Point clouds can be obtained by airborne or terrestrial LiDAR scanning or directly from street-view panoramic imagery. Point clouds obtained from airborne scanners cover most roof information but lack facade information. Similar but orthogonal to airborne scanning, street-view point clouds cover well of the facade information but lack roof points. Therefore, it is necessary to combine those point clouds in order to construct a complete building model. For this combination, simplification is needed in which edge preservation, outlier removal, noise smoothing and uniform density have to be considered as these properties are desired in many post-processing applications.
In this thesis, an algorithm pipeline is proposed that is able to take in an outlier-ridden, noisy and non-uniform roof and facade point clouds and generate an outlier-removed, edge-aware and uniform point cloud. The algorithm pipeline can be divided into two independent phases: outlier removal and simplification. The outlier removal algorithm can remove singly scattered and small cluster of outliers, whereas the simplification algorithm pipeline is able to generate a noise-reduced, uniform and edge-aware point cloud. The pipeline is validated to be able to achieve the objectives. Proof of efficiency in running-time is given, so that it can be used for processing large-size real-scene point clouds.