## Advances in Computational Stereo (2003)

### Cached

### Download Links

- [www.prip.tuwien.ac.at]
- [www.cs.jhu.edu]
- [www.cs.duke.edu]
- [www0.cs.ucl.ac.uk]
- DBLP

### Other Repositories/Bibliography

Venue: | IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE |

Citations: | 149 - 2 self |

### BibTeX

@ARTICLE{Brown03advancesin,

author = {Myron Z. Brown and Darius Burschka and Gregory D. Hager},

title = {Advances in Computational Stereo},

journal = {IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE},

year = {2003},

volume = {25},

number = {8},

pages = {993--1008}

}

### Years of Citing Articles

### OpenURL

### Abstract

Extraction of three-dimensional structure of a scene from stereo images is a problem that has been studied by the computer vision community for decades. Early work focused on the fundamentals of image correspondence and stereo geometry. Stereo

### Citations

8530 |
Introduction to Algorithms
- Cormen, Leiserson, et al.
- 1990
(Show Context)
Citation Context ...ted. 3.2.1 Dynamic Programming Dynamic programming is a mathematical method that reduces the computational complexity of optimization problems by decomposing them into smaller and simpler subproblems =-=[20]-=-. A global cost function is computed in stages, with the transition between stages defined by a set of constraints. For stereo matching, the epipolar monotonic ordering constraint allows the global co... |

3158 |
A.: “Multiple View Geometry in Computer Vision
- Hartley, Zisserman
- 2004
(Show Context)
Citation Context ...mber 115536. æ 0162-8828/03/$17.00 ß 2003 IEEE Published by the IEEE Computer Society methods, methods for dealing with occlusion, and real-time implementations. Recent books by Hartley and Zisserma=-=n [33]-=- and Faugeras and Luong [28] provide a wealth of information on the geometric aspects of multiple view stereo. We have also elected not to discuss recent developments in performance analysis but inste... |

1970 | An iterative image registration technique with an application in stereo vision
- Lucas, Kanade
- 1981
(Show Context)
Citation Context ...f pixels, then the disparity at a point may be estimated using least squares on the system of linear differential equations at each pixel in an n pixel window of points p1;p2; ...;pn about that point =-=[51], where 2 6 A-=- 6 4 rxE p1 rxE p2 rxE pnxn v A T A 1 A T b; ð3:2Þ ð ð : Þ Þ 3 7 5 and b 2 6 4 ð Þ ð ð : Þ Þ 3 7 5 : ð3:3Þ ð Þ Et p1 Et p2 Et pnxn The metric used here is the same sum of squared di... |

1736 | Schunck.: Determining optical flow
- Horn, B
- 1981
(Show Context)
Citation Context ...erations, making it independent of the template size. Variations include [26], [56], [57], and [76]; [57] provides a thorough discussion. 3.1.2 Gradient Methods Gradient-based methods or optical flow =-=[35]-=-, seek to determine small local disparities between two images by formulating a differential equation relating motion and image brightness. In order to do this, the assumption is made that the image b... |

1471 | Good features to track
- Shi, Tomasi
- 1994
(Show Context)
Citation Context ...nt between frames and the matrix A defines the in-plane rotation, scale, and shear transformations between frames. The parameters are computed directly from spatio-temporal intensity gradients, as in =-=[73]-=-. For epipolar-rectified imagery, only the horizontal parameters are computed. Segmentation and affine parameter estimation are computed iteratively, and patches with similar affine parameters are mer... |

1384 | Fast Approximate Energy Minimization via Graph Cuts, Proc
- Boykov, Veksler, et al.
- 2001
(Show Context)
Citation Context ...te Ford-Fulkerson style augmenting paths algorithm, which they show to be much faster in practice than standard push-relabel approaches (by factors of 2 to 5 for the examples provided). Boykov et al. =-=[13]-=- propose expansion move and swap move algorithms that can simultaneously modify labels of arbitrarily large pixel sets. Kolmogorov and Zabih [45] propose a graph architecture in which the vertices rep... |

1032 | A taxonomy and evaluation of dense two-frame stereo correspondence algorithms
- Scharstein, Szeliski
- 2002
(Show Context)
Citation Context ... aspects of multiple view stereo. We have also elected not to discuss recent developments in performance analysis but instead refer to a very complete and recent discussion by Scharstein and Szeliski =-=[69]-=- (See also http://www.middlebury.edu/stereo for implementations, test data, and results). No review of this nature can cite every paper that has been published. We have included what we believe to be ... |

797 | An experimental comparison of min-cut/max-flow algorithms for energy minimization
- Boykov, Kolmogorov
- 2004
(Show Context)
Citation Context ...c programming were 46.74 percent and 45.12 percent correct, respectively. Recent work on graph cuts has produced both new graph architectures and energy minimization algorithms. Boykov and Kolmogorov =-=[14]-=- have developed an approximate Ford-Fulkerson style augmenting paths algorithm, which they show to be much faster in practice than standard push-relabel approaches (by factors of 2 to 5 for the exampl... |

512 |
Introductory Techniques for 3-D Computer Vision. Upper Saddle River
- Trucco, Verri
- 1998
(Show Context)
Citation Context ...orizontal scan-lines, the two-dimensional correspondence search problem is again reduced to a scan-line search, greatly reducing both computational complexity and the likelihood of false matches. See =-=[82]-=- and [91] for details on algorithms to compute this rectification. 3 CORRESPONDENCE Asnotedpreviously,stereodisparitiesmaybedeterminedina numberof waysandbyexploiting anumber of constraints.All of the... |

457 | A theory of shape by space carving
- Kutulakos, Seitz
- 1992
(Show Context)
Citation Context ...model the scene as a mesh that is iteratively updated to minimize an objective function. Faugeras and Keriven [27] propose a similar method that models the scene using level sets. Kutulakos and Seitz =-=[48]-=- represent the scene as a volume and propose a space carving method to refine the surface. While these methods may be applied to binocular stereo, the object-centered representations are most powerful... |

322 | Determining the epipolar geometry and its uncertainty: A review
- Zhang
- 1998
(Show Context)
Citation Context ... scan-lines, the two-dimensional correspondence search problem is again reduced to a scan-line search, greatly reducing both computational complexity and the likelihood of false matches. See [82] and =-=[91]-=- for details on algorithms to compute this rectification. 3 CORRESPONDENCE Asnotedpreviously,stereodisparitiesmaybedeterminedina numberof waysandbyexploiting anumber of constraints.All of these method... |

316 | The Geometry of Multiple Images
- Faugeras, Luong, et al.
- 2001
(Show Context)
Citation Context ...$17.00 ß 2003 IEEE Published by the IEEE Computer Society methods, methods for dealing with occlusion, and real-time implementations. Recent books by Hartley and Zisserman [33] and Faugeras and Luong=-= [28]-=- provide a wealth of information on the geometric aspects of multiple view stereo. We have also elected not to discuss recent developments in performance analysis but instead refer to a very complete ... |

270 |
A Stereo Matching Algorithm with an Adaptive Window: Theory and Experiment
- Kanade, Okutomi
- 1994
(Show Context)
Citation Context ...ow sizes. Another approach to reduce sensitivity to occlusion is to adaptively resize the window size and shape in order to optimize the match similarity near occlusion boundaries. Kanade and Okutomi =-=[39]-=- propose an iterative method for determining the window size. The window size is initialized to be very small, and the match uncertainty is computed. The window size is then expanded by one pixel in e... |

266 | Computing visual correspondence with occlusions using graph cuts
- Kolmogorov, Zabih
- 2001
(Show Context)
Citation Context ...mage, ground truth, Muhlmann et al.’s area correlation algorithm [57], dynamic programming (similar to Intille and Bobick [36]), Roy and Cox’s maximum flow [65], and Kolmogorov and Zabih’s graph=-= cuts [45]-=-.s996 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 25, NO. 8, AUGUST 2003 TABLE 2 Common Block-Matching Methods (See Fig. 4 for Visual Description of Terms) cross correlation, ... |

234 | A maximum-flow formulation of the ncamera stereo correspondence problem
- Roy, Cox
- 1998
(Show Context)
Citation Context ... Szeliski [69]. Left to right: left stereo image, ground truth, Muhlmann et al.’s area correlation algorithm [57], dynamic programming (similar to Intille and Bobick [36]), Roy and Cox’s maximum f=-=low [65], -=-and Kolmogorov and Zabih’s graph cuts [45].s996 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. 25, NO. 8, AUGUST 2003 TABLE 2 Common Block-Matching Methods (See Fig. 4 for Visu... |

219 | Computing occluding and transparent motions - Irani, Rousso, et al. - 1994 |

197 | Non-parametric local transforms for computing visual correspondence
- Zabih, Woodfill
- 1994
(Show Context)
Citation Context ... example is the sum of absolute differences (SAD), which is often used for computational efficiency. See Aschwanden and Guggenbuhl [1] for an extensive comparison of these metrics. Zabih and Woodfill =-=[89]-=- propose an alternative method for computing correspondence by applying local nonparametric transforms to the images before matching. In order to eliminate sensitivity to radiometric gain and bias, a ... |

175 | Small vision systems: Hardware and implementation
- Konolige
- 1997
(Show Context)
Citation Context ...correlation computations [3]. Due to its simplicity and overall good performance at eliminating bad depth estimates, left-right consistency has been implemented in many real-time stereo systems [26], =-=[46]-=-, [55]. The ordering constraint in stereo may also be used to detect occlusion. The relative ordering of points along a pair of stereo scan-lines is generally monotonic, assuming that there are no nar... |

165 | Markov random fields with efficient approximations
- Boykov, Veksler, et al.
- 1998
(Show Context)
Citation Context ...stimation are computed iteratively, and patches with similar affine parameters are merged after each iteration. The segmentation algorithm used is based on the multiway cut algorithm of Boykov et al. =-=[12]-=-. A similar graph theoretic algorithm is discussed in Section 3.2.3 for global stereo correspondence. Unlike most feature-based methods, dense disparities are explicitly defined for this segmentation-... |

141 |
A survey of sensor planning in computer vision
- Tarabanis, Allen, et al.
- 1995
(Show Context)
Citation Context ...zes the number of views required to capture all points in the scene. A detailed discussion of sensor planning is beyond the scope of this paper. A review of this topic is provided in Tarabanis et al. =-=[78]-=-. 4.5 Comparisons Relatively little has been done to compare the variety of algorithms proposed for detecting, reducing sensitivity to, or measuring occlusions. Egnal and Wildes [25] recently conducte... |

138 | Depth Discontinuities by Pixel-to-Pixel Stereo
- Birchfield
- 1999
(Show Context)
Citation Context ...ine, in addition to time required to compute local cost functions. Using dynamic programming, the complexity is only OðN 2 Þ, much smaller than that of the other representation. Birchfield and Tomas=-=i [9] p-=-ropose a greedy algorithm for reducing this complexity to OðNDlogDÞ by pruning nodes when locally lower cost alternatives are available. In addition to minimizing the global cost for independent sca... |

138 | Adaptive Least Squares Correlation: A Powerful Image Matching T e c h n i q u e . S o u t h A f r i c a n J o u r n a l of P h o t o g r a m m e t r y , Remote Sensing and Cartography
- Gruen
- 1985
(Show Context)
Citation Context ...parable to exhaustive search. In addition to estimating translation (i.e., disparity) for each pixel, this framework may be extended to also estimate more general transformations. For instance, Gruen =-=[32]-=- estimates local affine transformations between images in order to compensate for perspective effects. It is worth noting that computation required to perform the optimization grows linearly with the ... |

135 |
Stereo vision for planetary rovers: Stochastic modeling to near real-time implementation
- Matthies
- 1992
(Show Context)
Citation Context ...his system to obstacle detection for unmanned ground vehicles may be found in [55]. A complete discussion of their algorithm development and details on an earlier version of the system is provided in =-=[54]-=-. Also in 1995, Kimura et al. reported on a video-rate stereo machine developed at CMU [41]. This was the first published stereo system capable of 30 fps (with 256 x 240 pixel images). Like the iWarp ... |

124 |
Depth from Edge and Intensity Based Stereo
- Baker, Binford
- 1981
(Show Context)
Citation Context ... alternatives are available. In addition to minimizing the global cost for independent scanlines (i.e., intrascanline search), interscanline constraints may also be applied to reduce ambiguity. Baker =-=[2]-=- proposed a dynamic programming method that first computes disparities independently for each scanline and then detects and corrects estimates that violate interscanline consistency constraints. Ohta ... |

119 | Object-centered surface reconstruction: Combining multiimage stereo and shading
- Fua, Leclerc
- 1995
(Show Context)
Citation Context ... shown to yield results comparable to the best graph cut methods in [69]. Another class of global methods seeks to reconstruct a scene without explicitly establishing correspondences. Fua and Leclerc =-=[30]-=- model the scene as a mesh that is iteratively updated to minimize an objective function. Faugeras and Keriven [27] propose a similar method that models the scene using level sets. Kutulakos and Seitz... |

113 | Multiway cut for stereo and motion with slanted surfaces
- Birchfield, Tomasi
- 1999
(Show Context)
Citation Context ...course, the improvement will vary with the percentage of higher-level features successfully matched. Another feature-based approach is to first segment the images and then match the segmented regions =-=[11]-=-, [62]. Birchfield and Tomasi [11] segment stereo images into small planar patches for which correspondence is then determined. As with most feature-based methods, this reduces the match sensitivity t... |

107 |
A Bayesian approach to binocular stereopsis
- Belhumeur
- 1996
(Show Context)
Citation Context ...sts over two-dimensional regions defined as intervals between vertical edges. The local cost function for each interval is defined as the variance of the pixel intensities in that interval. Belhumeur =-=[5]-=- proposed a two-stage approach, first computing intrascanline solutions using dynamic programming and then smoothing disparities between scanlines. This smoothing is done by taking each adjacent three... |

98 | A Cooperative Algorithm for Stereo Matching and Occlusion Detection
- Zitnick, Kanade
- 1999
(Show Context)
Citation Context ...t al. [52] propose a similar approach, but allow the diffusion equations to be anisotropic near intensity edges to better localize depth discontinuities. An interesting approach by Zitnick and Kanade =-=[90]-=- combines diffusion of the support region with inhibition of support for pixels along similar lines of sight. This ensures that the diffusion process does not violate the uniqueness constraint. Fusiel... |

97 | Stereo Matching with Nonlinear Diffusion
- Scharstein, Szeliski
- 1998
(Show Context)
Citation Context ...ion methods for global stereo matching, a number of other approaches have been used as well. Two of the most notable are nonlinear diffusion and belief propagation. Shah [71], Scharstein and Szeliski =-=[68]-=-, and Mansouri [52] aggregate support using various models for nonuniform diffusion, rather than using fixed-size, rectangular windows. We TABLE 3 Occlusion Methods discuss these methods more in Secti... |

87 |
A.: “Computer Vision - Threedimensional Data from Images
- Klette, Schlüns, et al.
- 1998
(Show Context)
Citation Context ...s (e.g., epipolar geometry) and assumptions (e.g., image brightness constancy and surface smoothness) are commonly exploited to make the problem tractable. For a complete discussion, see chapter 4 of =-=[43]-=-. The reconstruction problem consists of determining threedimensional structure from a disparity map, based on known camera geometry. The depth of a point in space P imaged by two cameras with optical... |

84 | Obstacle detection for unmanned ground vehicles: a progress report", Robotics Research
- Matthies, Kelly, et al.
- 1996
(Show Context)
Citation Context ...ation computations [3]. Due to its simplicity and overall good performance at eliminating bad depth estimates, left-right consistency has been implemented in many real-time stereo systems [26], [46], =-=[55]-=-. The ordering constraint in stereo may also be used to detect occlusion. The relative ordering of points along a pair of stereo scan-lines is generally monotonic, assuming that there are no narrow oc... |

77 | Surfaces from stereo: Integrating feature matching, disparity estimation, and contour detection
- Hoff, Ahuja
- 1989
(Show Context)
Citation Context ...ten caused by occlusion regions. Wildes [86] detects discontinuities in both surface orientation and depth by comparing local histograms of these values with a Kolmogorov-Smirnov test. Hoff and Ahuja =-=[34]-=- detect depth and orientation discontinuities by fitting local planar patches to semicircular regions of the data. If two of these planes making up a circular region differ in depth or orientation by ... |

76 | Efficient stereo with multiple windowing
- Fusiello, Roberto
- 1997
(Show Context)
Citation Context ...diffusion of the support region with inhibition of support for pixels along similar lines of sight. This ensures that the diffusion process does not violate the uniqueness constraint. Fusiello et al. =-=[31]-=- propose a method based on multiple windows. This method considers nine correlation windows centered about the desired point and takes the estimated disparity with the minimum error. The window associ... |

74 |
Structure from stereo-a review
- Dhond, Aggarwal
- 1989
(Show Context)
Citation Context ...aluating performance, and a survey of well-known approaches at that time. Stereo continued to be a significant focus of research in the computer vision community through the 1980s. Dhond and Aggarwal =-=[23]-=- reviewed many stereo advances in that decade, including a wealth of new matching methods, the introduction of hierarchical processing, and the use of trinocular constraints to reduce ambiguity in ste... |

71 | Ordinal measures for image correspondence
- Bhat, Nayar
- 1998
(Show Context)
Citation Context ... than the intensities themselves, they are somewhat robust to outliers. These rank measures, however, are sensitive to distortion due to perspective projection. To remedy this problem, Bhat and Nayar =-=[7]-=- propose an alternative ordinal measure based on the distance between rank value positions rather than the rank values themselves. This provides some robustness to perspective distortion. The fundamen... |

66 |
Detecting binocular half-occlusions : empirical comparisons of five approaches
- Egnal, Wildes
- 2002
(Show Context)
Citation Context ... Tarabanis et al. [78]. 4.5 Comparisons Relatively little has been done to compare the variety of algorithms proposed for detecting, reducing sensitivity to, or measuring occlusions. Egnal and Wildes =-=[25]-=- recently conducted an analysis of four of the simpler approaches. All of these fall into the detecting occlusion class of algorithms discussed above. The four methods compared are bimodality, match g... |

56 |
Experimental results from a comparative study on correlation-type registration algorithms
- Aschwanden, Guggenbuhl
- 1992
(Show Context)
Citation Context ...each with different normalization schemes have been used. One popular example is the sum of absolute differences (SAD), which is often used for computational efficiency. See Aschwanden and Guggenbuhl =-=[1]-=- for an extensive comparison of these metrics. Zabih and Woodfill [89] propose an alternative method for computing correspondence by applying local nonparametric transforms to the images before matchi... |

54 |
3-d surface description from binocular stereo
- Cochran, Medioni
- 1992
(Show Context)
Citation Context ...I1Þ and right ðI2Þ images to obtain a consistency constraint that may be used to detect occlusion or other sources of mismatch.sBROWN ET AL.: ADVANCES IN COMPUTATIONAL STEREO 1003 Cochran and Medio=-=ni [17]-=- incorporate an edge map into their postfiltering scheme. The disparity map is smoothed, keeping only the disparities associated with edges unaltered. Then, those points with large disparity differenc... |

53 | K.: Real-time tracking of multiple people using continuous detection
- Beymer, Konolige
- 1999
(Show Context)
Citation Context ...z Pentium II. Both the SVM II and the SVS use SAD on LoG transformed image pixels to determine the disparities and left/right checking for post-filtering. The SVS has been used for real-time tracking =-=[6]-=- and is distributed commercially by Videre Design. Another commercial product for real-time stereo is the Point Gray Digiclops trinocular stereovision system. The Triclops software development kit (SD... |

47 | Calculating Dense Disparity Maps from Color Stereo Images, an Efficient Implementation
- Muhlmann, Maier, et al.
- 2002
(Show Context)
Citation Context ...3. Comparative results on images from the University of Tsukuba, provided by Scharstein and Szeliski [69]. Left to right: left stereo image, ground truth, Muhlmann et al.’s area correlation algorith=-=m [57], dy-=-namic programming (similar to Intille and Bobick [36]), Roy and Cox’s maximum flow [65], and Kolmogorov and Zabih’s graph cuts [45].s996 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGE... |

43 | Automatic Extraction of Generic House Roofs from High Resolution Aerial Imagery
- Bignone, Henricsson, et al.
- 1996
(Show Context)
Citation Context ...so sensitive to regions of uniform texture in images. Feature-based methods seek to overcome these problems by limiting the regions of support to specific reliable features in the images (e.g., edges =-=[8]-=-, [84], curves [72], etc.). Of course, this also limits the density of points for which depth may be estimated. Throughout the 1980s, featurematching methods for stereo correspondence received signifi... |

38 |
Integrated Person Tracking Using Stereo
- Darrell, Gordon, et al.
- 1998
(Show Context)
Citation Context ...240 pixel images on a 700 MHz Pentium III. With inexpensive and compact realtime systems like these now commercially available, many applications that previously were impractical (e.g., tracking [6], =-=[22]-=- and virtual reality [40]) are now being intensely explored. 6 CONCLUSIONS After roughly 30 years of research on computational stereo (in the computer vision community), many elements of stereo algori... |

38 |
A Generalized Ordering Constraint for Stereo Correspondence7
- Yuille, Poggio
- 1984
(Show Context)
Citation Context ...used to detect occlusion. The relative ordering of points along a pair of stereo scan-lines is generally monotonic, assuming that there are no narrow occluding objects in the scene. Yuille and Poggio =-=[88]-=-, Little and Gillett [50], and Silva and Santos-Victor [74] have proposed methods that check for out of order matches, which may indicate the presence of occlusion. Another approach to detecting occlu... |

31 | Bias in robust estimation caused by discontinuities and multiple structures
- Stewart
- 1997
(Show Context)
Citation Context ... image differences (e.g., perspective differences and sensor noise). Sara and Bajcsy [66] propose a robust normalized cross correlation algorithm based on robust covariance matrix estimation. Stewart =-=[75]-=- characterizes the behavior of a number of robust correlation measures near discontinuities (i.e., occlusion regions). These include least median of squares (LMS), least trimmed squares (LTS), M-estim... |

30 | Occlusion detectable stereo - occlusion patterns in camera matrix
- Nakamura, Matsuura, et al.
- 1996
(Show Context)
Citation Context ...in another stereo pair in which the given point is jointly visible. Thus, by summing the SSD metrics over all stereo pairs, the effects of these ambiguities are significantly reduced. Nakamura et al. =-=[58]-=- and Satoh and Ohta [67] propose a similar approach in what they call stereo by eye array (SEA), but they directly model occlusion in an attempt to remove its effect altogether. They define occlusion ... |

27 |
Implementation and Performance of Fast Parallel Multi-baseline Stereo Vision
- Webb
- 1993
(Show Context)
Citation Context ...f 0.5 fps with 320 x 240 pixel images. One system came close to achieving frame-rate in 1993. Webb implemented the multibaseline stereo algorithms of Kanade et al. on the CMU Warp machine [21], [38], =-=[85]-=-. Three images were used for this system. For efficiency, the TABLE 4 Real-Time Stereo Implementations sum of SAD (SSAD) was implemented. 64 iWarp processors were used to achieve 15 fps with 256 x 240... |

26 | Constraint-Based Sensor Planning for Scene Modeling
- Reed, Allen
- 2000
(Show Context)
Citation Context ...sed on the jump lengths as measured from two independent positions, again requiring active camera motion. A more general approach to handling occlusion via active vision is proposed by Reed and Allen =-=[64]-=-, who describe a sensor planning system that minimizes the number of views required to capture all points in the scene. A detailed discussion of sensor planning is beyond the scope of this paper. A re... |

23 |
ªA Parallel Stereo Algorithm that
- Fua
- 1993
(Show Context)
Citation Context ...lar region differ in depth or orientation by more than a threshold, there is evidence of occlusion. Stereo match consistency may also be used to detect occlusion boundaries. Chang et al. [15] and Fua =-=[29]-=- compute two disparity maps, one based on the correspondence from Fig. 10. Examples of typical occlusion (left) and less common narrow occlusion (right) that may be observed when viewing a fence, for ... |

22 |
Direct evidence for occlusions in stereo and motion
- Little, Gillett
- 1990
(Show Context)
Citation Context ... The relative ordering of points along a pair of stereo scan-lines is generally monotonic, assuming that there are no narrow occluding objects in the scene. Yuille and Poggio [88], Little and Gillett =-=[50]-=-, and Silva and Santos-Victor [74] have proposed methods that check for out of order matches, which may indicate the presence of occlusion. Another approach to detecting occlusion boundaries is based ... |

22 | A Fast Stereo Matching Method
- Sun
(Show Context)
Citation Context ...lock sums, redundant computations may be avoided and block matching complexity may be reduced to OðNDÞ operations, making it independent of the template size. Variations include [26], [56], [57], an=-=d [76]-=-; [57] provides a thorough discussion. 3.1.2 Gradient Methods Gradient-based methods or optical flow [35], seek to determine small local disparities between two images by formulating a differential eq... |