Advertisement

DCF: Disparity computing framework for stereo vision systems

Open AccessPublished:November 14, 2022DOI:https://doi.org/10.1016/j.simpa.2022.100442

      Highlights

      • A computer program with the main components of a stereo vision system.
      • A software design that integrates the construction of disparity maps.
      • A software architecture to handle multiple scenarios in stereo vision systems.
      • A framework that integrates the main disparity calculation components discussed in the literature.

      Abstract

      Disparity maps are vital components of stereo vision systems as they encode the displacement of two or more images. However, previous works provide only a few implementation details, suggest processing steps that are not very well defined, and the software design is rarely discussed. Conversely, DCF applies the main components of a stereo vision system and integrates them to promote disparity map construction. As a result, DCF algorithms can be parameterized or executed with previously defined configurations. Thus, DCF outputs can be directed to different applications, such as benchmarking proposals, computer and robotic applications, triangulation, and 3D reconstruction.

      Keywords

      Code metadata
      Tabled 1
      Current code versionv2.0
      Permanent link to code/repository used for this code versionhttps://github.com/SoftwareImpacts/SIMPAC-2022-236
      Permanent link to Reproducible Capsulehttps://codeocean.com/capsule/4854862/tree/v1
      Legal Code LicenseGNU GPL-2.0 License
      Code versioning system usedgit
      Software code languages, tools, and services usedMATLAB
      Compilation requirements, operating environments & dependenciesLinux, Microsoft Windows, MATLAB R2014a
      If available Link to developer documentation/manualhttps://github.com/gabrieldgf4/disparity-computation-framework
      Support email for questions[email protected]

      1. Description of software

      The disparity calculation is one of the steps in the machine vision process using cameras. In the stereoscopic perspective, cameras record images from different points of view, usually simultaneously [
      • Furukawa Y.
      • Hernández C.
      Multi-view stereo: A tutorial.
      ,

      T. Schöps, J.L. Schönberger, S. Galliani, T. Sattler, K. Schindler, M. Pollefeys, A. Geiger, A multi-view stereo benchmark with high-resolution images and multi-camera videos, in: Conference on Computer Vision and Pattern Recognition, Vol. 2017, CVPR, 2017, pp. 2538–2547.

      ]. Although the cameras record the same scene, they are located in different positions in a previously configured displacement margin or an unrestrained configuration [

      G. Vieira, F. Soares, R. Parreira, G. Laureano, R. Costa, Depth map production: approaches, challenges and applications, in: Proceedings of XII Workshop de Visão Computacional, 2016, pp. 323–328.

      ]. The resulting images have some of the same scene elements, which are used to measure the shift of the images concerning the cameras [

      G. Vieira, F. Soares, N. Sousa, J. Gil, R. Parreira, G. Laureano, R. Costa, J. Ferreira, Stereo vision methods: from development to the evaluation of disparity maps, in: Proceedings of XIII Workshop de Visão Computacional, 2017, pp. 132–137.

      ]. This approach makes it possible to estimate depth from 2D images and three-dimensional scene construction using two-dimensional images [
      • Huertas A.
      • Matthies L.
      • Rankin A.
      Stereo-based tree traversability analysis for autonomous off-road navigation.
      ,
      • Roberts R.
      • Sinha S.N.
      • Szeliski R.
      • Steedly D.
      Structure from motion for scenes with large duplicate structures.
      ,
      • Dong H.
      • Yin S.
      • Jiang G.
      • Liu L.
      • Wei S.
      An automatic depth map generation method by image classification.
      ].
      The disparity calculation aims to measure the displacement of pixels between cameras [
      • Hirschmuller H.
      • Scharstein D.
      Evaluation of cost functions for stereo matching.
      ]. Pixels that move less have smaller disparity values. On the other hand, larger disparity values occur when pixel movement can be observed at non-adjacent positions. This measurement is computed from the coordinates of one or more target images with a reference image. The reference image, as its name suggests, is the starting point from which the displacement of pixels is observed.
      Disparity information indicates the proximity or distance of elements in the scene to the reference camera. The disparity is also treated as inverse depth since the disparity values are inversely proportional to depth [
      • Scharstein D.
      • Szeliski R.
      A taxonomy and evaluation of dense two-frame stereo correspondence algorithms.
      ,
      • Vieira G.
      • Soares F.A.A.M.N.
      • Laureano G.T.
      • Parreira R.T.
      • Ferreira J.C.
      A segmented consistency check approach to disparity map refinement.
      ]. When pixels have a smaller displacement between images, the disparity values indicate that objects in the scene are further away, i.e., they have greater depth values. Conversely, more significant disparities indicate that objects in the scene are closer to the reference camera, i.e., they have lower depth values.
      Figure thumbnail gr1
      Fig. 1Fixed Window method (FW) with different aggregation window sizes.
      Figure thumbnail gr2
      Fig. 2Segmented Based method (SB) with different cost functions and a fixed window aggregation size of 9 × 9.
      The disparity calculation process results in a new image named the disparity map. Stereo algorithms generally perform four steps in obtaining disparity maps [
      • Scharstein D.
      • Szeliski R.
      A taxonomy and evaluation of dense two-frame stereo correspondence algorithms.
      ]. The first is the matching cost computation that checks the similarity between pixels. The second one is the cost (support) aggregation which adds a neighborhood window in the pixel similarity evaluation. The third is the disparity computation/optimization step which calculates the displacement of pixels between the reference and target images. Finally, the fourth is the disparity refinement step, which applies adjustments to the disparity map to correct calculation errors.
      Evaluating the similarity between pixels is a challenging task for computer vision systems. Photometric distortions, specular surfaces, perspective distortions, ambiguous regions, repetitive patterns, occlusions, and noise can seriously compromise pixel matching [
      • Mattoccia S.
      • Giardino S.
      • Gambini A.
      Accurate and efficient cost aggregation strategy for stereo correspondence based on approximated joint bilateral filtering.
      ,
      • Mattoccia S.
      A locally global approach to stereo correspondence.
      ,

      S. Mattoccia, Stereo vision algorithms for fpgas, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2013, pp. 636–641.

      ]. As isolated pixels are susceptible to error, the support or aggregation window is used so that the neighbors of a central pixel reduce ambiguities and mismatches between pixels. This strategy (or area-based approach) presents more assertive results in relation to the pixel-by-pixel (or pixel-based approach) evaluation [
      • Hamzah R.A.
      • Ibrahim H.
      Literature survey on stereo vision disparity map algorithms.
      ]. However, the ideal size of the aggregation window is a parameter that needs to be found, and it is a parameter that can interfere with the execution time of the stereo vision algorithms.
      There are many disparity calculation algorithms, especially methods for the cost aggregation step [
      • Hamzah R.A.
      • Ibrahim H.
      Literature survey on stereo vision disparity map algorithms.
      ,
      • Lazaros N.
      • Sirakoulis G.C.
      • Gasteratos A.
      Review of stereo vision algorithms: from software to hardware.
      ,
      • Kumari D.
      • Kaur K.
      A survey on stereo matching techniques for 3D vision in image processing.
      ]. Some works discuss invariant approaches to the aggregation window size, such as the integral image [
      • Facciolo G.
      • Limare N.
      • Meinhardt-Llopis E.
      Integral images for block matching.
      ] and box-filtering [
      • Smith S.W.
      • et al.
      The Scientist and Engineer’s Guide to Digital Signal Processing.
      ,
      • McDonnell M.
      Box-filtering techniques.
      ]. Others propose variable-size windows [
      • Veksler O.
      Fast variable window for stereo correspondence using integral images.
      ], a perceptual grouping of color intensity levels [
      • Yoon K.-J.
      • Kweon I.-S.
      Locally adaptive support-weight approach for visual correspondence search.
      ], window displacement [
      • Scharstein D.
      • Szeliski R.
      A taxonomy and evaluation of dense two-frame stereo correspondence algorithms.
      ], multi-resolution images [
      • Laureano G.T.
      • de Paiva M.S.V.
      Disparities maps generation employing multi-resolution analysis and perceptual grouping.
      ], or image segmentation for disparity calculation [
      • Gerrits M.
      • Bekaert P.
      Local stereo matching with segmentation-based outlier rejection.
      ]. Although each method has particularities, some common elements can contribute to the coding, reusing, and comparing different depth estimation proposals.
      This work presents the Disparity Computation Framework, or DCF, designed to enable the coexistence of different disparity calculation proposals. DCF delimits the stereo vision scope, showing a solid base for attending different applications on-demand with a software design that allows attaching new stereo vision methods, algorithms, and metrics evaluation. Furthermore, DCF is a platform that allows the coexistence of different proposals for constructing disparity maps whose architecture integrates the main disparity calculation components discussed in the literature.
      In preparing the software design, we were concerned with designing modular software with a cohesive scope. In this sense, we raised the commonalities between different stereo vision methods and created structural software layers, including pre-processing, disparity calculation, post-processing, and performance evaluation modules. Also, in DCF design, all architecture components are available so that their components can be reused, new algorithms can be attached, and other structures can be prepared, including components to handle other computer vision tasks.
      DCF can raise research questions related to comparing different disparity calculation methods, error analysis between estimated and predicted disparities, the execution time of each algorithm, filtering input images, building disparity maps, and refinement of disparity calculation. Using DCF, researchers can perform comparative analyses, visually and numerically check the results, inspect the implemented code, and add new algorithms. Furthermore, they can perform disparity calculation methods with different test configurations using execution pipelines.
      Different execution pipelines can be configured with DCF. For example, an algorithm can be tested with different aggregation window sizes (Fig. 1). Other cost functions can easily replace a similarity evaluation function (Fig. 2). We can add pre-processing and post-processing steps to algorithms not originally designed with these steps (Fig. 3). We can apply image filtering and check the obtained results (Fig. 4), or add different refinement techniques to the disparity map (Fig. 5). Furthermore, we can compare different methods by looking at each error (Fig. 6).
      In the current version of DCF, there are ten disparity calculation methods, eleven cost functions, three evaluation metrics, and three disparity map refinement methods. Also, there are feature detectors, image filtering algorithms, and image rectification algorithms. Table 1 presents the disparity calculation methods, cost functions, and disparity map refinement methods already implemented in DCF.
      Figure thumbnail gr3
      Fig. 3Bilateral Stereo method (BL) with pre-processing and post-processing steps. From left to right, result without pre and post-processing steps, result with pre-processing (Guided Filter), result with post-processing (SCC), and result with both processing steps.
      Figure thumbnail gr4
      Fig. 4Guided Stereo method (GF) with different pre-processing filters.
      Figure thumbnail gr5
      Fig. 5Shiftable Window method (SW) with different post-processing approaches. From left to right, result without post-processing, result with the left-to-right consistency check (LR_check), result with the locally consistent method (LC), and result with the segmented consistency check (SCC).
      Figure thumbnail gr6
      Fig. 6Comparing the root mean square error (ϵ) of some disparity calculation methods.
      Finally, the algorithms present in the DCF can be parameterized or executed with previously defined configurations. The DCF calls are made through scripts where commands are presented sequentially, from the definition of parameters to function calls. In this sense, all the features present in the DCF can be accessed, parameterized, and executed in multiple configurations. The interfaces for accessing DCF functionality are public so that a single script can access the algorithms directly.

      2. Impact overview

      Previously published works related to the construction of disparity maps provide few implementation details, do not provide the source code, suggest processing steps that are not very well defined, and the software design for handling computer vision tasks is rarely discussed or adequately emphasized. Even when software is available, it is challenging to identify and decouple the steps of building disparity maps. Therefore, we developed the DCF to standardize the main components of disparity calculation and facilitate comparison between different stereo vision methods. DCF outputs can be directed to different applications, such as benchmarking proposals, computer and robotic applications, triangulation, and 3D reconstruction.
      DCF has been used to evaluate what cost function better fits each stereo-vision method [

      G. Vieira, F. Soares, N. Sousa, J. Gil, R. Parreira, G. Laureano, R. Costa, J. Ferreira, Stereo vision methods: from development to the evaluation of disparity maps, in: Proceedings of XIII Workshop de Visão Computacional, 2017, pp. 132–137.

      ], to design a refinement disparity method based on a segmentation process and adaptive support windows [
      • Vieira G.
      • Soares F.A.A.M.N.
      • Laureano G.T.
      • Parreira R.T.
      • Ferreira J.C.
      A segmented consistency check approach to disparity map refinement.
      ,
      • da Silva Vieira G.
      • Soares F.A.A.
      • Laureano G.T.
      • Parreira R.T.
      • Ferreira J.C.
      • Salvini R.
      Disparity map adjustment: a post-processing technique.
      ,
      • Vieira G.D.S.
      • Soares F.A.A.
      • Laureano G.T.
      • Parreira R.T.
      • Ferreira J.C.
      • Costa R.M.
      • Gonçalves C.
      Stereo matching enhancement by statistical analysis and weighted functions.
      ,
      • da Silva Vieira G.
      • Soares F.A.A.
      • Laureano G.T.
      • Parreira R.T.
      • Ferreira J.C.
      • Costa R.M.
      • Ferreira C.B.
      Disparity refinement through grouping areas and support weighted windows.
      ], and to construct disparity maps from unstructured environments where natural properties such as lighting and terrain shape provide various non-controlled conditions [
      • da Silva Vieira G.
      • Soares F.A.
      • de Lima J.C.
      • Laureano G.T.
      • Santos S.A.
      • Costa R.M.
      • Salvini R.
      Trunk detection and tree disparity calculation in uncontrolled environments.
      ]. In [
      • Vieira G.D.S.
      • Soares F.A.A.
      • De Lima J.C.
      • Do Nascimento H.A.
      • Laureano G.T.
      • Da Costa R.M.
      • Ferreira J.C.
      • Rodrigues W.G.
      A disparity computation framework.
      ], we presented the first insight into the DCF architecture to accommodate different stereo methods through a standard structure. Following that, we presented the modeling of the DCF architecture to integrate the main components of disparity calculation in [
      • da Silva Vieira G.
      • de Lima J.C.
      • de Sousa N.M.
      • Soares F.
      A three-layer architecture to support disparity map construction in stereo vision systems.
      ]. Unlike the previous version of DCF, the code has been revised, and different usage examples have been added in this version. In addition, we document the DCF source code to textually describe the functions’ goals, expected input parameters, and outputs.
      Table 1Functions implemented in DCF.
      Cost functionDisparity calculationDisparity refinement
      AFFBL
      • Yoon K.-J.
      • Kweon I.-S.
      Locally adaptive support-weight approach for visual correspondence search.
      LR_check
      • Hosni A.
      • Bleyer M.
      • Gelautz M.
      Secrets of adaptive support weight techniques for local stereo matching.
      BTSADBLNoSpatial
      • Hosni A.
      • Bleyer M.
      • Gelautz M.
      Secrets of adaptive support weight techniques for local stereo matching.
      LC
      • Mattoccia S.
      A locally global approach to stereo correspondence.
      BTSSDFW
      • Scharstein D.
      • Szeliski R.
      A taxonomy and evaluation of dense two-frame stereo correspondence algorithms.
      SSC
      • Vieira G.
      • Soares F.A.A.M.N.
      • Laureano G.T.
      • Parreira R.T.
      • Ferreira J.C.
      A segmented consistency check approach to disparity map refinement.
      LINGF
      • Hosni A.
      • Rhemann C.
      • Bleyer M.
      • Rother C.
      • Gelautz M.
      Fast cost-volume filtering for visual correspondence and beyond.
      NCCLO
      • Bobick A.F.
      • Intille S.S.
      Large occlusion stereo.
      SADMLMH
      • Cox I.J.
      • Hingorani S.L.
      • Rao S.B.
      • Maggs B.M.
      A maximum likelihood stereo algorithm.
      SSDMRPG
      • Laureano G.T.
      • de Paiva M.S.V.
      Disparities maps generation employing multi-resolution analysis and perceptual grouping.
      SSDNormSB
      • Gerrits M.
      • Bekaert P.
      Local stereo matching with segmentation-based outlier rejection.
      STADSW
      • Scharstein D.
      • Szeliski R.
      A taxonomy and evaluation of dense two-frame stereo correspondence algorithms.
      TAD_C+GVW
      • Veksler O.
      Fast variable window for stereo correspondence using integral images.
      ZSSD

      2.1 Ongoing research projects using the software

      A key component of stereo vision systems is the disparity map. If a disparity map is carefully constructed, it can attend industrial automation, autonomous navigation, and 3D reconstruction. In this sense, an autonomous vehicle can use the depth of a scene to move around without collisions; a visually impaired person can use a computer application developed to help him with his mobility; objects can be tracked in 3D, and augmented reality can bring better visual effects. We are currently using DCF to study stereo-vision algorithms and compare them visually and numerically. Furthermore, we use DCF to test processing steps like filtering and disparity map refinement.

      2.2 A list of all scholarly publications enabled by the software

      • 1.
        G. da Silva Vieira, J.C. de Lima, N.M. de Sousa, F. Soares, A three-layer architecture to support disparity map construction in stereo vision systems, Intell. Syst. Appl. 12 (2021) 200054.
      • 2.
        G. da Silva Vieira, F.A. Soares, J.C. de Lima, G.T. Laureano, S.A. Santos, R.M. Costa, R. Salvini, Trunk detection and tree disparity calculation in uncontrolled environments, in: 2019 IEEE Symposium on Computers and Communications, ISCC, IEEE, 2019, pp. 1–6.
      • 3.
        G.D.S. Vieira, F.A.A. Soares, J.C. De Lima, H.A. Do Nascimento, G.T. Laureano, R.M. Da Costa, J.C. Ferreira, W.G. Rodrigues, A disparity computation framework, in: 2019 IEEE 43rd Annual Computer Software and Applications Conference, Vol. 2, COMPSAC, IEEE, 2019, pp. 634–639.
      • 4.
        G. da Silva Vieira, F.A.A. Soares, G.T. Laureano, R.T. Parreira, J.C. Ferreira, R.M. Costa, C.B. Ferreira, Disparity refinement through grouping areas and support weighted windows, in: 2018 IEEE Canadian Conference on Electrical & Computer Engineering, CCECE, IEEE, 2018, pp. 1–4.
      • 5.
        G. da Silva Vieira, F.A.A. Soares, G.T. Laureano, R.T. Parreira, J.C. Ferreira, R. Salvini, Disparity map adjustment: a post-processing technique, in: 2018IEEE Symposium on Computers and Communications, ISCC, IEEE, 2018, pp.00580–00585.
      • 6.
        G. Vieira, F.A.A.M.N. Soares, G.T. Laureano, R.T. Parreira, J.C. Ferreira, A segmented consistency check approach to disparity map refinement, Can. J. Electr. Comput. Eng. 41 (4) (2018) 218–223.
      • 7.
        G.D.S. Vieira, F.A.A. Soares, G.T. Laureano, R.T. Parreira, J.C. Ferreira, R.M. Costa, C. Gonçalves, Stereo matching enhancement by statistical analysis and weighted functions, in: 2018 IEEE Canadian Conference on Electrical & Computer Engineering, CCECE, IEEE, 2018, pp. 1–4.
      • 8.
        G. Vieira, F. Soares, N. Sousa, J. Gil, R. Parreira, G. Laureano, R. Costa, J. Ferreira, Stereo vision methods: from development to the evaluation of disparity maps, in: Proceedings of XIII Workshop de Visão Computacional, 2017, pp.132–137

      Declaration of Competing Interest

      The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

      Acknowledgments

      We thank the Brazilian agency Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)–Finance Code 001– for the partial support to this work. We also acknowledge the partial support of the Universidade Federal de Goiás (Brazil) and Instituto Federal Goiano (Brazil).

      References

        • Furukawa Y.
        • Hernández C.
        Multi-view stereo: A tutorial.
        Found. Trends Comput. Graph. Vis. 2015; 9: 1-148
      1. T. Schöps, J.L. Schönberger, S. Galliani, T. Sattler, K. Schindler, M. Pollefeys, A. Geiger, A multi-view stereo benchmark with high-resolution images and multi-camera videos, in: Conference on Computer Vision and Pattern Recognition, Vol. 2017, CVPR, 2017, pp. 2538–2547.

      2. G. Vieira, F. Soares, R. Parreira, G. Laureano, R. Costa, Depth map production: approaches, challenges and applications, in: Proceedings of XII Workshop de Visão Computacional, 2016, pp. 323–328.

      3. G. Vieira, F. Soares, N. Sousa, J. Gil, R. Parreira, G. Laureano, R. Costa, J. Ferreira, Stereo vision methods: from development to the evaluation of disparity maps, in: Proceedings of XIII Workshop de Visão Computacional, 2017, pp. 132–137.

        • Huertas A.
        • Matthies L.
        • Rankin A.
        Stereo-based tree traversability analysis for autonomous off-road navigation.
        in: Application of Computer Vision, 2005. WACV/MOTIONS’05 Volume 1. Seventh IEEE Workshops on, Vol. 1. IEEE, 2005: 210-217
        • Roberts R.
        • Sinha S.N.
        • Szeliski R.
        • Steedly D.
        Structure from motion for scenes with large duplicate structures.
        in: Computer Vision and Pattern Recognition (CVPR), 2011 IEEE Conference on. IEEE, 2011: 3137-3144
        • Dong H.
        • Yin S.
        • Jiang G.
        • Liu L.
        • Wei S.
        An automatic depth map generation method by image classification.
        in: Consumer Electronics (ICCE), 2015 IEEE International Conference on. IEEE, 2015: 168-169
        • Hirschmuller H.
        • Scharstein D.
        Evaluation of cost functions for stereo matching.
        in: Computer Vision and Pattern Recognition, 2007. CVPR’07. IEEE Conference on. IEEE, 2007: 1-8
        • Scharstein D.
        • Szeliski R.
        A taxonomy and evaluation of dense two-frame stereo correspondence algorithms.
        Int. J. Comput. Vis. 2002; 47: 7-42
        • Vieira G.
        • Soares F.A.A.M.N.
        • Laureano G.T.
        • Parreira R.T.
        • Ferreira J.C.
        A segmented consistency check approach to disparity map refinement.
        Can. J. Electr. Comput. Eng. 2018; 41: 218-223
        • Mattoccia S.
        • Giardino S.
        • Gambini A.
        Accurate and efficient cost aggregation strategy for stereo correspondence based on approximated joint bilateral filtering.
        in: Asian Conference on Computer Vision. Springer, 2009: 371-380
        • Mattoccia S.
        A locally global approach to stereo correspondence.
        in: Computer Vision Workshops (ICCV Workshops), 2009 IEEE 12th International Conference on. IEEE, 2009: 1763-1770
      4. S. Mattoccia, Stereo vision algorithms for fpgas, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2013, pp. 636–641.

        • Hamzah R.A.
        • Ibrahim H.
        Literature survey on stereo vision disparity map algorithms.
        J. Sensors. 2016;
        • Lazaros N.
        • Sirakoulis G.C.
        • Gasteratos A.
        Review of stereo vision algorithms: from software to hardware.
        Int. J. Optomechatronics. 2008; 2: 435-462
        • Kumari D.
        • Kaur K.
        A survey on stereo matching techniques for 3D vision in image processing.
        Int. J. Eng. Manuf. 2016; 4: 40-49
        • Facciolo G.
        • Limare N.
        • Meinhardt-Llopis E.
        Integral images for block matching.
        in: Image Processing on Line, Vol. 4. 2014: 344-369
        • Smith S.W.
        • et al.
        The Scientist and Engineer’s Guide to Digital Signal Processing.
        California Technical Pub., San Diego1997
        • McDonnell M.
        Box-filtering techniques.
        Comput. Graph. Image Process. 1981; 17: 65-70
        • Veksler O.
        Fast variable window for stereo correspondence using integral images.
        in: Computer Vision and Pattern Recognition, 2003. Proceedings. 2003 IEEE Computer Society Conference on, Vol. 1. IEEE, 2003: I
        • Yoon K.-J.
        • Kweon I.-S.
        Locally adaptive support-weight approach for visual correspondence search.
        in: 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Vol. 2, CVPR’052005: 924-931https://doi.org/10.1109/CVPR.2005.218
        • Laureano G.T.
        • de Paiva M.S.V.
        Disparities maps generation employing multi-resolution analysis and perceptual grouping.
        in: 2008 First Workshops on Image Processing Theory, Tools and Applications. 2008: 1-6https://doi.org/10.1109/IPTA.2008.4743769
        • Gerrits M.
        • Bekaert P.
        Local stereo matching with segmentation-based outlier rejection.
        in: Computer and Robot Vision, 2006. the 3rd Canadian Conference on. IEEE, 2006: 66
        • Hosni A.
        • Bleyer M.
        • Gelautz M.
        Secrets of adaptive support weight techniques for local stereo matching.
        Comput. Vis. Image Underst. 2013; 117: 620-632
        • Hosni A.
        • Rhemann C.
        • Bleyer M.
        • Rother C.
        • Gelautz M.
        Fast cost-volume filtering for visual correspondence and beyond.
        IEEE Trans. Pattern Anal. Mach. Intell. 2013; 35: 504-511https://doi.org/10.1109/TPAMI.2012.156
        • Bobick A.F.
        • Intille S.S.
        Large occlusion stereo.
        Int. J. Comput. Vis. 1999; 33: 181-200
        • Cox I.J.
        • Hingorani S.L.
        • Rao S.B.
        • Maggs B.M.
        A maximum likelihood stereo algorithm.
        Comput. Vis. Image Underst. 1996; 63: 542-567
        • da Silva Vieira G.
        • Soares F.A.A.
        • Laureano G.T.
        • Parreira R.T.
        • Ferreira J.C.
        • Salvini R.
        Disparity map adjustment: a post-processing technique.
        in: 2018 IEEE Symposium on Computers and Communications, ISCC IEEE, 2018: 00580-00585
        • Vieira G.D.S.
        • Soares F.A.A.
        • Laureano G.T.
        • Parreira R.T.
        • Ferreira J.C.
        • Costa R.M.
        • Gonçalves C.
        Stereo matching enhancement by statistical analysis and weighted functions.
        in: 2018 IEEE Canadian Conference on Electrical & Computer Engineering, CCECE IEEE, 2018: 1-4
        • da Silva Vieira G.
        • Soares F.A.A.
        • Laureano G.T.
        • Parreira R.T.
        • Ferreira J.C.
        • Costa R.M.
        • Ferreira C.B.
        Disparity refinement through grouping areas and support weighted windows.
        in: 2018 IEEE Canadian Conference on Electrical & Computer Engineering, CCECE IEEE, 2018: 1-4
        • da Silva Vieira G.
        • Soares F.A.
        • de Lima J.C.
        • Laureano G.T.
        • Santos S.A.
        • Costa R.M.
        • Salvini R.
        Trunk detection and tree disparity calculation in uncontrolled environments.
        in: 2019 IEEE Symposium on Computers and Communications, ISCC IEEE, 2019: 1-6
        • Vieira G.D.S.
        • Soares F.A.A.
        • De Lima J.C.
        • Do Nascimento H.A.
        • Laureano G.T.
        • Da Costa R.M.
        • Ferreira J.C.
        • Rodrigues W.G.
        A disparity computation framework.
        in: 2019 IEEE 43rd Annual Computer Software and Applications Conference, Vol. 2, COMPSAC IEEE, 2019: 634-639
        • da Silva Vieira G.
        • de Lima J.C.
        • de Sousa N.M.
        • Soares F.
        A three-layer architecture to support disparity map construction in stereo vision systems.
        Intell. Syst. Appl. 2021; 12200054