![](bg1.png)
REAL-TIME MOVING OBJECT DETECTION IN VIDEO
SEQUENCES USING SPATIO-TEMPORAL ADAPTIVE GAUSSIAN
MIXTURE MODELS
Katharina Quast, Matthias Obermann and Andr´e Kaup
Multimedia Communications and Signal Processing, University of Erlangen-Nuremberg
Cauerstr. 7, 91058 Erlangen, Germany
Keywords:
Object detection, Background modeling.
Abstract:
In this paper we present a background subtraction method for moving object detection based on Gaussian
mixture models which performs in real-time. Our method improves the traditional Gaussian mixture model
(GMM) technique in several ways. It takes into account spatial and temporal dependencies, as well as a
limitation of the standard deviation leading to a faster update of the model and a smoother object mask. A
shadow detection method which is able to remove the umbra as well as the penumbra in one single processing
step is further used to get a mask that fits the object outline even better. Using the computational power of
parallel computing we further speed up the object detection process.
1 INTRODUCTION
The detection of moving objects in video sequences is
an important and challenging task in multimedia tech-
nologies. Most detection methods follow the princi-
ple of background subtraction. To segment moving
foreground objects from the background a pure back-
ground image has to be estimated. This reference
background image is then subtracted from each frame
and binary masks with the moving foreground objects
are obtained by thresholding the resulting difference
images.
In (Stauffer and Grimson, 1999; Power and
Schoonees, 2002) the values of a particular pixel over
time are modeled as a mixture of Gaussian distri-
butions. Thus, the background can be modeled by
a Gaussian mixture model (GMM). Once the pixel-
wise GMM likelihood is obtained, the final binary
mask is either generated by thresholding (Stauffer and
Grimson, 1999; Power and Schoonees, 2002; Kaew-
TraKulPong and Bowden, 2001) or according to more
sophisticated decision rules (Carminati and Benois-
Pineau, 2005; Li et al., 2004; Yang and Hsu, 2006).
Although the Gaussian mixture model technique is
quite successful the obtained binary masks are often
noisy and irregular. A main reason for this is that spa-
tial and temporal dependencies are neglected in most
approaches. In (Li et al., 2004) a Bayesian frame-
work for object detection is proposed that incorpo-
rates spectral, spatial, and temporal features. But the
spatial dependency is only deployed during post pro-
cessing mainly by applying morphological operations
which leads to poor object contours.
We improvethe standard GMM method by regard-
ing spatial and temporal dependencies and integrating
a limitation of the standard deviation into the tradi-
tional method. Combining this improved method with
our fast shadow removal technique, which is inspired
by the technique of (Porikli and Tuzel, 2003), leads to
good binary masks without adding any complex and
computational expensive extensions to the method.
Thus, better masks are obtained while the computa-
tional speed of the standard GMM method is kept and
further post processing can be omitted. Through par-
allelization of the algorithm we even achieve an enor-
mous performance speedup.
In the follwing, an overview of the GMM method
is given in Section 2. In Section 3 the proposed
method is first described explaining the use of spatial
and temporal dependencies, the limitation of the stan-
dard deviation, and the shadow removal technique.
Experimental results and implementation issues are
discussed in Section 4. Finally conclusions are drawn
in Section 5.
413
Quast K., Obermann M. and Kaup A. (2010).
REAL-TIME MOVING OBJECT DETECTION IN VIDEO SEQUENCES USING SPATIO-TEMPORAL ADAPTIVE GAUSSIAN MIXTURE MODELS.
In Proceedings of the International Conference on Computer Vision Theory and Applications, pages 413-418
DOI: 10.5220/0002816904130418
Copyright
c
SciTePress