An underwater robot engineering detection video image blurring restoration method and system

Abstract

The application discloses an underwater robot engineering detection video image blur restoration method and system, which is applied to the field of underwater high-end equipment technology and specifically comprises the following steps: obtaining constraint factors, establishing a constraint model of motion vectors and blurred images according to constraint relationships among the constraint factors; fusing the constraint model and gradient constant prior and solving the motion vectors; determining the influence of the estimation error of the motion parameters on the ringing effect, correcting the solving error of the motion parameters through the ringing condition in the restoration result image; determining the influence of the difference of the frequency responses among different code words on the restoration result and the ringing effect, and obtaining a clear restoration result. Compared with the traditional exposure mode, the encoding exposure can effectively avoid the problems of irreversible deconvolution and inconsistent frequency information responses.

Description

Technical Field

[0001] This invention relates to the field of high-end underwater equipment technology, and more specifically to a method and system for restoring blurred video images from underwater robot engineering inspections. Background Technology

[0002] The implementation of the national infrastructure strategy has led to the creation of numerous underwater structures on coastal and inland lakes that are difficult to maintain manually. The construction, safe operation, and maintenance of large bridges and dams on these lakes have created an increasingly urgent need for underwater robots with rapid, real-time operational capabilities. Currently, foreign countries have accumulated rich experience in the research and application of underwater engineering robots, and have successively developed various arm-type and tracked deep-sea operation robots, which provide convenience for deep-sea resource exploration and operations.

[0003] Having solved the problem of ensuring stable operation of robots in complex flow fields in coastal and inland lakes, the issue of high-quality underwater imaging remains the core constraint restricting the application of underwater robots in underwater structure inspection and operations. It is also a core technology that urgently needs to be addressed in this high-end equipment.

[0004] Machine vision, as a crucial component of underwater robot inspection and operations, is responsible for perceiving and recognizing the surrounding environment. However, the quality of images acquired by robots in underwater environments is affected by water quality, current velocity, and robot speed due to the limitations of the working environment. This impact manifests in two main ways: firstly, water quality affects image clarity; secondly, the robot's motion causes motion blur. During underwater robot inspections, the influence of robot motion on image quality is particularly significant, even severely impacting subsequent image processing and recognition tasks. Therefore, researching the restoration of motion-blurred images in underwater engineering inspections is of paramount importance for intelligent image detection by underwater robots and is one of the decisive issues for the widespread application of underwater robots in engineering infrastructure.

[0005] Motion blur can be viewed as a process of convolving the sharp image with a blur kernel, while deblurring is its inverse process: deconvolution. The traditional exposure process of a robotic camera is represented in the real domain as a rectangular window, such as... Figure 1As shown in (a), the presence of zeros in the frequency domain makes the deconvolution process an ill-conditioned problem, and the acquired image loses high-frequency information. Currently, there are many commonly used methods for blurry image restoration, including inverse filtering, Wiener filtering, blind deconvolution algorithms, and the Lucy-Richardson algorithm. Different algorithms have different effects and applicable ranges, but each has its advantages and disadvantages. Encoded exposure can effectively solve the problems of traditional exposure methods in motion blur restoration, providing a new approach to motion blur restoration. However, several technical problems make the research on coded exposure motion blur methods of great practical significance.

[0006] Encoded exposure refers to controlling the opening and closing of the shutter according to a specific binary code during the exposure process, such as... Figure 1 As shown in (b), the process of acquiring moving targets using coded exposure is equivalent to performing time sampling on the signal. Since time-domain sampling causes aliasing in the frequency domain, coded exposure converts the low-pass filter in the frequency domain into a band-pass filter. Zero removal makes the deconvolution problem a well-formed problem, and the acquired blurred image retains a large amount of high-frequency information. The advantages of coded exposure determine its broad application prospects in the industrial field.

[0007] Encoded exposure is significantly superior to traditional exposure methods in terms of imaging effect and effective information acquisition. This conclusion also applies to motion-blurred image restoration. Its high signal-to-noise ratio makes it more valuable for applications in target tracking, fault detection, and other fields.

[0008] Encoded exposure makes the deconvolution process reversible, but accurately estimating the blur kernel is key to the deblurring problem. Although the codewords during exposure are determined, different relative motions between the target and the camera will cause variations in the point spread function during imaging. Therefore, estimating the direction and magnitude of the relative motion between the camera and the target is crucial for obtaining the point spread function. Researchers have proposed several different methods to address this problem, broadly categorized into two types: active methods and passive methods.

[0009] Active methods acquire motion parameters by detecting the relative motion between the camera and the scene using mechanical devices. These methods rely on motion detection equipment to obtain the relative motion between the camera and the target, resulting in a complex system structure. While the acquired motion parameters allow for precise derivation of the point spread function (PSF), they are primarily used in robot vision and aerospace photography, limiting their application scope. For imaging systems where motion parameters cannot be directly acquired, algorithms are needed to estimate the relative motion between the camera and the scene from the acquired blurred image to derive the PSF. This is the passive method, and therefore, the estimation of the motion blur PSF is crucial.

[0010] However, it is difficult to directly estimate motion blur parameters from blurred images. There is currently little research on this problem, and there is no unified theoretical model or mathematical framework. Summary of the Invention

[0011] In view of this, the present invention provides a method and system for blur restoration of underwater robot engineering inspection video images. Compared with traditional exposure methods, coded exposure can effectively avoid the problems of irreversible deconvolution and inconsistent response of information at different frequencies. In addition, coded exposure has strong applicability, can restore blurred images with various motion modes, and has more codeword selection, providing possibilities for operation on target information at different frequencies. The acquired images have a higher signal-to-noise ratio and better quality.

[0012] To achieve the above objectives, the present invention provides the following technical solution:

[0013] A method for restoring blurred images from underwater robot engineering inspection videos, the specific steps of which are as follows:

[0014] Obtain constraint factors, and establish a motion vector expression as a constraint model with the blurred image based on the constraint relationship between the constraint factors;

[0015] The constraint model is fused with the gradient constant prior, and the motion vector is solved;

[0016] The influence of motion parameter estimation error on ringing effect is determined. By analyzing the ringing situation in the restored image, the solution error of motion parameters is corrected. The influence of frequency response differences between different codewords on the restored result and ringing effect is determined, and a clear restored result is obtained.

[0017] Optionally, in the above-mentioned method for restoring blurred video images of underwater robot engineering inspection, the specific steps for establishing a constraint model of motion vectors and blurred images are as follows:

[0018] A blurred image is represented as a superposition of a blurred foreground and a sharp background, i.e.: I b =αF b +(1-α)B, where, I b F b B and B represent the blurred image, blurred foreground, and sharp background, respectively, and α represents the transparency information of the foreground image;

[0019] Determine the direction of relative motion, as well as the motion vector and motion fuzzy kernel within the corresponding time period;

[0020] The gradient vector of the blurred image at a certain point is calculated based on the motion blur kernel, and the relationship between the gradient vector and the motion vector is obtained.

[0021] The sharp image is represented by a moving foreground image and a stationary background image through a linear combination of α and (1-α). The blurred image is derived from the linear combination of the moving foreground image and the stationary background image through (α*h) and (1-α*h). The relationship between the moving foreground image and the stationary background image is fused with the relationship between the gradient vector, the motion vector, and the pixel value of the sharp image to obtain the relationship between the α image gradient and the motion vector, which is the constraint model for encoding exposure motion blur.

[0022] Optionally, in the above-mentioned method for restoring blurred video images of underwater robot engineering inspection, the specific steps for solving the problem are as follows:

[0023] According to the design of the constraints in the constraint model, the motion vector is only related to the gradient of the α image. The values ​​of the foreground and background in the α image are constant values ​​of 1 and 0, respectively, and the gradient is zero. Only the values ​​at the motion blur edges are in the range of (0,1) and the gradient value is non-zero. By constructing a pre-filter before optimization, the set of points that satisfy the constraint conditions is used as the feasible region for solving the motion vector.

[0024] Based on the influence of noise on the gradient of the α image along the motion direction, the constraint model and gradient constant prior are fused and introduced into intermediate variables, transforming it into an optimization problem for the motion vector. Then, for the motion vector and the gradient affected by noise, the feasible region is selected, and the precise motion vector and intermediate variables are iteratively optimized to solve.

[0025] Optionally, in the above-mentioned underwater robot engineering inspection video image blur restoration method, in view of the acquisition error of α image, the optimal codeword design scheme is determined by the influence of the frequency characteristics and noise characteristics of codeword on the restoration result; local pixels with the same constraint characteristics are retained, and a local consistency detection method is designed to eliminate the solution error caused by local inconsistency.

[0026] Based on the impact of motion parameter solution errors on the restoration results, the corresponding error sources are inferred from the ringing situation in the restored image, the motion vector solution process is corrected, and finally a clear image unaffected by the ringing effect is restored.

[0027] A system for restoring blurred images from underwater robot engineering inspection videos includes:

[0028] The constraint model construction module obtains constraint factors and establishes a constraint model of motion vectors and blurred image based on the constraint relationship between the constraint factors.

[0029] The solution module integrates the constraint model with the gradient constant prior and solves for the motion vector;

[0030] The correction module determines the impact of motion parameter estimation errors on the ringing effect, corrects the motion parameter solution errors by analyzing the ringing patterns in the restored image, and determines the impact of frequency response differences between different codewords on the restored result and the ringing effect, thereby obtaining a clear restored result.

[0031] Optionally, in the above-mentioned underwater robot engineering inspection video image blur restoration system, the constraint model construction module includes:

[0032] The representation unit, a blurred image, is the superposition of a blurred foreground and a sharp background, i.e.: I b =αF b +(1-α)B, where, I b F b B and B represent the blurred image, blurred foreground, and sharp background, respectively, and α represents the transparency information of the foreground image;

[0033] The calculation unit determines the direction of relative motion, as well as the motion vector and motion fuzzy kernel within the corresponding time period;

[0034] The first relationship determination unit calculates the gradient vector of the blurred image at a certain point based on the motion blur kernel, and obtains the relationship between the gradient vector and the motion vector.

[0035] The second relation determination unit derives from the clear image, which is represented by a moving foreground image and a stationary background image through a linear combination of α and (1-α), that the blurred image is the result of a linear combination of a moving foreground image and a stationary background image through (α*h) and (1-α*h).

[0036] The fusion unit fuses the relationship between the moving foreground image and the static background image with the relationship between the obtained gradient vector, motion vector, and pixel value of the clear image to obtain the relationship between the α image gradient and the motion vector, which is a constraint model for encoding exposure motion blur.

[0037] Optionally, in the above-mentioned underwater robot engineering inspection video image blur restoration system, the solution module includes:

[0038] The feasible region unit, according to the design of the constraints in the constraint model, the motion vector is only related to the gradient of the α image. The values ​​of the foreground and background in the α image are constant values ​​of 1 and 0 respectively, and the gradient is zero. Only the values ​​at the motion blur edges are in the range of (0,1) and the gradient value is non-zero. By constructing a pre-filter before optimization, the set of points that satisfy the constraint conditions is used as the feasible region for solving the motion vector.

[0039] The solution unit, based on the influence of noise on the gradient of the α image along the motion direction, integrates the constraint model and the gradient constant prior and introduces intermediate variables, transforming it into an optimization problem for the motion vector. Then, for the motion vector and the gradient affected by noise, it selects the feasible region and iteratively optimizes the solution for the accurate motion vector and intermediate variables.

[0040] Optionally, in the above-mentioned underwater robot engineering inspection video image blur restoration system, the correction module includes:

[0041] The first error correction unit, in response to the acquisition error of the α image, determines the optimal codeword design scheme by considering the influence of the frequency characteristics and noise characteristics of the codeword on the restoration result; it retains local pixels with the same constraint characteristics and designs a local consistency detection method based on the constraint characteristics to eliminate the solution error caused by local inconsistencies;

[0042] The second error correction unit, based on the impact of motion parameter solution errors on the restoration result, infers the corresponding error sources through the ringing situation in the restored image, corrects the motion vector solution process, and finally restores a clear image unaffected by the ringing effect.

[0043] As can be seen from the above technical solution, compared with the prior art, this invention discloses a method and system for restoring blurred video images in underwater robot engineering inspection. Compared with traditional exposure methods, coded exposure can effectively avoid the problems of irreversible deconvolution and inconsistent response of information at different frequencies. Furthermore, coded exposure has strong applicability, can restore blurred images with various motion modes, and offers more codeword selection, providing possibilities for operation on target information at different frequencies. The acquired images have a higher signal-to-noise ratio and better quality. With the progress of my country's economy and society, blurred image restoration is a key and challenging problem in intelligent transportation, criminal investigation, and aerospace photography. Underwater robots also face motion blur problems in intelligent inspection, which affects the efficiency and accuracy of intelligent monitoring, severely restricting the application and promotion of underwater robots in underwater engineering. This invention can not only greatly improve the image clarity of underwater robots during operation, thereby meeting the needs of rapid inspection and operation of large underwater structures, but also extend the application of the results to related fields such as underwater monitoring and underwater photography under high-velocity conditions, possessing significant practical value. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0045] Figure 1 (a)-(b) are schematic diagrams of the exposure process of a traditional exposure camera and an coded exposure camera;

[0046] Figure 2 This is a flowchart of the method of the present invention;

[0047] Figure 3 This is a flowchart of the constraint model construction method of the present invention;

[0048] Figure 4 This is a schematic diagram of the motion vector solving process of the present invention;

[0049] Figure 5 This is a flowchart of the ringing effect suppression method of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] To achieve motion fuzz restoration and correction for underwater robots, the general implementation steps of this method are as follows: Figure 2 As shown, it can be completed in the following three steps:

[0052] (1) Establishment of the exposure motion blur constraint model. By acquiring factors related to the direction and speed of motion, the motion vector is expressed as a constraint model related to the blurred image information.

[0053] (2) Solving for motion vectors. For different regions in the blurred image, the uncertainties in the constraint model are further restricted, and based on this, the constraint model is transformed into a problem of solving for the optimization of motion vectors.

[0054] (3) Suppression of ringing effect in the restoration result of motion blur caused by exposure. In order to avoid secondary degradation of the restored image, the influence of the estimation error of motion parameters on the ringing effect is first determined; then, the solution error of motion parameters is corrected by the ringing situation in the restored image; finally, on this basis, the influence of the difference in frequency response between different codewords on the restoration result and the ringing effect is determined, and a clear restoration result is obtained.

[0055] The specific implementation steps are as follows:

[0056] The generative model for underwater robot-encoded motion-blurred images is represented as: I b=I*h+n, where, I b Let I be the acquired blurred image, h be the blurred kernel, n be the additive noise, and * be the convolution operation.

[0057] (1) Establishment of the exposure motion blur constraint model. Using vector (u,v) T It represents the relative velocity of the target in the horizontal and vertical directions in the image, and the direction of motion and the magnitude of the velocity are constrained as parameters in the model equation.

[0058] A blurred image can be understood as the superposition of a blurred foreground and a sharp background, that is: I b =αF b +(1-α)B, where, I b F b B and B represent the blurred image, blurred foreground, and sharp background, respectively, while α represents the transparency information of the foreground image (referred to as the α image). Since the α image is 1 and 0 at the sharp background and blurred foreground, respectively, and transitions from 1 to 0 at the edge of the moving target, it can intuitively reflect the direction and magnitude of the motion causing the blur.

[0059] First, a blurred image can be represented by an α image, a blurred foreground, and a sharp background. The gradient of the blurred image can be expressed as the relationship between the gradient of the α image and the blurred foreground and sharp background. The gradient of the blurred image along the motion direction is related to the blurred foreground, sharp background, motion vector, and the gradient of the α image.

[0060] Secondly, two pixels of a clear image can also be represented by a blurred foreground, a clear background, and an alpha image; furthermore, codewords can determine the energy distribution of the blurred foreground at different locations, and the gradient of the alpha image at a certain point along the direction of motion has only a finite number of cases.

[0061] Based on the above relationships, determine the constraints that only include motion vectors. Examples include: the relationship between the gradient of a point in the alpha image along the motion direction and the pixel value; the representation of the gradient along the motion direction of the alpha image at different locations in the image; and the influence of the alternation of codewords '0' and '1' in the binary codeword on the gradient of a point in the alpha image along the motion direction. The specific implementation steps for this part are as follows: Figure 3 As shown.

[0062] The relationship between the motion vector and the pixel within a certain time interval during the exposure process is determined below. Assume the exposure time of a segment "101" in the codeword is [t1, t1+3Δt], and the direction of relative motion is horizontal to the right. The motion vector within this time interval is denoted as b=(u,0). T The corresponding motion fuzzy kernel is:

[0063]

[0064] Where x and y are respectively, and N is the number of '1's in the codeword, then:

[0065]

[0066] Let δ be the Dirichlet function. In this case, there exists P = (x0, y0). T ,

[0067]

[0068] in, Let P be the gradient vector of the blurred image. It can be seen that, within a certain time period, in the motion blur result caused by the truncation of a certain codeword, the gradient vector of the blurred image at a certain point is related to the motion vector and the pixel value of the clear image. The blurring effect of all codewords on the image during the entire exposure time can be regarded as the superposition of multiple time periods and the effect of codeword truncation.

[0069] A sharp image can be understood as a linear combination of a moving foreground image and a stationary background image through α and (1-α). Therefore, a blurred image can be seen as the result of a linear combination of a moving foreground image and a stationary background image through (α*h) and (1-α*h). By fusing this relationship with the previously obtained relationship between the motion vector, the pixel value of the sharp image, and the gradient vector, we can obtain the relationship between the α image gradient and the motion vector. This relationship is defined as the coded exposure motion blur constraint model (Model A).

[0070] (2) Solving for motion vectors. According to the design method of constraints in the motion blur constraint model, the motion vectors are only related to the gradient of the α image. The values ​​of the foreground and background in the α image are constant values ​​of 1 and 0, respectively, and the gradient is zero. Only the values ​​at the motion blur edges are in the range of (0,1) and the gradient value is non-zero. By constructing a pre-filter before optimization, the set of points that satisfy the constraint conditions is used as the feasible region (ScopeΩ) for solving the motion vectors.

[0071] Motion blurring is equivalent to dispersing the energy originally concentrated at a single point in an image to different locations, resulting in a reduced signal-to-noise ratio and loss of high-frequency information. Furthermore, the choice of codewords can cause discontinuous blur edges. To address this, this invention designs a non-zero gradient constant prior (PriorP) along the motion direction.

[0072] Considering that the gradient of the α image along the motion direction may be affected by noise, the constraint model "A" and the gradient constant prior "P" are fused and introduced as intermediate variables, transforming the problem into an optimization problem for the motion vector. Then, for the motion vector and the noise-affected gradient, a feasible region "Ω" is selected, and the exact motion vector and intermediate variables are iteratively optimized. The schematic diagram of the solution process is shown below. Figure 4As shown.

[0073] (3) Methods to suppress the ringing effect. Because the estimation error of motion parameters will seriously affect the quality of the restored image, mainly in terms of the direction and length of relative motion. When the estimated motion length is greater than the actual length, the ringing effect will be very obvious, and conversely, the image restoration effect will be poor.

[0074] To address the acquisition error of the α image, the optimal codeword design scheme is determined by analyzing the impact of codeword frequency and noise characteristics on the reconstruction result. Retaining local pixels with the same constraint characteristics is a necessary prerequisite for improving solution accuracy. This invention will design a local consistency detection method based on constraint characteristics to eliminate solution errors caused by local inconsistencies. Specific analysis details are as follows: Figure 5 As shown.

[0075] Based on the analysis of the impact of motion parameter solution errors on the restoration results, the corresponding error sources are inferred by the ringing situation in the restored image, the motion vector solution process is corrected, and finally a clear image unaffected by the ringing effect is restored.

[0076] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0077] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for restoring blurred images from underwater robot engineering inspection videos, characterized in that, The specific steps are as follows: Obtain constraint factors, and express the motion vector as a constraint model related to the blurred image based on the constraint relationship between the constraint factors; The constraint model is fused with the gradient constant prior, and the motion vector is solved; The influence of motion vector estimation error on ringing effect is determined, and the solution error of motion vector is corrected by observing the ringing situation in the restored image; the influence of frequency response differences between different codewords on the restored result and ringing effect is determined, and a clear restored result is obtained. The specific steps to express motion vectors as a constraint model related to a blurred image are as follows: A blurred image is represented as a superposition of a blurred foreground and a sharp background, that is: I b = αF b + (1- α ) B ,in, I b , F b and B These represent a blurred image, a blurred foreground, and a sharp background, respectively. α Transparency information for the foreground image; Determine the direction of relative motion, as well as the motion vector and motion fuzzy kernel h within the corresponding time period; The gradient vector of the blurred image at a certain point is calculated based on the motion blur kernel, and the relationship between the gradient vector and the motion vector is obtained. A sharp image is represented by a moving foreground image and a static background image. α With (1- α A linear combination of ) leads to the conclusion that a blurred image is a moving foreground image and a stationary background image obtained through ( α h ) and (1- α h The result of linear combination is the fusion of the relationship between the moving foreground image and the static background image, the obtained gradient vector, the relationship between the motion vector and the pixel value of the sharp image, to obtain... α The relationship between image gradient and motion vector is defined as a constraint model that encodes exposure motion blur. In this image, the α value is 1 and 0 in the clear background and the blurred foreground, respectively, and transitions from 1 to 0 at the edge of the moving target, reflecting the direction and magnitude of the motion that causes the blur.

2. The method for restoring blurred video images of underwater robot engineering inspection according to claim 1, characterized in that, The specific steps to solve the problem are as follows: Based on the constraint design in the constraint model, the motion vector only interacts with... α It is related to the image gradient, and α The values ​​at the foreground and background in the image are constant values ​​of 1 and 0, respectively, and the gradient is zero. Only the values ​​at the motion blur edges are in the range of (0,1) and the gradient value is non-zero. By constructing a pre-filter before optimization, the set of points that satisfy the constraints is used as the feasible region for solving the motion vector. according to α The gradient of the image along the direction of motion is affected by noise. The constraint model and gradient constant prior are fused and intermediate variables are introduced to transform it into an optimization problem for the motion vector. Then, for the motion vector and the gradient affected by noise, the feasible region is selected, and the precise motion vector and intermediate variables are iteratively optimized to solve.

3. The method for restoring blurred video images of underwater robot engineering inspection according to claim 1, characterized in that, against α The image acquisition error is analyzed by examining the impact of the frequency and noise characteristics of the codewords on the restoration result, and the optimal codeword design scheme is determined. Local pixels with the same constraint characteristics are retained, and a local consistency detection method is designed to eliminate the solution error caused by local inconsistencies. Based on the impact of motion vector solution errors on the restoration results, the corresponding error sources are inferred from the ringing situation in the restored image, the motion vector solution process is corrected, and finally a clear image unaffected by the ringing effect is restored.

4. A system for restoring blurred video images from underwater robot engineering inspections, characterized in that, include: The constraint model construction module obtains constraint factors and expresses the motion vector as a constraint model related to the blurred image based on the constraint relationship between the constraint factors. The solution module integrates the constraint model with the gradient constant prior and solves for the motion vector; The correction module determines the impact of motion vector estimation error on ringing effect, corrects motion vector solution error by analyzing ringing patterns in the restored image, and determines the impact of frequency response differences between different codewords on the restoration result and ringing effect, thus obtaining a clear restoration result. The constraint model building module includes: The representation unit, a blurred image, is the superposition of a blurred foreground and a sharp background, that is: I b = αF b + (1- α ) B ,in, I b , F b and B These represent a blurred image, a blurred foreground, and a sharp background, respectively. α Transparency information for the foreground image; The calculation unit determines the direction of relative motion, as well as the motion vector and motion fuzzy kernel h within the corresponding time period; The first relationship determination unit calculates the gradient vector of the blurred image at a certain point based on the motion blur kernel, and obtains the relationship between the gradient vector and the motion vector. The second relationship determination unit, represented by a clear image as a moving foreground image and a stationary background image, is determined by... α With (1- α A linear combination of ) leads to the conclusion that a blurred image is a moving foreground image and a stationary background image obtained through ( α h ) and (1- α h The result of a linear combination; The fusion unit fuses the relationship between the moving foreground image and the static background image, along with the obtained gradient vector, motion vector relationship, and pixel values ​​of the sharp image, to obtain... α The relationship between image gradient and motion vector is defined as a constraint model that encodes exposure motion blur. In this image, the α value is 1 and 0 in the clear background and the blurred foreground, respectively, and transitions from 1 to 0 at the edge of the moving target, reflecting the direction and magnitude of the motion that causes the blur.

5. The underwater robot engineering inspection video image blur restoration system according to claim 4, characterized in that, The solution module includes: The feasible region element, based on the design of the constraints in the constraint model, has motion vectors that only interact with... α It is related to the image gradient, and α The values ​​at the foreground and background in the image are constant values ​​of 1 and 0, respectively, and the gradient is zero. Only the values ​​at the motion blur edges are in the range of (0,1) and the gradient value is non-zero. By constructing a pre-filter before optimization, the set of points that satisfy the constraints is used as the feasible region for solving the motion vector. Solving unit, according to α The gradient of the image along the direction of motion is affected by noise. The constraint model and gradient constant prior are fused and intermediate variables are introduced to transform it into an optimization problem for the motion vector. Then, for the motion vector and the gradient affected by noise, the feasible region is selected, and the precise motion vector and intermediate variables are iteratively optimized to solve.

6. The underwater robot engineering inspection video image blur restoration system according to claim 4, characterized in that, The correction module includes: The first error correction unit, for α The image acquisition error is analyzed by examining the impact of the frequency and noise characteristics of the codewords on the restoration result, and the optimal codeword design scheme is determined. Local pixels with the same constraint characteristics are retained, and a local consistency detection method is designed to eliminate the solution error caused by local inconsistencies. The second error correction unit, based on the impact of motion vector solution errors on the restoration result, infers the corresponding error sources through the ringing situation in the restored image, corrects the motion vector solution process, and finally restores a clear image unaffected by the ringing effect.

Images