Monocular vision-based structure displacement measurement method under high-temperature working condition

Abstract

The invention discloses a monocular vision-based structure displacement measurement method under a high-temperature working condition, and particularly relates to the field of monocular vision-based image analysis from a bit image to a non-bit image, and the method comprises the steps: extracting a local gray level feature set and carrying out the statistics of gray level distribution parameters based on a continuous frame gray level image sequence collected by a monocular vision system; judging and screening preliminary connected regions, establishing a region degradation probability graph, and constructing a standard Markov random domain graph structure; based on a standard Markov random domain graph structure, a gray gradient feature set is extracted, a local gradient disturbance vector field is established, and a disturbance potential energy graph is generated. By constructing a variation Markov random domain model based on local gray features and mutual information entropy adjustment and combining dynamic degradation characteristic extraction and reference point time sequence trajectory analysis, structural information recovery and displacement measurement of a local degradation region under a high-temperature working condition are realized. And the problems of image connectivity loss and structure tracking failure caused by optical window degradation are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis from positional image to non-positional image based on monocular vision, and more specifically, to a method for measuring structural displacement under high temperature conditions based on monocular vision. Background Art

[0002] Under high temperature conditions, monocular vision systems are widely used in the field of steel structure displacement monitoring, relying on stable image acquisition to ensure the continuity and accuracy of displacement measurement. Usually, the system uses an optical window with high temperature resistance to protect the camera lens, such as quartz glass, to cope with the strong radiation heat flow in the environment. However, under long-term high temperature exposure conditions, the surface of the optical window will inevitably undergo local carbonization, fogging and coking, resulting in a decrease in optical transmittance and an increase in transmission non-uniformity, forming randomly distributed local degradation areas. As the use time increases, this degradation effect tends to expand dynamically, thereby causing non-uniform light transmittance distortion during image acquisition. Affected by this, the bitmap (i.e., binary contour feature) information of the corresponding degraded area in the image is broken or lost, and the original edge contour is partially obscured. The traditional image segmentation processing method based on global threshold or edge detection cannot effectively extract the complete boundary; at the same time, the non-bitmap grayscale field and local texture features in the degraded area are prone to grayscale drift and texture blur due to carbonization occlusion, which seriously interferes with the image structure information extracted based on grayscale or morphological operators; Especially under monocular vision conditions, the lack of multi-view redundant information compensation and local optical degradation directly lead to the decline of structural integrity in the image processing stage, which cannot support the reliable calculation of subsequent displacement; Furthermore, the imaging anomaly caused by optical window degradation is not fixed, but evolves dynamically with environmental thermal disturbances, sediment generation rate and window aging, forming a "degradation drift" phenomenon with non-uniform expansion in space and time. This phenomenon not only destroys the regional connectivity of static images, but also makes it difficult for fixed templates or static segmentation strategies to adapt to local changes, ultimately causing the failure of the reference benchmark in the displacement measurement process. In summary, under high temperature conditions, the existing monocular vision system suffers from the loss of local image information and non-image texture blurring caused by the degradation of the optical window surface, which seriously limits the accuracy and reliability of steel structure displacement measurement based on image processing, and becomes the core problem restricting the application performance of monocular vision in high temperature environments. Summary of the invention

[0003] To overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for measuring structural displacement under high-temperature conditions based on monocular vision. By constructing a variant Markov random field model based on local gray-scale features and mutual information entropy adjustment, combined with dynamic degradation characteristics extraction and reference point time-series trajectory analysis, the structural information recovery and displacement measurement of local degradation regions under high-temperature conditions are realized, overcoming the problems of image connectivity loss and structural tracking failure caused by the degradation of the optical window.

[0004] To achieve the above object, the present invention provides the following technical solution: A method for measuring structural displacement under high-temperature conditions based on monocular vision, including a monocular vision system for collecting image data; S1. Based on the continuous frame gray-scale image sequence collected by the monocular vision system, extract the local gray-scale feature set and statistically calculate the gray-scale distribution parameters, complete the determination and screening of the initial connected regions, establish a regional degradation probability map, and build a standard Markov random field map structure; S2. Based on the standard Markov random field map structure, extract the gray-scale gradient feature set and establish a local gradient perturbation vector field, generate a perturbation potential energy map, superimpose the mutual information entropy adjustment matrix to construct a coupled potential energy structure, combine the gray-scale time-series change sequence to construct a time-series coherence constraint matrix, and jointly update the neighborhood transition probability to form a variant Markov random field model; S3. Apply the variant Markov random field model to the current frame gray-scale image, solve the local region state configuration to generate a segmentation map, perform morphological closing operation to generate a morphological structure restoration map, extract the connected regions and screen the effective boundary set, calculate the coordinates of the boundary center points to form an initial reference point set for displacement measurement, and track the characteristics of the reference point set frame by frame in combination with the continuous frame gray-scale image sequence to establish a reference point time-series trajectory matrix; S4. Calculate the displacement vector sequence based on the reference point time-series trajectory matrix, perform smoothing processing to generate a smoothed displacement vector field, determine local scale drift in combination with the change trend of the mutual information entropy of the variant Markov random field model, and correct the displacement vector field to output the final structural displacement measurement result.

[0005] In a preferred embodiment, in S1, a continuous frame gray-scale image sequence is collected by the monocular vision system, fixed-region sliding window division is performed to form a continuous frame local gray-scale region group, and according to the pixel neighborhood relationship of the continuous frame local gray-scale region group, the gray-scale distribution vectors of each local region are calculated to form a local gray-scale feature set; Based on the local gray-scale feature set, the mean value and second moment of the internal gray-scale distribution of the region are statistically calculated to generate a local gray-scale statistical matrix. The pixel-level connectivity determination is performed using the local gray-scale statistical matrix, the connected regions are extracted to form an initial connected region set, the coefficient of variation of the neighborhood gray-scale is calculated for the initial connected region set, and the regions with the coefficient of variation greater than the preset coefficient of variation threshold are screened to construct a regional degradation probability map; Based on the regional degradation probability graph, the neighborhood consistency rule is set to form the initial neighborhood dependency structure and generate the standard Markov random field graph structure.

[0006] In a preferred embodiment, in S2, based on the standard Markov random field graph structure, the first-order grayscale gradient vector of the local area is extracted to form a grayscale gradient feature set, the grayscale gradient feature set is jointly modeled with the neighborhood grayscale difference matrix to generate a local gradient perturbation vector field, based on the local gradient perturbation vector field, a local perturbation energy function is established, the neighborhood potential energy function is updated, and a perturbation potential energy graph is formed; Introduce regional mutual information entropy distribution, count the interaction between local regional mutual information entropy and neighborhood, construct mutual information entropy adjustment matrix, superimpose the disturbance potential energy map with mutual information entropy adjustment matrix to form a coupled potential energy structure, and define the coupled potential energy neighborhood update rule; The grayscale distribution vector groups extracted from the local grayscale region groups of continuous frames are arranged in time order to generate a local region grayscale temporal change sequence, and a regional temporal continuity constraint matrix is constructed based on the change sequence; Taking the coupled potential energy structure of the neighborhood and the temporal coherence constraint matrix as joint constraints, the neighborhood transition probability is reconstructed to generate a mutation Markov random field model.

[0007] In a preferred embodiment, in S3, the variant Markov random field model is applied to the grayscale image of the current frame, the local region state configuration is solved based on the neighborhood state transition probability matrix, a local region segmentation map is generated, and a morphological closing operation is performed on the local region segmentation map based on the grayscale statistical matrix to generate a morphological structure recovery map; The connected areas in the morphological structure restoration graph are extracted, and the effective boundary set is screened based on the boundary connectivity to form a boundary restoration vector group. The coordinates of the boundary center point are calculated through the boundary restoration vector group to form an initial reference point set for displacement measurement; Based on the initial reference point set of displacement measurement and combined with the continuous frame grayscale image sequence, the local grayscale features and connectivity of the reference point set are tracked frame by frame to establish the reference point temporal trajectory matrix.

[0008] In a preferred embodiment, in S4, based on the reference point timing trajectory matrix, the reference point displacement vector group is calculated frame by frame to form a displacement vector sequence, local timing smoothing is performed on the displacement vector sequence, local non-physical jump trajectories are eliminated, and a smooth displacement vector field is output. The smooth displacement vector field is used according to the mutual information entropy change trend of the variant Markov random field model in the local area to determine the local scale drift and construct a local scale drift matrix; Based on the local scale drift matrix, the displacement vector field is corrected in reverse to form a corrected displacement measurement result set, and the corrected displacement measurement result set is weighted and averaged by the regional confidence weight matrix to form the final structural displacement measurement output result.

[0009] In a preferred embodiment, in S1, it is defined that represents the local gray-scale statistical feature vector extracted at the frame at the position, and the local gray-scale statistical feature vector is constructed as follows: ; where: ; ; ; where is the local gray-scale mean; is the local gray-scale standard deviation; is the local gray-scale skewness; represents the th pixel gray value extracted with as the center in the local gray-scale region group of consecutive frames of the th frame; is the number of pixels in the sliding window; is the frame number of the consecutive-frame gray-scale image sequence; is the coordinate of the center pixel of the sliding window in the image; It is defined that represents the set of filtered connected regions, and the connected regions are extracted and screened by the coefficient of variation; ; where represents the th connected subset in the preliminary connected region; the pixel represents the pixel coordinates within; represents the neighborhood pixel coordinates of the pixel ; represents the neighborhood set of the pixel ; is the Euclidean distance of the local statistical feature vector; is the skewness difference adjustment factor; is the energy threshold preset for connectivity determination; is the gray value of the th frame pixel ; represents the average gray value of; is a preset gray mutation energy threshold; is the local gray skewness corresponding to the pixel position in the th frame; is the local gray skewness corresponding to the pixel position in the th frame; represents the local gray statistical feature vector extracted at the pixel position in the th frame; represents the local gray statistical feature vector extracted at the pixel position in the th frame;

[0010] Define as the standard Markov random field graph structure, which is composed of and ; ; Among them: ; ; Among them is the degradation probability value at the pixel coordinate ; is the gray mutation energy of ; is used to control the attenuation rate of the degradation probability Gaussian mapping in the above formula; is the neighborhood degradation probability difference adjustment factor; is the neighborhood consistency potential value at the pixel coordinate ; is the gradient vector of the th frame at the pixel coordinate ; is the gradient vector of the th frame at the pixel ; is the neighborhood gradient weight adjustment factor; is the set of neighborhood pixels at the pixel coordinate ; is the natural exponential function; is the two-norm; is the degradation probability value at the pixel ;

[0011] In a preferred embodiment, in S2, define as the first-order gray gradient vector of the pixel in the th frame: ; Define as the local gradient perturbation vector field: ; ; ; where is the gradient vector of the th frame at pixel coordinates ; is the Laplacian enhanced gradient vector at pixel coordinates ; is the Laplacian enhanced gradient vector at pixel ; is the second-order derivative of the pixel point; is the set of neighboring pixels of pixel coordinates ; is the local perturbation energy function; is the exponential decay adjustment factor of the local perturbation energy function; is the non-linear amplification order of the perturbation energy; Define as the mutual information entropy value of the th frame pixel coordinates : ; Define to represent the neighborhood mutual information entropy interaction matrix centered on pixel coordinates , which is used to quantify the mutual information entropy difference and its co-variation degree between local pixels and neighboring pixels; ; Define as the coupling potential field: ; where is the probability distribution of the pixel coordinates with the gray value of ; is the set of gray value ranges; is the mutual information entropy value of the th frame pixel ; is the mutual information second co-term adjustment factor; is the mutual information entropy adjustment weight; Define to represent the continuous gray vector sequence of the th frame and its neighboring front and back frame pixel coordinates : ; Define as the grayscale temporal coherence metric of pixel coordinates : ; Define to represent the transition probability of pixel : ; where is the local grayscale vector of the pixel coordinates of the th frame ; is the local grayscale vector of the pixel coordinates of the th frame ; is the local grayscale vector of the pixel coordinates of the th frame ; is the time frame number; represents the local grayscale vector extracted at the pixel coordinates of the th frame ; is the temporal change constraint adjustment factor; is the non - linear order of the temporal coherence constraint; represents the neighboring pixel coordinates traversed in the normalization denominator; is the grayscale temporal coherence metric of ; is the coupling potential energy value at the position of the th frame .

[0012] In a preferred embodiment, in S3, define as the local state label of the pixel coordinates of the th frame ; ; ; ; where is the set of state labels; is the candidate state label; is the current state label of pixel ; is the state consistency constraint adjustment factor; is the Kronecker function. The Kronecker function is an indicator function that takes the value 1 when is equal to , and 0 otherwise; is the grayscale feature similarity constraint adjustment factor; For the Pixel coordinates in the frame Grayscale statistical feature vector of ; For the Pixels in frame Grayscale statistical feature vector of ; Represents the morphological closing operation based on the gray-level statistical matrix; It is the image structure restored after morphological closing operation; Indicates the first connected subsets; is the preset connected region area threshold; is a set of connected regions; express The pixel area; definition Recover the set of vectors for the boundaries: ; Calculate the boundary center point set and generate the initial reference point set : ; definition As reference point Trajectory matrix on the time axis: ; in for The boundary set of ; is the boundary pixel coordinate; is the number of boundary pixels; Represents the neighborhood matching function based on grayscale feature vector; is the number of consecutive frames; As reference point In the The spatial coordinate position in the frame image; As reference point In the The grayscale statistical feature vector corresponding to the new coordinate position determined by the neighborhood matching of the frame; As reference point In the The grayscale statistical feature vector corresponding to the new coordinate position determined by neighborhood matching.

[0013] In a preferred embodiment, in S4, it is defined For the Frame reference point The displacement vector is: ; definition is the smoothed displacement vector: ; where is the width of the smoothing time window; is the time frame number; is the reference point at the spatial coordinate position in the frame image; defines as the local drift metric of the reference point : ; defines as the corrected displacement vector: ; where is the sign of the discrete change rate at adjacent frame numbers; is at the frame at the reference point current mutual information entropy value; is the drift correction adjustment factor; defines as the confidence weight of the reference point : ; defines the final structural displacement measurement output result of the frame as : ; where is the weight matrix adjustment factor.

[0014] Technical effects and advantages of the present invention: 1. By introducing an image analysis mechanism from bit-image to non-bit-image, combining the variant Markov random field model with the gray-scale - mutual information entropy joint modeling method, it is possible to extract the structural features and gray-scale information of the degraded area under the local optical degradation conditions caused by carbonization, atomization, and coking in the optical window, realize the dynamic restoration of the local structure, improve the continuity and reliability of monocular vision displacement measurement under high-temperature working conditions, and provide an effective solution to the problems of local structure fracture and reference failure caused by optical degradation in the existing monocular vision system; 2. Through the local gray-scale distribution vector and the second-order moment statistical characteristics, combined with the connectivity determination and degraded probability map generation mechanism, it can effectively identify the local gray-scale variation feature area, avoid the problems of misjudgment and information loss in the global threshold segmentation method under high-temperature degradation conditions, and enhance the adaptability of the image preprocessing stage to local degradation perturbations; 3. By constructing a gradient perturbation vector field and a mutual information entropy adjustment matrix, coupling them to form a potential energy structure, and combining with the local gray - level temporal coherence constraint, the dynamic change trend of the image degradation region is captured and modeled, overcoming the problem of spatio - temporal non - uniform expansion caused by degradation drift, and improving the stability and continuity of region segmentation and reference extraction; 4. Based on the combined processing method of morphological closing operation and gray - level statistical features, under the background of image degradation, the boundary fracture and detail occlusion are repaired to ensure the complete extraction of connected regions. Further, a stable set of reference points is calculated through the boundary recovery vector group, providing highly reliable reference information for displacement measurement and reducing the displacement calculation error caused by degradation; 5. By constructing a reference point time - series trajectory matrix and a local drift matrix, combining the drift detection and dynamic correction mechanism of the change rate of mutual information entropy, the non - physical jumps caused by local optical degradation changes are eliminated, forming a smooth and physically consistent displacement vector field, and improving the stability of displacement measurement; 6. By introducing a regional confidence weight matrix, adaptively weighting the displacement measurement results according to the local drift stability, avoiding the influence of local abnormal drift reference points on the overall measurement accuracy, reflecting the structural deformation trend under high - temperature complex environments, and improving the reliable measurement ability of the monocular vision system under harsh working conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flowchart of the method steps of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0016] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0017] Refer to the attached Figure 1 description. A structural displacement measurement method under high - temperature working conditions based on monocular vision according to an embodiment of the present invention includes a monocular vision system for collecting image data; S1. Based on the continuous - frame gray - level image sequence collected by the monocular vision system, extract the local gray - level feature set and statistically calculate the gray - level distribution parameters, complete the determination and screening of the preliminary connected regions, establish a regional degradation probability map, and build a standard Markov random field map structure; S2. Based on the standard Markov random field graph structure, extract the gray gradient feature set, establish the local gradient perturbation vector field, generate the perturbation potential energy map, superimpose the mutual information entropy adjustment matrix to construct the coupled potential energy structure, combine the gray time series change sequence to construct the time series coherence constraint matrix, and jointly update the neighborhood transition probability to form a variant Markov random field model; S3. Apply the variant Markov random field model to the current frame gray image, solve the local area state configuration to generate a segmentation map, perform morphological closing operation to generate a morphological structure restoration map, extract the connected regions and screen the effective boundary set, calculate the boundary center point coordinates to form the initial reference point set for displacement measurement, and combine the continuous frame gray image sequence to track the characteristics of the reference point set frame by frame to establish the reference point time series trajectory matrix; S4. Calculate the displacement vector sequence based on the reference point time series trajectory matrix, perform smoothing processing to generate a smooth displacement vector field, combine the change trend of the mutual information entropy of the variant Markov random field model to determine the local scale drift, correct the displacement vector field, and output the final structural displacement measurement result; Among them, the monocular vision system is an image acquisition device based on a single imaging sensor, which is used to obtain two-dimensional image data from a single perspective.

[0018] In S1, collect a continuous frame gray image sequence through the monocular vision system, perform fixed area sliding window division to form a continuous frame local gray area group, calculate the gray distribution vector of each local area according to the pixel neighborhood relationship of the continuous frame local gray area group, and form a local gray feature set; Based on the local gray feature set, statistically calculate the mean and second moment of the gray distribution inside the area to generate a local gray statistical matrix, use the local gray statistical matrix to perform pixel-level connectivity determination, extract the connected regions to form a preliminary connected region set, calculate the coefficient of variation of the neighborhood gray level for the preliminary connected region set, screen the regions with the coefficient of variation greater than the preset coefficient of variation threshold, and construct a region degradation probability map; Based on the region degradation probability map, set the neighborhood consistency rule to form an initial neighborhood dependence structure and generate a standard Markov random field graph structure.

[0019] In S2, based on the standard Markov random field graph structure, extract the first-order gray gradient vectors of the local area to form a gray gradient feature set, jointly model the gray gradient feature set and the neighborhood gray difference matrix to generate a local gradient perturbation vector field, and based on the local gradient perturbation vector field, establish a local perturbation energy function, update the neighborhood potential energy function, and form a perturbation potential energy map; Introduce the regional mutual information entropy distribution, statistically calculate the local area mutual information entropy and neighborhood interaction relationship, construct a mutual information entropy adjustment matrix, superimpose the perturbation potential energy map and the mutual information entropy adjustment matrix to form a coupled potential energy structure, and define the coupled potential energy neighborhood update rule; Arrange the gray - level distribution vector groups extracted from the local gray - level region groups of consecutive frames in chronological order to generate a local - region gray - level time - series change sequence, and construct a regional time - series coherence constraint matrix based on the change sequence; Taking the coupling potential energy structure of the neighborhood and the time - series coherence constraint matrix as joint constraints, reconstruct the neighborhood transition probability to generate a variant Markov random field model.

[0020] In S3, apply the variant Markov random field model to the gray - level image of the current frame, solve the local - region state configuration based on the neighborhood state transition probability matrix to generate a local - region segmentation map. On the local - region segmentation map, perform morphological closing operation based on the gray - level statistical matrix to generate a morphological structure restoration map; Extract the connected regions in the morphological structure restoration map, screen the effective boundary set based on boundary connectivity to form a boundary restoration vector group, and calculate the coordinates of the boundary center points through the boundary restoration vector group to form an initial displacement measurement reference point set; Based on the initial displacement measurement reference point set, combined with the consecutive - frame gray - level image sequence, track the local gray - level features and connectivity relationships of the reference point set frame by frame to establish a reference - point time - series trajectory matrix.

[0021] In S4, based on the reference - point time - series trajectory matrix, calculate the reference - point displacement vector group frame by frame to form a displacement vector sequence. Perform local time - series smoothing processing on the displacement vector sequence, remove local non - physical jump trajectories, output a smoothed displacement vector field, and determine local scale drift based on the change trend of the mutual information entropy of the variant Markov random field model in the local region to construct a local scale drift matrix;

[0022] Based on the local scale drift matrix, reverse - correct the displacement vector field to form a corrected displacement measurement result set. Weight - average the corrected displacement measurement result set through the regional confidence - weight matrix to form the final structural displacement measurement output result.

[0023] It should be noted that in the formula structure involved in this solution, dimensionless terms can be used as proportional or structural adjustment factors. When combined with quantities with units, they only play a role in numerical scaling and do not introduce new physical dimensions, so they will not change or confuse the overall unit system of the expression; such combinations of "dimensionless terms and terms with units" can be understood as the composite - structure expression forms commonly used in mathematical - physical modeling, which conform to the principle of dimensional consistency and have a clear physical - interpretation basis; Secondly, in the formula structure of this solution, if there are multiple variable terms with different physical units, including but not limited to time - related, mass - related, or energy - related variables, their combined appearance is for expressing the co - modeling relationship of multiple physical mechanisms. Each variable can form a unified structure through function mapping, ratio combination, or normalization adjustment, with clear units and definite meanings. The overall expression conforms to the principle of dimensional consistency and the common paradigm of engineering modeling; In this solution, if there are constants, weights, adjustment factors, threshold parameters, proportionality coefficients, etc., they all belong to adjustable control parameters for different application environments. Their values depend on the target device configuration, data input characteristics, and performance optimization goals, and converge and are set within a reasonable range through methods such as model verification, performance constraints, or engineering calibration during the implementation stage. Although these parameters do not have a preset unique value, they have clear adjustment logic and calculation paths, belonging to a deterministic setting process in engineering implementation. The purpose of such setting is to ensure that the solution has both general adaptability and reproducibility and operability, without affecting its technical clarity and implementability; In S1, define to represent the local gray - scale statistical feature vector extracted at the position of the th frame, and construct the local gray - scale statistical feature vector: ; Where: ; ; ; Where is the local gray - scale mean value, and in the above formula, the local gray - scale mean value represents the arithmetic average of the pixel gray - scale values within the window; is the local gray - scale standard deviation, and in the above formula, the local gray - scale standard deviation represents the square root of the sum of the squares of the pixel gray - scale values deviating from the mean within the window; is the local gray - scale skewness, and in the above formula, the local gray - scale skewness represents the third - order central moment of the gray - scale values within the window about the mean; represents the th pixel gray - scale value extracted with as the center in the local gray - scale region group of consecutive frames of the th frame; is the number of pixels within the sliding window; is the frame number of the consecutive - frame gray - scale image sequence; is the coordinate of the center pixel of the sliding window in the image; Define to represent the set of filtered connected regions, and extract and screen the connected regions with the coefficient of variation; ; Where represents the th connected subset in the preliminary connected region; the pixel represents the pixel coordinates within; represents the neighborhood pixel coordinates of the pixel; represents the neighborhood set of the pixel; is the Euclidean distance of the local statistical feature vector; is the skewness difference adjustment factor, which is used to weigh the influence of skewness. In practical applications, it can be selected within the range of [0.1, 10] through cross-validation according to the requirement of the asymmetry sensitivity of the local gray-scale distribution. The larger the value, the more it tends to enhance the weight of the local gray-scale skewness, and the smaller the value, the more it tends to balance the influence of the gray-scale mean and variance. Among them, skewness itself is the third-order central moment, and its dimension is more volatile compared with variance. Taking 0.1 as the lower limit can avoid the influence of skewness being smoothed out by numerical precision when it is less than 0.1. Taking 10 as the upper limit is because skewness usually does not exceed [-5, 5] in the local area of natural images. Exceeding 10 times magnification will lead to the loss of neighborhood gradient stability and amplify noise errors. Therefore, from 0.1 to 10 can ensure the effective expression of skewness features and avoid numerical loss under the actual gray-scale distribution scale; is the energy threshold preset for connectivity determination; is the gray value of the pixel in the th frame; represents the average gray scale of; is the preset gray-scale variation energy threshold, which is used to screen areas with a high degree of degradation; is the local gray-scale skewness corresponding to the pixel at the position in the th frame; is the local gray-scale skewness corresponding to the pixel at the position in the th frame; represents the local gray-scale statistical feature vector extracted at the pixel position in the th frame; represents the local gray-scale statistical feature vector extracted at the pixel position in the th frame; defines as the standard Markov random field graph structure, which is composed of and ; ; Among them: ; ; where is the degradation probability value at pixel coordinate ; is 's gray - scale variation energy. The gray - scale variation energy represents the overall fluctuation intensity of pixel gray - scale values relative to the mean within a local area, and is used in practical applications to measure the severity of gray - scale changes and the structural complexity within a region; is used in the above formula to control the attenuation rate of the degradation probability Gaussian mapping; is the neighborhood degradation probability difference adjustment factor; additionally and are parameters set artificially based on empirical rules and task characteristics, and can be selected through experimental verification to adapt to different image degradation characteristics, where the larger it is, the higher the discrimination degree of the degradation probability, the smaller it is, the smoother the change of neighborhood potential energy, and there is a trade - off between the adjustment accuracy and robustness of the two; is the neighborhood consistency potential energy value at pixel coordinate ; is the gradient vector of the th frame at pixel coordinate ; is the gradient vector of the th frame at pixel ; is the neighborhood gradient weight adjustment factor, is a parameter set artificially based on experience, and is used to control the influence weight of the gradient difference within the neighborhood on the potential energy function, where the larger it is, the stronger the suppression of neighborhood gradient differences, which helps to enhance the connectivity of smooth regions, the smaller it is, the more sensitive to edge details are maintained, and the structural resolution ability is improved; is the set of neighborhood pixels at pixel coordinate ; is the natural exponential function, that is, the exponential operation with the natural constant e as the base, and is used to describe the exponential decay or growth process; is the two - norm, that is, the Euclidean norm of a vector, which is used to measure the Euclidean distance between two vectors; is the degradation probability value at pixel ;

[0024] In S2, define as the first - order gray - scale gradient vector of the th - frame pixel : ; Define is the local gradient perturbation vector field: ; ; ; where is the gradient vector of the th frame at pixel coordinates ; is the Laplacian enhanced gradient vector at pixel coordinates ; is the Laplacian enhanced gradient vector at pixel ; is the second - order derivative of the pixel, the Laplacian operator; is the set of neighboring pixels of pixel coordinates ; is the local perturbation energy function, which is used to reflect the severity of local perturbations in the above formula; is the exponential decay adjustment factor of the local perturbation energy function. The exponential decay adjustment factor is used to control the influence degree of the perturbation intensity on the energy value. The larger the value, the faster the perturbation energy decays with the gradient change, and the smaller the value, the slower the decay, enhancing the sensitivity of local perturbations to the energy field; is the non - linear amplification order of the perturbation energy, which is used to control the amplification amplitude of the perturbation intensity. The larger the value, the more obvious the non - linear strengthening of local perturbations, and the smaller the value, the weaker the strengthening effect, adjusting the sensitivity of perturbation energy to large - scale gradient changes; Define as the mutual information entropy value of the pixel coordinates of the th frame: ; Define to represent the neighborhood mutual information entropy interaction matrix centered on pixel coordinates , which is used to quantify the mutual information entropy difference and its co - variation degree between local pixels and neighboring pixels; ; Define as the coupling potential energy field: ; where is the probability distribution of the gray value of pixel coordinates being ; is the set of gray - level values; is the mutual information entropy value of the pixel of the th frame; is the second-order co-term regulator of mutual information; is the mutual information entropy regulation weight, where the mutual information entropy regulation weight is used to control the influence intensity of mutual information entropy on the coupling potential field. The larger the value, the more significant the contribution of neighborhood information difference to the potential energy. In practical applications, it can be adaptively set according to the complexity of local gray-scale changes or determined by optimizing empirical parameters; Define to represent the sequence of consecutive gray-scale vector of the pixel coordinates of the frame and its adjacent frames before and after: ; Define as the gray-scale temporal coherence measure of the pixel coordinate : ; Define to represent the transition probability of the pixel : ; where is the local gray-scale vector of the pixel coordinate of the frame; is the local gray-scale vector of the pixel coordinate of the frame; is the local gray-scale vector of the pixel coordinate of the frame; is the time frame number, and the value range of in the formula is { , , }; represents the local gray-scale vector extracted at the pixel coordinate of the frame; is the temporal change constraint regulator, and the temporal change constraint regulator is used to control the influence intensity of temporal coherence on the transition probability; is the non-linear order of the temporal coherence constraint, and the non-linear order of the temporal coherence constraint is used to adjust the amplification or compression effect of the coherence influence; represents the neighborhood pixel coordinates traversed in the normalization denominator, and is used to normalize all neighborhood transition probabilities; is the gray-scale temporal coherence measure; is the coupling potential value at the position of the frame .

[0025] In S3, define as the local status label of the frame pixel coordinates ; ; ; ; where is the set of status labels; is the candidate status label; is the pixel current status label; is the status consistency constraint adjustment factor, which is used to balance the impact of the status label consistency of neighboring pixels on the overall energy minimization. The larger the value, the more consistent the status within the neighborhood. It can be determined according to the segmentation smoothness requirement or through cross-validation; is the Kronecker function, and the Kronecker function is an indicator function that takes the value of 1 when is equal to , and 0 otherwise; is the gray-scale feature similarity constraint adjustment factor, which is used to control the impact of the difference in the gray-scale statistical features of neighboring pixels on the energy minimization. The larger the value, the more consistent the gray-scale features within the segmented region. It can be set according to the image texture complexity or experimental experience; is the gray-scale statistical feature vector of the pixel coordinates in the frame; is the gray-scale statistical feature vector of the pixel represents the morphological closing operation based on the gray-scale statistical matrix; is the image structure restored after the morphological closing operation; represents the th connected subset in the preliminary connected region; is the preset connected region area threshold; is the set of connected regions; represents pixel area, defined as the number of pixels contained in the region, used to quantify the region size; Define as the boundary recovery vector group: ; Calculate the set of boundary center points to generate the initial reference point set : ; Define As the reference point The trajectory matrix on the time axis: ; Where is the boundary set of; is the boundary pixel coordinate; is the number of boundary pixels; represents the neighborhood matching function based on the gray - level feature vector. In the above formula, the neighborhood matching function of the gray - level feature vector is used to measure the similarity of the gray - level feature vectors at two pixel positions in consecutive frames, and determines the corresponding relationship of the reference point in the time series by minimizing the feature difference, supporting the tracking of the reference point; is the number of consecutive frames; is the reference point at the spatial coordinate position in the - th frame image, used to record the trajectory information of the reference point changing with time; is the reference point at the gray - level statistical feature vector corresponding to the new coordinate position determined by neighborhood matching in the - th frame; is the reference point at the gray - level statistical feature vector corresponding to the new coordinate position determined by neighborhood matching in the - th frame.

[0026] In S4, define as the displacement vector of the reference point at the - th frame: ; Define as the smoothed displacement vector: ; Where is the width of the smoothing time window; is the time - frame number. In the formula, the value range of is { , }; is the spatial coordinate position of the reference point at the - th frame image; Define as the local drift metric of the reference point : ; Define as the corrected displacement vector: ; wherein is the symbol of the discrete change rate on adjacent frame numbers, used to measure the change speed of the mutual information entropy in the frame number direction; is the frame's mutual information entropy value at the reference point current , used to describe the uncertainty of the local image gray distribution; is the drift correction adjustment factor, and the drift correction adjustment factor is used to control the influence intensity of the local drift measurement on the displacement vector correction; Define as the confidence weight of the reference point : ; Define the final structural displacement measurement output result of the frame as : ; wherein is the weight matrix adjustment factor, and in practical applications, the weight matrix adjustment factor is used to control 's influence degree on , which determines the drift sensitivity and the weight decay speed. The larger the value, the stronger the inhibitory effect of the drift measurement on the weight, the faster the weight drops, and the stronger the ability to eliminate local drift sensitive points; the smaller the value, the weaker the influence of the drift on the weight, allowing a greater drift tolerance. In practical applications, it can be selected through cross-validation, including but not limited to adjusting within the range of 0.01 - 1, and manually adjusting according to the image noise level and structural stability.

[0027] It should be noted as a whole that this solution is designed for the key problems faced by the application of a monocular vision system in steel structure displacement measurement under high-temperature conditions; in a high-temperature environment, carbonization, atomization, and coking phenomena inevitably occur on the surface of the optical window, resulting in local optical degradation during the imaging process; this degradation effect causes phenomena such as bit image fracture and non-bit image blurring in the collected images, seriously interfering with structure recognition and displacement measurement based on traditional image processing methods; for this reason, this solution is based on the image analysis technology of a monocular vision from bit image to non-bit image, and overcomes the influence brought by optical degradation by constructing a dynamic degradation adaptive image processing and displacement measurement link; ​First, in the image acquisition stage, a monocular vision system is used to obtain a sequence of continuous-frame grayscale images, and the images are divided into sliding windows in a fixed area; for the local area, the gray distribution vector is statistically calculated to form a set of local gray features, and a local gray statistical matrix is constructed through the mean value and the second moment; pixel-level connectivity determination is performed by virtue of the gray distribution characteristics, and the connected regions with a coefficient of variation higher than the set threshold are screened, and then a regional degradation probability map is constructed; by introducing the neighborhood consistency rule, a standard Markov random field graph structure is established to describe the probability correlation between local regions; the purpose of this part is to extract regional degradation information through local statistical characteristics, provide a reliable basis for the subsequent dynamic model construction, avoid misjudgment caused by local gray fluctuations, and improve the anti-degradation robustness of image analysis; On this basis, further relying on the standard Markov random field graph structure, the local gray first-order gradient vector is extracted, and combined with the neighborhood gray difference matrix for joint modeling to generate a local gradient perturbation vector field; for local perturbations, an exponential decay adjustment factor and a nonlinear enhancement mechanism are introduced to construct a local perturbation energy function, thereby forming a perturbation potential energy map; to enhance the adaptability to complex gray changes, a regional mutual information entropy adjustment matrix is superimposed to construct a coupled potential energy structure; by defining the coupled potential energy neighborhood update rule and combining the continuous-frame gray time-series change sequence, a regional time-series coherence constraint matrix is generated; finally, jointly constrained by the coupled potential energy and the time-series coherence, the neighborhood transition probability is reconstructed to form a variant Markov random field model; this part integrates spatial perturbation information and time coherence information, not only improving the sensitivity to degradation dynamic changes, but also effectively suppressing the loss of structural information caused by local degradation drift, and constructing a dynamic adaptive regional division basis; Based on the above variant Markov random field model, it is applied to the current-frame grayscale image, and the local region state configuration is solved by minimizing the local energy function to generate a local region segmentation map. Subsequently, morphological closing operations are performed based on the gray statistical matrix to repair local fractures and detail losses caused by degradation, and a morphological structure restoration map is generated. Through connected region extraction and boundary connectivity screening, an effective boundary restoration vector group is obtained, the coordinates of the center points of each boundary are calculated, and an initial benchmark point set for displacement measurement is formed; based on the initial benchmark point set and combined with the continuous-frame grayscale image sequence, the local gray features and connected relationships of the benchmark point set are tracked frame by frame to establish a benchmark point time-series trajectory matrix; through multi-level structure restoration and benchmark point screening, it is ensured that under optical degradation conditions, the benchmark information points reflecting the real physical structure changes can still be stably extracted and tracked, laying a foundation for the subsequent construction of the displacement vector field; Finally, based on the timing trajectory matrix of the reference points, the displacement vector groups are calculated frame by frame to form a displacement vector sequence. For the locally non-physical jump trajectories that may occur under high-temperature working conditions, a local timing smoothing processing strategy is designed to effectively eliminate abnormal motion vectors and generate a smooth displacement vector field. Further, by combining the change trend of the mutual information entropy in the variant Markov random field model, the local scale drift condition is determined, a local scale drift matrix is constructed, and based on this, the displacement vector field is corrected backward to form a corrected displacement measurement result set. By defining a regional confidence weight matrix and performing weighted averaging based on local drift stability, a stable and reliable structural displacement measurement result is finally output. This process ensures the accuracy and robustness of the displacement measurement results under complex degradation conditions through a three-level mechanism of timing smoothing, drift determination, and confidence weighting. Overall, through the design of a multi-stage and multi-level image analysis and processing link from bit-image information extraction, non-bit-image feature restoration, timing coherence modeling, local drift determination to confidence weighting, this solution breaks through the problems of local occlusion and feature loss caused by optical degradation in monocular vision displacement measurement under high-temperature degradation conditions. The introduction of the standard Markov random field graph and its variant model, combined with the gray gradient perturbation analysis and the mutual information entropy constraint mechanism, enhances the adaptability to complex gray changes and improves the stability of structure recognition and tracking.

[0028] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for measuring structural displacement under high-temperature conditions based on monocular vision, including a monocular vision system for collecting image data, characterized in that: S1. Based on the continuous-frame grayscale image sequence collected by the monocular vision system, extract the local grayscale feature set and statistically calculate the grayscale distribution parameters, complete the determination and screening of the initial connected regions, establish the regional degradation probability map, and build the standard Markov random field map structure; S2. Based on the standard Markov random field map structure, extract the grayscale gradient feature set and establish the local gradient perturbation vector field, generate the perturbation potential energy map, superimpose the mutual information entropy adjustment matrix to construct the coupling potential energy structure, combine the grayscale time-series change sequence to construct the time-series coherence constraint matrix, and jointly update the neighborhood transition probability to form the variant Markov random field model; S3. Apply the variant Markov random field model to the current-frame grayscale image, solve the local region state configuration to generate the segmentation map, perform the morphological closing operation to generate the morphological structure restoration map, extract the connected regions and screen the effective boundary set, calculate the boundary center point coordinates to form the initial displacement measurement reference point set, and combine the continuous-frame grayscale image sequence to track the characteristics of the reference point set frame by frame to establish the reference point time-series trajectory matrix; S4. Calculate the displacement vector sequence based on the reference point time-series trajectory matrix, smooth it to generate the smooth displacement vector field, combine the change trend of the mutual information entropy of the variant Markov random field model to determine the local scale drift, and correct the displacement vector field to output the final structural displacement measurement result.

2. A method for measuring structural displacement under high-temperature conditions based on monocular vision according to claim 1, characterized in that: In S1, collect the continuous-frame grayscale image sequence through the monocular vision system, perform fixed-region sliding window division to form a continuous-frame local grayscale region group, and calculate the grayscale distribution vector of each local region according to the pixel neighborhood relationship of the continuous-frame local grayscale region group to form the local grayscale feature set; Based on the local grayscale feature set, statistically calculate the mean and second moment of the grayscale distribution inside the region to generate the local grayscale statistical matrix, use the local grayscale statistical matrix to perform pixel-level connectivity determination, extract the connected regions to form the initial connected region set, calculate the coefficient of variation of the neighborhood grayscale for the initial connected region set, screen the regions with the coefficient of variation greater than the preset coefficient of variation threshold, and construct the regional degradation probability map; Based on the regional degradation probability map, set the neighborhood consistency rule to form the initial neighborhood dependence structure and generate the standard Markov random field map structure.

3. A method for measuring structural displacement under high-temperature conditions based on monocular vision according to claim 2, characterized in that: In S2, based on the standard Markov random field map structure, extract the first-order grayscale gradient vectors of the local regions to form the grayscale gradient feature set, jointly model the grayscale gradient feature set and the neighborhood grayscale difference matrix to generate the local gradient perturbation vector field, and based on the local gradient perturbation vector field, establish the local perturbation energy function and update the neighborhood potential energy function to form the perturbation potential energy map; Introduce the regional mutual information entropy distribution, statistically analyze the relationship between the local mutual information entropy and neighborhood interaction, construct a mutual information entropy adjustment matrix, superimpose the perturbation potential energy map and the mutual information entropy adjustment matrix to form a coupled potential energy structure, and define the neighborhood update rule of the coupled potential energy; Arrange the gray distribution vector groups extracted from the local gray regions of consecutive frames in chronological order to generate a local region gray time-series change sequence, and construct a regional time-series coherence constraint matrix based on the change sequence; Taking the coupled potential energy structure of the neighborhood and the time-series coherence constraint matrix as joint constraints, reconstruct the neighborhood transition probability to generate a variant Markov random field model.

4. A method for measuring structural displacement under high-temperature conditions based on monocular vision according to claim 3, characterized in that: In S3, apply the variant Markov random field model to the gray image of the current frame, solve the local region state configuration based on the neighborhood state transition probability matrix, generate a local region segmentation map, and perform a morphological closing operation on the local region segmentation map based on the gray statistical matrix to generate a morphological structure restoration map; Extract the connected regions in the morphological structure restoration map, screen the effective boundary set based on boundary connectivity to form a boundary restoration vector group, calculate the coordinates of the boundary center points through the boundary restoration vector group to form an initial set of displacement measurement reference points; Based on the initial set of displacement measurement reference points, combined with the consecutive frame gray image sequence, track the local gray features and connectivity of the reference point set frame by frame to establish a reference point time-series trajectory matrix.

5. A method for measuring structural displacement under high-temperature conditions based on monocular vision according to claim 4, characterized in that: In S4, based on the reference point time-series trajectory matrix, calculate the reference point displacement vector group frame by frame to form a displacement vector sequence, perform local time-series smoothing processing on the displacement vector sequence, remove local non-physical jump trajectories, output a smoothed displacement vector field, and determine local scale drift based on the change trend of the mutual information entropy of the variant Markov random field model in the local region to construct a local scale drift matrix; Based on the local scale drift matrix, reverse-correct the displacement vector field to form a corrected displacement measurement result set, and perform weighted averaging on the corrected displacement measurement result set through a regional confidence weight matrix to form the final output result of the structural displacement measurement.

6. A method for measuring structural displacement under high-temperature conditions based on monocular vision according to claim 5, characterized in that: In S1, define to represent the local gray-scale statistical feature vector extracted at the frame at the position, and construct the local gray-scale statistical feature vector: ； Wherein: ； ； ； where is the local grayscale mean; is the local grayscale standard deviation; is the local grayscale skewness; represents the th grayscale value of the pixel extracted with as the center in the th local grayscale region group of consecutive frames; is the frame number of the sequence of consecutive-frame grayscale images; are the coordinates of the center pixel of the sliding window in the image; Definition Represents the set of filtered connected regions, and performs extraction and coefficient of variation screening on the connected regions; ； where represents the th connected subset in the preliminary connected region; pixel represents the pixel coordinates within; represents the neighboring pixel coordinates of pixel represents the neighborhood set of pixel is the Euclidean distance of the local statistical feature vector; is the skewness difference adjustment factor; is the energy threshold preset for connectivity determination; is the gray value of the pixel in the th frame; represents the average gray value of is the preset gray variation energy threshold; is the local gray skewness corresponding to the pixel position in the th frame; is the local gray skewness corresponding to the pixel position in the th frame; represents the local gray statistical feature vector extracted at the pixel position in the th frame; represents the local gray statistical feature vector extracted at the pixel position in the Definition is a standard Markov random field graph structure, which is composed of and ; ； Wherein: ； ； where is the degradation probability value at the pixel coordinate ; is 's gray-scale variation energy; is used to control the attenuation rate of the degradation probability Gaussian mapping in the above formula; is the neighborhood degradation probability difference adjustment factor; is the pixel coordinate 's neighborhood consistency potential value; is the th frame's gradient vector at the pixel coordinate ; is the th frame's gradient vector at the pixel ; is the neighborhood gradient weight adjustment factor; is the pixel coordinate 's neighborhood pixel set; is the natural exponential function; is the two-norm; is the degradation probability value at the pixel ; 7. A method for measuring structural displacement under high-temperature conditions based on monocular vision according to claim 6, characterized in that: In S2, define as the gray-scale first-order gradient vector of the frame pixels: ； Definition is the local gradient perturbation vector field: ； ； ； where is the gradient vector of the th frame at pixel coordinates ; is the Laplacian-enhanced gradient vector at pixel coordinates ; is the Laplacian-enhanced gradient vector at pixel ; is the second-order derivative of the pixel point; is the set of neighboring pixels of pixel coordinates ; is the local perturbation energy function; is the exponential decay adjustment factor of the local perturbation energy function; is the non-linear amplification order of the perturbation energy; Definition be the frame pixel coordinates mutual information entropy value of: ； Definition Denote the neighborhood mutual information entropy interaction matrix centered at pixel coordinates which is used to quantify the difference in mutual information entropy between local pixels and neighborhood pixels and their degree of co-variation; ； Definition is the coupling potential field: ； wherein is the pixel coordinate and the gray value is the probability distribution; is the set of gray value ranges; is the mutual information entropy value of the pixels in the th frame; is the adjustment factor of the second-order co-term of mutual information; is the adjustment weight of mutual information entropy; Definition Indicating the pixel coordinates of the frame and its adjacent frames before and after, as well as a sequence of consecutive grayscale vectors: ； Definition is the grayscale temporal coherence metric for pixel coordinates : ； Definition Denote the migration probability of pixel as follows: ； wherein is the local gray vector of the pixel coordinates of the th frame; is the local gray vector of the pixel coordinates of the th frame; is the local gray vector of the pixel coordinates of the th frame; is the time frame number; represents the local gray vector extracted at the pixel coordinates of the th frame; is the temporal variation constraint adjustment factor; is the non - linear order of the temporal coherence constraint; represents the neighborhood pixel coordinates traversed in the normalization denominator; is the gray - scale temporal coherence measure of; is the coupling potential energy value at the position of the th frame ; ​ 8. A method for measuring structural displacement under high-temperature conditions based on monocular vision according to claim 7, characterized in that: In S3, define as the local state label of the frame pixel coordinates ; ； ； ； Wherein is a set of status tags; is a candidate status tag; is a pixel 's current status tag; is a status consistency constraint adjustment factor; is a Kronecker function. The Kronecker function is an indicator function that takes the value of 1 when is equal to , and 0 otherwise; is a gray-scale feature similarity constraint adjustment factor; is the th frame's gray-scale statistical feature vector of pixel coordinates ; is the th frame's pixel 's gray-scale statistical feature vector; represents a morphological closing operation based on a gray-scale statistical matrix; is the image structure restored after the morphological closing operation; represents the th connected subset in the preliminary connected region; is a preset connected region area threshold; is a set of connected regions; represents 's pixel area; Definition is the boundary recovery vector group: ； Calculate the set of boundary center points to generate an initial set of reference points : ； Definition as the reference point the trajectory matrix on the time axis: ； wherein is the boundary set; is the boundary pixel coordinate; is the number of boundary pixels; represents the neighborhood matching function based on the gray feature vector; is the number of consecutive frames; is the reference point in the spatial coordinate position in the frame image; is the reference point in the gray statistical feature vector corresponding to the new coordinate position determined by neighborhood matching in the frame; is the reference point in the gray statistical feature vector corresponding to the new coordinate position determined by neighborhood matching in the frame.

9. A method for measuring structural displacement under high-temperature conditions based on monocular vision according to claim 8, characterized in that: In S4, define as the displacement vector of the frame reference point : ； Definition is the smoothed displacement vector: ； wherein is the smoothing time window width; is the time frame number; is the reference point at the spatial coordinate position in the frame image; Definition as the reference point of the local drift metric: ； Definition is the corrected displacement vector: ； wherein is the symbol of the discrete change rate on adjacent frame numbers; is the frame's mutual information entropy value at the reference point current ; is the drift correction adjustment factor; Definition As the reference point Confidence weight of: ； Define the final structural displacement measurement output result of the frame as : ； Among them is the weight matrix adjustment factor.

Images