Perception of procedurally generated virtual buildings

Sammanfattning: Due to the high amount of time it takes to manually design virtual buildings and cities it is often desired to automate the process. Procedural modeling is a way to achieve this by algorithmically generating buildings. By using inverse generation procedural descriptions of buildings can be extracted from a given model. When using a procedural generator to create a city some variance is needed during generation which might lead to unrealistic buildings if not handled correctly. This thesis investigates which features of buildings are most sensitive to change in regards to their realism. An inverse procedural modeler is implemented in C# using Unity, which can generate split grammar rules from input pictures of building facades and their defined layout. Generated rules can then be used to generate buildings. Photographs of facades on residential buildings are edited in Adobe Photoshop and fed to the implemented generator to recreate real buildings. These buildings and manually altered versions of them are used to carry out an experiment involving independent participants in order to find which features on facades detracts most from their realism when changed. The findings are that certain features impact a buildings realism more than others when changed. Color and window styles on a facade are especially sensitive. This knowledge can in the future be used to improve building generators such that they are careful in editing the window style and colors of created buildings.