Scene Reconstruction from Community Photo Collections, Michael Goesele
Sprache des Titels:
With the recent rise in popularity of Internet photo sharing sites, community photo collections---collections of images contributed by users online---have emerged as a powerful new type of image dataset for computer vision and computer graphics research. Compared to standard datasets captured under laboratory conditions, community photo collections show an extreme variability since the images were captured by a large number of photographers with various cameras and at different times. In this talk, I will first describe our work on geometry reconstruction from community photo collections. We employ a robust multi-view stereo approach which yields high quality results even under such difficult conditions. I will then discuss how we can use images captured from different viewpoints and under varying lighting conditions to reconstruct a more complete scene model including geometry, reflectance, and distant illumination. I will close with an outlook on open research problems in the area.