Building facade interpretation from uncalibrated wide-baseline image sequences

We propose an approach for building facade interpretation ranging from uncalibrated wide-baseline image sequences to the extraction of windows. The approach comprises several novel features, such as determination of the facade planes by robust least squares matching, learning of implicit shape models for objects, particularly windows, and the determination of the latter by means of Markov Chain Monte Carlo (MCMC) employing an abstraction hierarchy generated via mathematical morphology. Results for the fully automatic approach show its potential and shortcomings.