This page has been created in order to distribute a prototype software
implementing the Spatially Coherent Self-Organizing Background Subtraction
(SC-SOBS) algorithm presented in [3]:
L. Maddalena, A. Petrosino, The SOBS
Algorithm: What Are the Limits?, 2012 IEEE Computer Society Conference
on Computer Vision and Pattern Recognition Workshops (CVPRW), DOI:
10.1109/CVPRW.2012.6238922, pp.21-26, 16-21 June 2012
Click here to download the Windows
executable (WinZip compressed) together with the needed OpenCV .dll's.
Please,
observe that, compared to the previous version, it includes a bug fix (for
properly handling also image sequences with "unusual" resolution,
e.g., 238x158) and it has a slightly different syntax for specifying the input
parameters (see "Usage with generic image sequences").
If you have
problems downloading, please contact lucia.maddalena "at" cnr.it; if
you use the software, please cite the above mentioned paper.
Basic usage for Change Detection
Competition:
Unzip the SC-SOBS.exe and the .dll's into the directory holding the
"dataset" directory containing the Change Detection Challenge
sequences [1]. To obtain all masks for the
"baseline/highway" sequence, just type:
SC-SOBS dataset/baseline/highway/input/in
Click here to download the masks computed by SC-SOBS for
the whole dataset.
Usage with generic image sequences:
SC-SOBS <SeqName> [Parameters]
where
o <SeqName>: sequence
name (complete path), not including frame numbers. Image sequences consist
of .jpg image frames with
consecutive 6 digit numbers, named in the following form
<SeqName><number>.jpg
o [parameters]: optional, including:
-nini #:
number of first sequence frame to be considered.
-nend #:
number of last sequence frame to be considered.
-n #: (square root of) number of weight vectors
for each pixel. Default: 3
-K #: Number of initial frames for training.
Default: fromIdx-1 (fromIdx read from file ‘temporalROI.txt’ as in [1])
-e1 #: Distance threshold e1 for training phase (Eq.
(12)). Default: 1.0
-e2 #: Distance threshold e2 for testing phase (Eq.
(12)). Default: 0.008
-c1 #: Learning rate c1 for training phase
(Eq. (14)). Default: 1.0
-c2 #: Learning rate c2 for testing phase
(Eq. (14)). Default: 0.05
-Cw #:
Size of the neighbourhood for Spatial Coherence (Eq. (10)). Default: 5
-s
#: To apply shadow removal (as in [2]).
Default: 1 (apply)
-g #: Shadow detection value for g in Eq. (5) in [2]. Default: 0.7
-b #: Shadow detection value for b in Eq. (5) in [2]. Default: 1.0
-tS #:
Shadow detection value for tS
in Eq. (5) in [2]. Default: 0.1
-tH #: Shadow detection value for tH in Eq. (5) in [2]. Default: 10.0
-ROI #:
To use ROI.bmp mask as in [1]. Default: 1 (do use)
-Med #:
Size of the neighbourhood for Median Filtering Post-Processing (through OpenCV
function cvSmooth). Default: 1
-m #: To save background model images. Default: 0 (do
not save
-l
#: To save only last detection mask. Default 0 (save all in the
temporal ROI)
1) SC-SOBS
Provides the above information on usage.
2) SC-SOBS c:/Sequences/WavingTrees/WavingTrees -nini 1000 -nend 1247 -K
200 -e1 0.1 -e2 0.03 -c1 1.0 -c2 0.05 –l
where sequence WavingTrees, coming
from sequences adopted in K. Toyama, J. Krumm, B. Brumitt, and
B. Meyers, “Wallflower: principles and practice of background maintenance,” in Proc.
7th IEEE Conf. Computer Vision, 1999, vol. 1, pp. 255–261, has been saved in .jpg image files named:
WavingTrees001000.jpg, …,
WavingTrees001247.jpg
and stored in
directory c:/Sequences/WavingTrees.
This gives the moving
object detection mask for last frame (named bin001247.png) as well the background
model (named Model001199.ppm) achieved by training on the first 200 frames and
the updated background model (named Model001247.ppm) for the last frame.
3) SC-SOBS c:/Sequences/WavingTrees/WavingTrees -nini 1000 –nend 1247 -K
200 -e1 0.1 -e2 0.03 -c1 1.0 -c2 0.05 –l
same as before, but
without applying median filtering post-processing.
References:
[1] http://www.changedetection.net/
[2] L.
Maddalena, A. Petrosino, A Self-Organizing Approach to
Background Subtraction for Visual Surveillance Applications, IEEE Transactions on
Image Processing, DOI 10.1109/TIP.2008.924285, Vol. 17, no. 7, pagg. 1168-1177,
July 2008.
[3] L.
Maddalena, A. Petrosino, The SOBS Algorithm: What Are the
Limits?, 2012 IEEE Computer Society Conference on Computer Vision and
Pattern Recognition Workshops (CVPRW), DOI: 10.1109/CVPRW.2012.6238922, pp.21-26,
16-21 June 2012
Last update: November 5,
2015.