DATA DRIVEN RULE PROPOSAL FOR GRAMMAR BASED FACADE RECONSTRUCTION

Today the demands on 3d models are steadily growing. At the same time, the extraction of man-made objects from measurement data is quite traditional. Often, the processes are still point based, with the exception of a few systems, which allow to automatically fit simple primitives to measurement data. The need to be able to automatically transform object representations, for example, in order to generalize their geometry, enforces a structurally rich object description. Likewise, the trend towards more and more detailed representations requires to exploit structurally repetitive and symmetric patterns present in man-made objects, in order to make extraction cost-effective. In this paper, we address the extraction of building facades in terms of a structural description. We extend our former work on facade reconstruction, which is based on a formal grammar to derive a structural facade description in the form of a derivation tree and uses a stochastic process based on reversible jump Markov Chain Monte Carlo (rjMCMC) to guide the application of derivation steps during the construction of the tree. We use measurements to improve the control of the rjMCMC process. This data driven approach reduces the number of false proposals and therefore the execution time.