Full video results from GOSUS

Overview

  • For the results of background subtraction: each frame consists of six sub-images: top left shows the original image; top middle shows the background estimated by GOSUS; top right shows the foreground segmentation by GOSUS; bottom left shows the foreground segmentation by RPCA; bottom middle shows the foreground segmentation by RPMF, bottom right shows the result by GRASTA. The foreground thresholds for each algorithm are selected by cross-validation. This dataset is from Perception Test Images Sequences [1] and Wallflower Test Images Sequences [2].
  • For the results of multiple face tracking: faces marked with the same number are from the same track; frame number is shown on the left top corner.

    • Background Subtraction

    WavingTrees (Download video 1.2 MB)


    WaterSurface (Download video 4MB)


    Campus (Download video 2.6 MB)


    Bootstrap (Download video 5.1 MB)


    Escalator (Download video 5.7 MB)


    ShoppingMall (Download video 3.7 MB)


    Lobby (Download video 2.8 MB)


    Curtain (Download video 5.0 MB)


    Fountain (Download video 1.1 MB)


    TimeOfDay (Download video 9.5 MB)


    LightSwitch (Download video 4.6 MB)


    Multiple Face Tracking


    The Big Bang Theory - The Bad Fish Paradigm (Download video 22 MB)


    The Big Bang Theory - Leonard's Secret Weapon (Download video 22 MB)

    Reference

  • [1] L. Li, W. Huang, I. Y. H. Gu, and Q. Tian. Statistical modeling of complex backgrounds for foreground object detection. TIP, 13(11):1459–1472, 2004.
  • [2] K. Toyama, J. Krumm, B. Brumitt, and B. Meyers. Wallflower: Principles and practice of background maintenance. In ICCV, 1999.