A parallel stereo algorithm that produces dense depth maps and preserves image features

Abstract

To cite this version:Pascal Fua. A parallel stereo algorithm that produces dense depth maps and preserves image features.

Animated Heads from Ordinary Images: A Least-Squares Approach

“…Given the camera geometry and distances of the heads to the cameras, an error of one pixel in disparity corresponds to an error in measured range of 8 to 10 millimeters. The precision of the correlation based algorithm we use is in the order of half a pixel, outliers excluded [7]. We therefore conclude that our fitting algorithm performs an effective and robust averaging of the input data.…”

“…We start the process by using a simple correlation-based algorithm [7] to compute a disparity map for each pair or triplet and by turning each valid disparity value into a 3-D point. When using many stereo pairs, as in the the case of a video sequence, this could result in a very large number of such 3-D points that form a noisy and irregular sampling of the underlying global 3-D surface.…”

“…We typically start with a set of stereo image pairs or a video sequence and use a correlation-based algorithm [7] to compute disparity maps. We then derive clouds of 3-D points and fit the animation mask to these.…”

Image-based spatio-temporal modeling and view interpolation of dynamic events

“…Plenoptic modeling [McMillan and Bishop, 1995] captured cylindrical panoramas from two nearby positions and after disparities were estimated, they could be warped to novel views. Lightfield rendering [Levoy and Hanrahan, 1996] and Lumigraph [Gortler et al, 1996] adopt a different standpoint. Using a large number of input images from arbitrary positions, they reconstruct the complete lightfield or lumigraph, which is just the plenoptic function as a function of position and orientation in regions free of occlusions.…”

“…[Ohta and Kanade, 1985] use dynamic programming to determine correspondences between edge features. [Fua, 1993] described a correlation based multi resolution algorithm to compute dense depth maps. [Okutomi and Kanade, 1993] developed a multibaseline stereo algorithm that can uses multiple baselines between different pairs of cameras.…”