International Core Journal of Engineering 2020-26 | Page 99

The architecture of this paper is arranged as follows:
Some related works are briefly reviewed in Section 2. Then,
the proposed method based on the combination of Gabor
filter and Lab features is introduced in Section 3. The
experiment results are shown in Section 4. Finally,
conclusions are given in Section 5.
C. Detection based on optical flow
Methods based on optical flow make use of the optical
flow field to abstract the moving targets [5-6]. In optical flow
method, the motion vector of each pixel is established, and
then the optical flow field of the whole image is established.
If there is no moving target, the motion vectors of all pixels
in the image should be continuously changed. Otherwise, due
to the relative motion between the target and the background,
the motion vector at the position of the target must be
different from that of the background. Then the moving
target will be detected.
II. R ELATED WORK
A. Inter-frame difference detection based on texture for
moving target
Texture features [13] are different from color features.
Color features are based on the features of pixel points.
Texture features are global features, which need to be
statistically calculated in the region containing multiple pixel
points. It reflects the visual characteristics of the
homogeneity phenomenon in the image. Besides, it reflects
the organization and arrangement properties of the surface
structure with slow change or periodic change. Texture
features are resistant to noise and often have rotational
invariance.
While in the actual situation, because of the influence of
light and other factors, the surface brightness of the target
does not remain constant. It does not meet the premise of the
basic constraint equation of optical flow, resulting in an error
in the calculation. What’s more, the method is computation
and time consuming. This method cannot deal with sudden
drastic lighting changes and its initialization is important.
The optical flow method is rarely used in practice due to
these drawbacks.
Gabor is a filter that extracts texture features from images
[14]. In [15], it is used to binarize and normalize the frame.
Gabor filter approximate the receptive field cells (the transfer
function under light intensity stimulation) of a single cell.
Also, Gabor wavelet [16] is sensitive to the edge of the
image and provides a good direction selection and scale
selection characteristics. Moreover, it is not sensitive to light
changes, so it provides good adaptability to light changes.
D. Detection based on background subtraction
Background subtraction method is similar to inter-frame
difference method. They both use the image difference
algorithm to extract the target area, while the background
subtraction method subtracts the current frame image from a
constantly updated background model [7]. It includes four
steps: background modeling, background update, target
detection, and post processing, as shown in Fig. 2.
Background modeling and background updating are the two
core problems in background subtraction. The commonly
used methods in the background modeling step are GMM [7]
and Vibe [9-10].
In fact, the impulse response of a Gabor filter can be
defined as a sinusoidal wave which is multiplied by a
Gaussian function [17]. In the spatial domain, a two-
dimensional Gabor filter is actually a Gaussian kernel
function modulated by a sinusoidal plane wave. The Gabor
filter can be expressed as:
,
Fig. 2. The background subtraction algorithm flow chat.
where
GMM assumes that each pixel obeys the normal
distribution in the time domain. The pixel is judged as the
background within a certain threshold, while the pixel that
does not conform to the distribution is judged as the
foreground. Vibe is mainly characterized by random
background update strategy which is a kind of non-
parametric estimation. It is based on the assumption that
when the model of pixel change cannot be determined, the
random model is more suitable to simulate the uncertainty of
the pixel change. Both method have advantages and
shortcomings. There still exist inaccurate detection problems
under the conditions, such as dynamic background, light
change, and camouflage phenomenon [11].
and
.
In this equation, λ represents the wavelength of the
sinusoidal factor, θ represents the orientation of the normal to
the parallel stripes of a Gabor function, ψ is the phase offset,
σ is the sigma/standard deviation of the Gaussian envelope,
and γ is the spatial aspect ratio which specifies the ellipticity
of the support of the Gabor function. Fig. 3 shows the
construction of Gabor Filter Array. It visualizes wavelength,
orientation and the real part of the spatial convolution kernel
of each Gabor filter in the array. Therefore, in order to
realize the texture based difference method, Gabor filter is
applied to the inter-frame difference method in the proposed
method.
B. Inter-frame difference detection based on color for
moving targets
The traditional inter-frame difference method may lose
some information by converting the color images into gray-
scale images. So, the proposed method takes color features
into account. It applies the inter-frame difference method to
calculate different components of colors respectively [18].
The contribution of this paper is to dispose the problems
of the dynamic background and light change. It is proposed
that a new method for moving target detection uses the
combination of texture based inter-frame difference method
and color based inter-frame difference method. The Gabor
filter is used to extract the image texture features, and the
color features are extracted in the Lab color space [12]. In
this way, the proposed method has certain adaptability to the
dynamic background and illumination changes in the
detection of moving targets.
Lab color space is designed to approximate human vision
[12], unlike the RGB and CMYK color models. It describes
all the colors visible to the human eye and is created to serve
77