pose parameters. Building reconstruction is implemented by adjusting the model parameters to fit model with images. We proposed a Least-squares Model-image Fitting (LSMIF) algorithm to obtain the optimal fit between model and images. In this paper, the performance of LSMIF is investigated. First, the convergence rate and pull-in range of the algorithm is analyzed. Then, how to use constraints to increase the convergence rate is introduced. Finally, 10 sets real data were tested to assess the theoretical accuracy of parameter determination. In order to assess the empirical accuracy, the results are also compared with manually measured data using an analytical plotter. This study reveals that the function of LAMIF is table and can generate qualified 3D information of building comparable to manually measured data.