Research on blurred image restoration algorithm based on best iteration for vision guidance vehicle

Abstract

In order to solve the problem of blurred image when the visual guidance vehicle used to locate a two-dimensional code to acquire images, a blurred image restoration algorithm based on the optimal number of iterations for visual guidance vehicles was proposed. Firstly, the process of image blurring was modeled according to the above method. The Lucy-Richardson algorithm based on the optimal number of iterations was designed and the experimental results were verified by the independently developed visual guidance vehicle. The experimental results show that the algorithm based on best iteration number of visual guidance vehicle blurred image restoration algorithm is superior to the traditional Wiener filter algorithm in suppressing noise, improving the quality of the restored image and enhancing the fastness, the best number of restores is 20 times, it improves the speed of correcting and control precision of the visual guide vehicle, the vision tracking vehicle tracking accuracy is stabilized at 3.5 mm or less.

Keywords

Blurred image restoration Lucy-Richardson algorithm optimal iteration times visual guidance

1. Introduction

With the rapid development of modern logistics and transportation industry, more and more logistics equipment has also emerged and developed. As an unmanned intelligent logistics equipment, the visual guidance vehicle has been widely applied to all walks of life in the national economy involving logistics and transportation. It is a comprehensive intelligent vehicle that integrates science and technology such as mechanics, computer technology, control theory, intelligent technology, and electronics. Since its inception, the visual guidance vehicle has attracted the attention and research of many scholars at home and abroad because of its highly intelligent and flexible characteristics. This article studies on visual guidance vehicles, visual guidance technology is used in the visual guidance vehicle to control operation, it has a higher information dimension and guidance flexibility which are more and more widely used in engineering practice. The visual fuzzy problem appears during the process of vision guided vehicle, it will affect the running accuracy and stability of the vehicle, it is one of the basic problems of visual guidance and also the core issue of vehicle self-guidance technology. Therefore, it is necessary to restore the image with motion blurred images [17].

Wiener filter algorithm [19] is often used to reconstruct the motion blurred image. However, the visual guidance vehicle has the characteristics of time-varying uncertainty as the control object. It is difficult to apply the Wiener filter algorithm to reconstruct the image accuracy. The L-R algorithm [16, 18, 20]. recovers blurred images based on the number of iterations, for time-varying visual guidance vehicles with high uncertainty, blurred image recovery time is difficult to control, the reason is: when the L-R algorithm [5, 8, 7] recovers an image, the restoration effect is significantly improved with the increase in the number of iterations, and the image noise is amplified and the loss time is increased. Therefore, how to balance the degree of image restoration [1, 10], suppress noise and reduce the recovery time becomes the key to the restoration of blurred images. Therefore, in order to meet the requirements of engineering applications, many scholars are currently studying how to improve the recovery effect of blurred images [13]. Xiao and others restored the image through the improved Wiener filtering motion blurr; Wang et al. restored blurred images by L-R algorithm; Chen et al. rotated the two-dimensional blurred image to the horizontal axis to perform a first-order differential to obtain the discriminant curve. The fuzzy kernel function was obtained by discriminating the distance of the conjugate correlation peak on the curve, and the Wiener filter was used for image restoration [6].

L-R algorithm restores the image [14] based on the number of iterations, and the effect of image restoration varies greatly between different iterations; however, the visual guidance car is moving at a constant speed, so the degree of blurring of the blurred image [2] is certain, Therefore, experiments can be used to determine the optimal number of iterations for reconstructing blurred images of visual guided vehicles. Based on the optimal iteration number, the blurred image restoration algorithm of visual guidance vehicle is improved on the basis of the conventional L-R algorithm. The optimal iteration times are used to optimize the restoration effect of the blurred image and restrain the noise balance of the blurred image restoration time, which is simple, easy to implement, and good control effect. Based on this, this paper proposes a blurred image restoration algorithm based on the optimal number of iterations to improve the control accuracy of visual guidance vehicles and enhance the stability of vehicle operation.

2. Visual guidance principle

The visual guidance vehicle dynamically acquires the two-dimensional code image through the CCD (Charge Coupled Device) camera installed in the vehicle. The two-dimensional code image captured by the camera is a perspective projection of a three-dimensional space in a two-dimensional space. In the process of recognizing picture information by machine vision, a process of inversion is required to obtain the yaw angle and offset during the running of the vehicle. After a specific algorithm, we can be obtain the speed, position, and heading information of the visual guidance vehicles at different time.

The visual guidance vehicle uses the differential correction control strategy. First, the visual guidance vehicle acquires the two-dimensional code image as the input of the best iteration number LR algorithm to obtain a clear two-dimensional code image. According to the laying position of the two-dimensional code, the visual guidance vehicle real-time speed, position and heading information are calculated. This information is used as a feedback signal to compare with the input signal, then output the deviation change amount, and this value is transmitted to the motion guidance system of the visual guidance vehicle to control the stable operation of the vehicle.

3. Establishing vision guidance vehicle blurred image recovery model

3.1 Fuzzy kernel function design

Image restoration technology is to model the process of image blurring. This process is equivalent to blurring the image through a degenerate model with a clear image. The original image can be recovered by taking the opposite process on the premise that the degradation model is known. The degradation process of the image is usually described as the convolution of the original image with the point spread function, the image degradation process is shown in Eq. (1).

$\displaystyle f(x,y)\to H[\ ]\to\begin{array}[]{c}n(x,y)\\ \downarrow\\ \bigoplus\\ \uparrow\\ g(x,y)\\ \end{array}$ (1)

Among them, $g(x,y)$ represents the blurred image, $f(x,y)$ represents the original image, $n(x,y)$ represents additive noise, $H[\ ]$ represents the point spread function [4], the mathematical expression of the degenerate model is:

$\displaystyle g(x,y)=H[f(x,y)]+n(x,y)$ (2)

Let $X{\_}0(t)$ and $Y{\_}0(t)$ represent the motion blur components of the trolley in the $x$ and $y$ directions respectively, integrating Eq. (2) within the exposure time $T$ results in:

$\displaystyle g(x,y)=\int_{0}^{T}f(x-X_{0}(t),y-Y_{0}(t))dt+n(x,y)$ (3)

Fourier transform for Eq. (3) and change the integration order, the final variable is:

$\displaystyle G(\omega,\mu)=F(\omega,\mu)\int_{0}^{T}e^{-j2\pi(\omega X_{0}(t)% +\mu Y_{0}(t))}dt+N(\mu+\omega)\int_{0}^{T}e^{-j2\pi(\omega X_{0}(t)+\mu Y_{0}% (t))}dt$ (4)

Then $\int_{0}^{T}e^{-j2\pi(\omega X_{0}(t)+\mu Y_{0}(t))}dt$ is the required fuzzy kernel function $H(\omega,\mu)$ . The guided vehicle travels at a constant speed, in the exposure time $T$ , the blur length is $L$ , ambiguity direction is $\beta$ , travel time is $t$ , therefore, the fuzzy kernel function [12] can be expressed as:

$\displaystyle H(\omega,\mu)=\int_{0}^{T}e^{-j2\pi(\omega\frac{L\cos\beta t}{T}% +\mu\frac{L\sin\beta t}{T})}dt$ (5)

$L$ is the fuzzy length, and it is the moving distance of the visual guidance vehicle during the exposure time, for the reason that the visual guidance vehicle is a uniform motion $L=vt$ ; $\beta$ is the fuzzy angle, the blur angle is the angle between the actual movement direction of the visual guidance vehicle and the planning direction, the actual direction of motion can be calculated by the position information stored in the two-dimensional code to obtain $\beta$ .

3.2 L-R algorithm to restore image model

For nonlinear restoration of images, the L-R iterative restoration algorithm is more accurate than the Wiener filter algorithm, so this algorithm is chosen. The L-R algorithm assumes that the image noise obeys the Poisson distribution, using the maximum likelihood method for estimation, assuming that the value of the probability density function is the largest, it can be derived that the iterative expression is:

$\displaystyle f(x,y)^{(k+1)}=f(x,y)^{k}\times\left\{F^{-1}[H(\omega,\mu)]% \oplus\frac{g(x,y)}{F^{-1}[H(\omega,\mu)]\otimes f(x,y)^{(k)}}\right\}$ (6)

There $\oplus$ denotes the correlation operation, $\otimes$ denotes the convolution operation, and $k$ is the number of iterations [3]. You can make $f(x,y)^{(0)}=g(x,y)$ , when noise is negligible, As $k$ increases, $f(x,y)^{(1+k)}$ converges on the probability of $f(x,y)$ , as a result, we achieve image restoration. However, if there is noise, the L-R iterative process will amplify the noise [10], and use the iterative residual method to substitute $r^{-(k)}$ into Eq. (6):

$\displaystyle f(x,y^{(k+1)}=f(x,y)^{k}\times\left\{F^{-1}[H(\omega,\mu)]\oplus% \frac{g(x,y)+r^{-(k)}(x,y)}{F^{-1}[H(\omega,\mu)]\otimes(x,y)^{(k)}}\right\}$ (7)

When the noise $r^{-(k)}(x,y)$ is not negligible, Eq. (7) will no longer have convergence, so the L-R algorithm has the problem of noise amplification and will affect the effect of the restored image.

4. Experiment and result verification

The verification of this conclusion is based on an independently developed visual guidance vehicle. As shown in Fig. 1, two-dimensional codes including coordinate information, path information, and station information are laid every 2 meters. According to the analysis of fuzzy length and fuzzy angle [9], the blurred image restoration experiment is designed. According to the experimental data, such as the running speed of the guided vehicle, the maximum blur degree, etc., the number of iterations of the image restoration is set to 10, 20, 30, 40, and 50 times, and the results were analyzed.

Figure 1.

Visual guidance car.

4.1 Analysis of experimental results of motion blur restoration without noise mode

As shown in the figure below, Fig. 4 shows the motion blurred image acquired by the uniform speed motion of the guided vehicle [19]. Figure 4 is an reconstructed image by using LR algorithm 10 times, Fig. 4 is an reconstructed image by using LR algorithm 20 times, Fig. 7 is an reconstructed image by using LR algorithm 30 times, Fig. 7 is an image reconstructed 40 times by LR algorithm, and Fig. 7 is an image reconstructed 50 times by LR algorithm.

Table 1
L-R algorithm different iteration times to restore the parameters of the image evaluation parameters

Algorithm	The number of	Mean square error	Peak signal to noise	Recovery time t/s
	iterations	MSE/dB	ratio PNSR/dB
L-R iteratively restored image	Original blurred image	253.5	9.4	0.0
	Iteration 10 times	176.6	13.5	3.2
	Iteration 20 times	84.0	23.8	3.8
	Iteration 30 times	79.3	25.4	4.9
	Iteration 40 times	73.9	25.9	5.6
	Iteration 50 times	69.2	26.4	7.3

Figure 2.

Motion blur image.

Figure 3.

Iterate 10 times to restore the image.

Figure 4.

Iterate 20 times to restore the image.

Figure 5.

Iterate 30 times to restore the image.

Figure 6.

Iterate 40 times to restore the image.

Figure 7.

Iterate 50 times to restore the image.

The PNSR value is used to evaluate the approximation degree of the image. Larger PNSR value means less distortion. The value of MSE is used to evaluate the difference between the blurred image and the restored image. The smaller the MSE is, the better the image quality is. The recovery time represents rapidity and evaluates the visual guidance vehicle’s visual guidance in real-time. Figure 4 is a Wiener filtering algorithm to recover the image [15], take NSR $=$ 0 and NSR $=$ 0.01 Wiener filter to restore the image, Fig. 9 is NSR $=$ 0 Wiener filter to restore the image, Fig. 9 is NSR $=$ 0.01 Wiener filter recovery. The corresponding evaluation parameters are shown in Table 2.

Table 2

Wiener filter restored image evaluation parameters

Algorithm	Recovery time t/s	Peak signal to noise ratio	Mean square
		PNSR/dB	error/MSE
NSR $=$ 0 Wiener filter recovery recorded	4.1	20.2	152.4
NSR $=$ 0.01 Wiener filter recovery result	4.1	11.3	272.2

Figure 8.

NSR $=$ 0 Wiener filter to reconstruct the image.

Figure 9.

NSR $=$ 0.01 Wiener filter to reconstruct the image.

As can be seen from Tables 1 and 2, L-R algorithm with more than 10 iterations is significantly better than Wiener filter. If we use LR algorithm to reconstruct the noisy image without any noise, With the increase of the number of iterations, the recovery effect becomes better and better, but the number of iterations increases and the time consumption also increases, which seriously affects the real-time performance of the visual guidance vehicle guidance and the rapidity of the system response. When the number of iterations is 10 times and 20 times, the sharpness of the image is obviously improved and the restoration effect is better. When the number of iterations is more than 20 times, the image restoration effect is improved less, and the recovery time will increase sharply, it seriously affects the system fast response. Therefore, the best iteration times of blurred image restoration for visual guidance vehicles can be obtained through experiments, and the LR algorithm based on the iteration number of 20 times can fully meet the requirements of visual guidance vehicles for noiseless motion blurred image restoration claim.

4.2 Experimental analysis of motion blur image restoration with noise

Figure 12 is a noisy motion blurred image, Fig. 12 is an image reconstructed 10 times by LR algorithm, Fig. 12 is an image reconstructed by 20 LR algorithms, Fig. 15 is an image reconstructed by 30 LR algorithms, Fig. 15 is an iterative 40 times the LR algorithm to restore the image, Figure 15 is an iterative 50 times LR algorithm to restore the image.

Table 3
Parameters of L-R algorithm to reconstruct the noisy image with different iteration times

Algorithm	The number of	Mean square	Peak signal to noise	Recovery time
	iterations	error MSE/dB	ratio PNSR/dB	t/s
L-R iteratively restored image	Original blurred image	382.9	4.4	0.0
	Iteration 10 times	196.0	9.8	3.4
	Iteration 20 times	99.8	16.3	4.1
	Iteration 30 times	107.5	15.7	5.0
	Iteration 40 times	175.0	11.3	6.3
	Iteration 50 times	285.3	6.1	7.5

Figure 10.

Motion blurred image with noise.

Figure 11.

Iterate 10 times to restore the image.

Figure 12.

Iterate 20 times to restore the image.

Figure 13.

Iterate 30 times to restore the image.

Figure 14.

Iterate 40 times to restore the image.

Figure 15.

Iterate 50 times to restore the image.

From the above table, when the L-R algorithm recovers a blurred image with noise, the image restoration effect is significant when the number of iterations is 20 times or less; when it is more than 20 times, the noise is also amplified as the number of iterations increases, and the recovery effect becomes worse. It can be seen from blurred image restoration and system rapidity of the visual guidance vehicles, the iterative 20 times L-R algorithm can effectively solve the requirement of the visual guidance vehicle to recover the blurred image, and has better fastness and robustness.

Conclusion: Based on the noise-free and noise-containing image restoration experiment, it can be concluded that the LR algorithm can meet the requirement of fastness, restoration accuracy and noise suppression. The best iteration is 20 times.

4.3 Result verification

The Wiener filter algorithm and the blurred image restoration algorithm based on the best iteration number of visual guidance vehicle proposed in this paper are respectively carried out by the self-developed visual guidance vehicle at a working speed of 0.5 m/s [16]. The maximum trajectory detected is the track tracking accuracy; the straight line distance from the start of the offset to the return guide path is the correction distance, and the experimental results are shown in Table 4.

Table 4
Experimental data of different algorithms

Algorithm	Tracking accuracy/mm	Correction distance/m
Wiener filter algorithm	6.2	2.3
The best to take the number of L-R algorithm	3.5	1.6

The Wiener filter algorithm is used to recover the blurred image. For the non-linear blurred image with noise, the poor recovery effect is not conducive to the control stability and the trajectory tracking accuracy is reduced. This paper proposes a blurred image restoration algorithm for visual guidance vehicle based on the optimal number of iterations, and improves the L-R algorithm to obtain the optimal number of iterations based on this paper. The blurred image restoration time is shortened, the stability of the control system is improved, and the final tracking accuracy is stable within 3.5 mm.

5. Conclusion

Based on the noise-free and noise-containing image restoration experiment, the L-R algorithm is obtained. Through the independently developed visual guided vehicle experiment, the superiority of the L-R algorithm with the best number of iterations was verified, and the tracking accuracy was controlled within 3.5 mm. but the tracking accuracy is still lower than that of the laser guidance. The next step is to optimize the motion controller and further improve Vision guided vehicle control accuracy.

Footnotes

Acknowledgments

First of all, I would like to thank my teacher, Luo Zai, from China Jiliang University. Prof. Luo has provided guiding opinions and recommendations for the research direction of my dissertation. In addition, I would also like to thank my friends and classmates for their great support and assistance in the preparation of the essay, which gave me great inspiration. I would also like to thank the authors of the reference literature for their excellent research starting point through their research articles. Finally, thank you for reviewing the teachers’ hard work.

References

Qian

X.M.

et al., Optimization design of frame structure of automatic guide car [J], China Mechanical Engineering 25(19) (2014), 2653–2657, 2664.

G.L.

Wang

X.X.

and Wan

Z.Y.

, Image dehazing method based on fourth-order partial differential equations [J], Electronic Technology 29(2) (2016), 62–65.

Shi

H.L.

and Qiu

X.H.

, Research on Motion Restrained Vehicle Image Restoration Method [J], Computer Technology and Development 26(8) (2016), 60–64.

H.P.

, Research on Method of Restoration of Blurred Image Based on Wavelet Frame [J], Software Herald 15(7) (2016), 186–189.

Zhang

, Image restoration based on improved MDAL algorithm [J], Electronic Technology and Software Engineering (2) (2018), 78–79.

Zhao

L.F.

Wang

A.L.

Wang

et al., A Poisson-Gaussian Noise Image Restoration Algorithm Based on Variational Models [J], Journal of Computer-Aided Design & Computer Graphics 29(3) (2017), 428–435.

Zhao

M.H.

Zhang

Shi

Z.H.

et al., Suppression Method of image restoration boundary ringing effect by sine integration fitting [J], Chinese Journal of Image and Graphics 22(2) (2017), 249–256.

Lai

M.Q.

and Cai

G.C.

, Total variation image restoration based on alternating direction multipliers [J], Computer Technology and Development 27(4) (2017), 60–63.

Liang

R.L.

Chen

C.Q.

et al., Relationship between 3D point spread function layer distance and image restoration [J], Journal of Guangxi University (Natural Science Edition) 41(3) (2016), 810–815.

10.

Yang

Feng

Wang

S.Z.

et al., Degradation image restoration method based on optical model [J], Journal of China Agricultural University 22(1) (2017), 112–119.

11.

Zhang

S.Q.

and Liu

Q.H.

, Research on Uniform Velocity Motion Image Restoration Algorithm [J], Industrial Control Computer 29(7) (2016), 109–110.

12.

Zhang

X.D.

and Zhang

, Image restoration algorithm for improving image quality in special scenes [J], Computer Systems& Applications 25(6) (2016), 147–153.

13.

M.Y.

and Zou

, Image restoration of BP neural network based on cuckoo algorithm [J], Journal of Computer Applications 37(a01) (2017), 173–175.

14.

Y.P.

and Ma

Z.X.

, Motion Blur parameter estimation for image restoration [J], Industrial Control Computer 29(1) (2016), 86–87.

15.

Jiang

Peng

Y.P.

and Wang

, Image restoration based on stochastic resonance and wiener filtering [J], Science Technology and Engineering 16(8) (2016), 223–228.

16.

Wang

Y.K.

Liu

F.P.

and Lu

Z.P.

, Research on blind deconvolution image restoration algorithm [J], Journal of Beijing Institute of Graphic Communication 25(8) (2017), 24–26.

17.

Zhang

Y.Y.

M.Z.

Wang

C.X.

et al., Research on restoration of motion blurred image [J], China Equipment Engineering (6) (2017), 182–183.

18.

Zhang

Y.Y.

Zhou

S.M.

Zhao

Y.L.

et al., Research on motion blurred image restoration for high-speed moving targets [J], Infrared and Laser Engineering 46(4) (2017), 250–255.

19.

Xiao

Z.Y.

S.C.

Liu

et al., Image restoration algorithm for motion blurred instrument based on improved Wiener filtering [J], Electrical Measurement and Instrument 54(2) (2017), 88–91.

20.

Gong

Z.L.

Wang

X.W.

and Du

W.T.

, Blurred image restoration of navel orange based on Lucy-Richardson algorithm with gain control [J], Journal of Jiangxi Agricultural University (1) (2018), 42–46.

Research on blurred image restoration algorithm based on best iteration for vision guidance vehicle

Abstract

Keywords

1. Introduction

2. Visual guidance principle

3. Establishing vision guidance vehicle blurred image recovery model

3.1 Fuzzy kernel function design

Table 1 L-R algorithm different iteration times to restore the parameters of the image evaluation parameters

Table 3 Parameters of L-R algorithm to reconstruct the noisy image with different iteration times

Table 4 Experimental data of different algorithms

Footnotes

Acknowledgments

References

Table 1
L-R algorithm different iteration times to restore the parameters of the image evaluation parameters

Table 3
Parameters of L-R algorithm to reconstruct the noisy image with different iteration times

Table 4
Experimental data of different algorithms