3D Reconstruction and Visualization of Urban Scenes from Uncalibrated Wide-Baseline Image Sequences

In this paper we show how based on a number of different techniques it is possible to fully automatically generate basic ingredients for high quality visualizations of urban areas characterized by vertical facade planes from nothing but uncalibrated wide-baseline image sequences without using any markers or ground control. At the core of our algorithms are least-squares matching, projective geometry based reconstruction, robust estimation based on random sample consensus – RANSAC, auto-calibration, projective and Euclidean bundle adjustment, plane to plane homographies, as well as the robust estimation of image mosaics. Results for the Hradschin in Prague, Czechia, and Plaza Real in Barcelona, Spain, show the potential and shortcomings of the employed algorithms.