290 likes | 399 Views
This presentation delves into the potential of visualizing photographic coverage in spaces, proposing innovative approaches to enhance photo tourism and augmented reality experiences. By employing blind camera calibration techniques, image feature matching, and triangulation algorithms, we explore effective methods to render augmented reality coverage. Additionally, we discuss next steps such as integrating virtual lighting and projectors, and using passive markers to guide photographers to unique shots. The fusion of technology and creativity promises to transform how we perceive and capture the world.
E N D
Visualizing Photo Coverage Phillip Isola 12/3/10
Big Idea What would be possible if we could see sight?
Big Idea What would be possible if we could see sight?
Big Idea How can we effectively visualize thephotographic coverage of a space
Background • Photo tourism • e.g. Snavely et al. 2007 • Camera + projector augmented reality • e.g. Raskar et al. 2003
Background • Multi-camera tracking • e.g. Khan and Shah 2003 • Structure from motion • e.g. Mooser et al. 2009
Possible approaches • Physical projection • Virtual environments • GPS, compass, etc • Image feature matching
My current approach • Use blind camera calibration techniques to solve correspondence problem • Interpolate and render augmented reality coverage / modulation
Algorithm Extract features Image processing KNN to generate candidate matches RANSAC on fundamental matrix Correspondence Triangulate Interpolate based on matches Smooth Interpolation and rendering
Features Extract features SURF (Bay et al. 2008)
KNN to generate candidate matches Correspondence Find highest probability set of unique matches Reject weak matches
KNN to generate candidate matches Correspondence
KNN to generate candidate matches Correspondence
KNN to generate candidate matches Correspondence Matches in both directions agree, so probably fairly unambiguous
RANSAC on fundamental matrix Correspondence
RANSAC on fundamental matrix Epipolar geometry (From http://en.wikipedia.org/wiki/Epipolar_geometry)
RANSAC on fundamental matrix RANSAC matches Reject matches that don’t well fit the epipolar constraints (fundamental matrix) Red box shows estimate of a slice of camera 1’s view frustum (using planar best fit)
Triangulation Triangulate
Interpolation Interpolate based on matches
Smooth Rendering
Non-planar geometry View 1 View 2
Two cameras View 1 View 2
Virtual laser pointer View 1 View 2
Virtual paintbrush + + +
Next steps • Virtual lights and projectors • Passive markers • Matches with internet • Guide photographers to unique shots or point out what was interesting in the past • Sikuli in the real world • ‘Program’ augmented reality systems by assigning functions to views (from:http://sikuli.org/)
Summary • Photographs mark views of the world • They tell us that view is perhaps interesting, is definitely already captured • Cameras can be used as semantic projectors