Digital Signal Processing Reference
In-Depth Information
applications; frame difference method is relatively simple, real-time and has better
adaptability to the dynamic environment, but can not completely extract all the
relevant points of the target. Compare to the frame difference method, the background
difference method can get more complete target, and its computation complexity is
much lower than the optical flow computation, so the background difference method
has been widely used.
2
Background Subtraction Method
Background subtraction [3-4] is simply using the current image subtracting the
background image, select a suitable threshold for binarization of the differential
image, you can accurately identify the location of moving targets. Its outstanding
features are simple, fast and having good detection results, so it can be used in the
occasion with high real-time requirement.
Suppose video image sequence at time t is I(x,y,t), the background image
simulated by model is B(x,y,t), to do the subtraction of two images, we can get the
difference image D(x,y,t). As the formula (1).
D(x
,
y
,
t)
=
|
I(x
,
y
,
t)
-
B(x
,
y
,
t)
|
(1)
Then passed the threshold processing to get the two binary image BW(x,y,t).
1
if(D(x,y,t
>
)
T
)
BW
(
x
,
y
,
t
)
=
(2)
0
otherwise
In formula (2), if the difference of the current frame and the background is greater
than the threshold T, the binarized image is set to 1, indicating that the pixel in the
movement area;otherwise the binary image is set to 0, indicating that the pixel is part
of the background.
Although moving target detection based on the background difference method can
get more complete extraction and moving object related informations than frame
difference method, but with the passage of time, the scene may appear light change,
wave reflection ,shaking trees and other external conditions changes , there may be
some false movement, affect the effect of moving target detection.
3
Adaptive Mixture Gaussian Background Modeling
Mixture Gauss background model [5-6] is firstly proposed by Staufer and Grimson.
This method uses Gaussian mixture modeling for each pixel, and uses pixel iterative
for model parameters updateso as to effectively overcome the disturbance caused
by the background image change, light gradient. After extracting and updating the
background, the moving target can be very easy get by deducting the background.
