• DocumentCode
    2583522
  • Title

    Occlusion detection in early vision

  • Author

    Toh, Peng-Seng ; Forrest, Andrew K.

  • Author_Institution
    Centre for Robotics & Autom. Syst., Imperial Coll., London, UK
  • fYear
    1990
  • fDate
    4-7 Dec 1990
  • Firstpage
    126
  • Lastpage
    132
  • Abstract
    The authors show that occlusion can be detected by early visual processes derived directly from the image data. Two methods are demonstrated. The first method is based on detecting binocular rivalry during binocular fixation. The feasibility of this technique also supports the hypothesis that stereo fusion and rivalry coexist in human vision. Stereo fusion gives rise to smooth disparity whereas rivalry signals depth discontinuity. The second method uses focal information that results from a change in depth of field. This method does not involve changes of viewpoint and hence avoids the correspondence problem. These two methods together with cues from motion, as has been suggested elsewhere, form robust occlusion detection in early visual processing
  • Keywords
    computer vision; computerised picture processing; binocular fixation; binocular rivalry; correspondence problem; depth discontinuity; image data; occlusion detection; stereo fusion; Educational institutions; Humans; Interpolation; Layout; Machine vision; Motion detection; Object recognition; Robotics and automation; Robustness; Surface reconstruction;
  • fLanguage
    English
  • Publisher
    ieee
  • Conference_Titel
    Computer Vision, 1990. Proceedings, Third International Conference on
  • Conference_Location
    Osaka
  • Print_ISBN
    0-8186-2057-9
  • Type

    conf

  • DOI
    10.1109/ICCV.1990.139509
  • Filename
    139509