This page has been created in order to distribute a prototype software
implementing 3dSOBS+, an enhanced version of the 3D Self-Organizing Background Subtraction algorithm [1],
presented in [2]
L. Maddalena, A. Petrosino, The 3dSOBS+ Algorithm for Moving Object Detection, Computer Vision and Image
Understanding, DOI: 10.1016/j.cviu.2013.11.006, Vol. 122, pp. 65-73, 2014.
Click here to download the Windows
executable (WinZip compressed) together with the needed OpenCV .dll's.
If you have problems downloading, please contact lucia.maddalena "at"
cnr.it; if you use the software, please cite the paper [2].
Usage:
3dSOBSplus <SeqName> [Parameters]
where
o <SeqName>: sequence
name (complete path), not including frame numbers. Image sequences consist
of .png image frames with
consecutive numbers, named in the form
o [parameters]: optional, including:
-nini #:
Number
of first sequence frame to be considered.
-nend #:
Number
of last sequence frame to be considered.
-n #: Number
of model layers. Default: 5
-F #: Number
of initial frames for model initialization. Default: 100
-e #: Segmentation
threshold e in Eq. (10). Default:
0.005
-w2d #: Halfwidth
of 2D neighborhood for model update in Eq. (4). Default 1
-w1d #: Halfwidth
of 1D neighborhood for model update in Eq. (6). Default 1
-lr
#: Learning
rate g=n in Eqs. (5) and (7).
Default 0.05
-Boot
#: Read from file the initial background
(1) or
compute it through temporal median
(0). Default 0
-SmaskMOD: To save the MOD masks. Default: save
-SmodBG: To save the BG models. Default: do
not save
1) 3dSOBSplus
Provides the above information on usage.
2) 3dSOBSplus C:\Sequenze\BMC2012\111_png\input -nini 1 -nend 450
where sequence 111_png\input,
coming from the Background Models Challenge (BMC2012) dataset
[3]
,
consists of .png image files named:
1.png, …, 1499.png
and stored in
directory C:\Sequenze\BMC2012.
This gives the moving
object detection mask for the first 450 frames (named bin000001.png, …,
bin000450.png) as well as the initial model (named InitialModel.png and stored
in the same directory of the input sequence) achieved by temporal median on the
first 100 frames, adopted for the neural model initialization.
3) 3dSOBSplus C:\Sequenze\BMC2012\111_png\input\ -ini 1 -end 450 -SmodBG
same as before, but saving
in the current directory a representation of the background model for each frame
(-SmodBG). For each frame, the representation is built as an image where each
pixel is the neural model weight vector that is closest to the corresponding
pixel of the frame.
4) 3dSOBSplus C:\Sequenze\BMC2012\111_png\input\ -ini 1 -end 450
-ShowMask 1 -Boot 1
same as before, but
showing the foreground masks (-ShowMask 1) and using the already computed
initial model InitialModel.png, stored in the same directory of the input
sequence (-Boot 1).
References:
[1] L. Maddalena, A. Petrosino, Stopped Object Detection by Learning Foreground Model in Videos, IEEE Transactions on Neural Networks and Learning Systems, DOI: 10.1109/TNNLS.2013.2242092, vol.24, no.5, pp.723-735, May 2013.
[2] L. Maddalena, A. Petrosino, The 3dSOBS+ Algorithm for Moving Object Detection, Computer Vision and Image Understanding, DOI: 10.1016/j.cviu.2013.11.006, Vol. 122, pp. 65-73, 2014.
[3] http://bmc.iut-auvergne.com/
Last update: November 5,
2015.