An Inverse Procedural Modeling and Hybrid Rendering Approach for Large-Scale Urban Scenes Visualization

A novel procedural modeling and hybrid level-of-detail rendering method is introduced for visually lossless urban scenes visualization. Our user-assisted inverse procedural modeling approach allows the user to exploit repetitions and symmetries of facades to create a split grammar representation of the input, which achieved a 70% compression factor averagely. We extract lines and points from the input models and provide their simplifications encoded in a data structure that allows for a quick and automatic LOD selection. Projected area is used as a LOD selector to combine points, lines, and polygon models that contain procedural textures. Our implementation shows a 10 times speed-up as compared to the ground truth rendered as full geometry, and is about 5 times faster compared to the geometric LOD. The quality of the results is indistinguishable from the original that was confirmed by a user study conducted with 72 subjects and an automated dynamic range independent metric.