4.2 Image Processing in Multi-tag Movement
117
Table 4.3 Image test results of round RFID tags
Figure Time/s Feature points Number of matches Matching rate/%
SURF
8
1.760
5
4
80.0
Improved SURF 8
0.764
8
7
87.5
Table 4.4 Comparison of tag
image test results of different
algorithms
Average time/s
Average matching
rate/%
SURF
2.027
43.6
Literature [2]
algorithm
1.183
80.2
Improved SURF 1.056
87.4
position data of the tag. However, in the actual simulation process, relative motion
between tag and camera would affect the position data collected by the reader, so
that the system cannot accurately obtain the position of the tag, and then affect the
coordinate distribution of the subsequent acquisition of RFID multi-tag. In order to
ensure the smooth progress of the follow-up research, the tag position data must
be collected by a method that can remove the generated motion blur. Based on
the system designed in the previous article, regarding the possibility of multiple
distribution states of tags, this paper proposed an algorithm that can meet the needs
of use and provides an effective data guarantee for obtaining the three-dimensional
coordinates of RFID tags.
4.2.1 Image Deblur Theory
When shooting a moving target, due to the relative motion between the subject and
the camera, the acquired image is degraded, and such a blurred image is a motionblurred image. In many practical applications, motion blur needs to be removed.
Under normal circumstances, the process of motion degradation can be modeled as
a two-dimensional linear displacement invariant process. In this process, the blurred
image g(x, y) can be expressed as the convolution of the original image f (x, y)
points and the diffusion function (PSF) [87]; therefore, the restoration process from
the blurred image is actually the operation of deconvolution. The degradation model
is shown in Fig. 4.8.
The mathematical expression of the model is:
g(x, y) = f (x, y) ∗ h(x, y) + η(x, y)
(4.9)
In Eq. (4.9), * represents a two-dimensional linear convolution, η(x, y) is additive
noise, and h(x, y) represents a point spread function. The expression of the model in
117
Table 4.3 Image test results of round RFID tags
Figure Time/s Feature points Number of matches Matching rate/%
SURF
8
1.760
5
4
80.0
Improved SURF 8
0.764
8
7
87.5
Table 4.4 Comparison of tag
image test results of different
algorithms
Average time/s
Average matching
rate/%
SURF
2.027
43.6
Literature [2]
algorithm
1.183
80.2
Improved SURF 1.056
87.4
position data of the tag. However, in the actual simulation process, relative motion
between tag and camera would affect the position data collected by the reader, so
that the system cannot accurately obtain the position of the tag, and then affect the
coordinate distribution of the subsequent acquisition of RFID multi-tag. In order to
ensure the smooth progress of the follow-up research, the tag position data must
be collected by a method that can remove the generated motion blur. Based on
the system designed in the previous article, regarding the possibility of multiple
distribution states of tags, this paper proposed an algorithm that can meet the needs
of use and provides an effective data guarantee for obtaining the three-dimensional
coordinates of RFID tags.
4.2.1 Image Deblur Theory
When shooting a moving target, due to the relative motion between the subject and
the camera, the acquired image is degraded, and such a blurred image is a motionblurred image. In many practical applications, motion blur needs to be removed.
Under normal circumstances, the process of motion degradation can be modeled as
a two-dimensional linear displacement invariant process. In this process, the blurred
image g(x, y) can be expressed as the convolution of the original image f (x, y)
points and the diffusion function (PSF) [87]; therefore, the restoration process from
the blurred image is actually the operation of deconvolution. The degradation model
is shown in Fig. 4.8.
The mathematical expression of the model is:
g(x, y) = f (x, y) ∗ h(x, y) + η(x, y)
(4.9)
In Eq. (4.9), * represents a two-dimensional linear convolution, η(x, y) is additive
noise, and h(x, y) represents a point spread function. The expression of the model in