Quantitative evaluation strategies for urban 3D model generation from remote sensing data

Over the last decade, several automatic approaches have been proposed to reconstruct 3D building models from aerial laser scanning (ALS) data. Typically, they have been benchmarked with data sets having densities of less than 25 points/m2. However, these test data sets lack significant geometric points on vertical surfaces. With recent sensor improvements in airborne laser scanners and changes in flight path planning, the quality and density of ALS data have improved significantly. The paper presents quantitative evaluation strategies for building extraction and reconstruction when using dense data sets. The evaluation strategies measure not only the capacity of a method to detect and reconstruct individual buildings but also the quality of the reconstructed building models in terms of shape similarity and positional accuracy. The paper presents the evaluation strategies for 3D building detection and building model reconstruction based on dense ALS data to benchmark the results in terms of quantifying the quality of the models, with respect to geometrical accuracy and the desired level of detail.Display Omitted High density aerial laser scanning data, approximate 225 points/m2.Building detection and building reconstruction from point clouds of urban building.Evaluation strategies for 3D building detection and building model reconstruction.Evaluation examining identical location, shape similarity and positional accuracy.Proposed method for generating building outlines.