Bim geometry generation from low-resolution aerial photographs for building energy modeling

Building Energy Modeling (BEM) is a time consuming and costly process, primarily due to the extensive manual inputs required for model development. The definition of geometry inputs for models of existing buildings is particularly difficult. We present an automatic procedure for the analysis and segmentation of building facades in low-resolution aerial photographs in order to recover their semantic structure as well as to synthesize geometric models for reuse in the BEM process. An iterative approach for facade parsing is developed for the detection of facade structures which is robust to image noise and under-sampling artifacts as they are common in aerial photographs. In each step, a further refined shape grammar representing the high-level facade topology is inferred from the low-level facade element detection. Then, the lowlevel detection is improved by using feedback from the previously generated high-level structure. Based on this iterative analysis, we recover high-level facade topology by deriving a compact procedural description which takes into account symmetries and repeating patterns corresponding to floors and windows. The facade structure is encoded into common BEM formats. Moreover, synthesis of realistic facade textures is achieved by replacing low-resolution segments from the aerial images with best matching high-resolution pictures from a repository of facade elements.