A method for comprehensive measurement of three-dimensional topography and temperature field of high-temperature target

Abstract

The application discloses a kind of high-temperature target three-dimensional topography and temperature field comprehensive measurement method, comprising: utilize binocular structured light gray image to obtain three-dimensional point cloud;High-resolution non-structured light gray image is carried out contrast enhancement, and enhanced image is obtained;Threshold processing obtains threshold processing image;High-resolution gray image and color image are carried out edge detection, and profile drawing is obtained;Calculate transformation matrix;Color image is transformed by transformation matrix, and new high-resolution color image is generated;Temperature change is carried out colorimetric temperature calibration, and high-temperature object temperature image is obtained based on high-resolution color image;Obtain the high-temperature object point cloud data of fusion temperature information.The application can better reflect the three-dimensional information and temperature distribution of high-temperature object surface, directly observe the abnormal area of object surface, make up the deficiency that three-dimensional reconstruction in traditional high-temperature cannot obtain the three-dimensional topography information and temperature information of high-temperature object simultaneously, and it is difficult to directly judge material anomaly.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional topography measurement technology. More specifically, this invention relates to a method for comprehensively measuring the three-dimensional topography and temperature field of a high-temperature target. Background Technology

[0002] Morphological and temperature field measurements of high-temperature objects are crucial testing technologies in the aerospace equipment field. In aerospace, the unique operating environment of aircraft places extremely high demands on their high-temperature resistance. During actual operation, the wall temperature and pressure of aircraft engines rise significantly due to high-intensity combustion. Hot-end components such as combustion chambers and nozzles must withstand thermo-mechanical coupling fields at temperatures of 2000°C or even higher, making them prone to material deformation. However, even minute deformations of materials under high-temperature conditions can affect engine performance and even damage the overall engine system structure, leading to aircraft malfunctions or crashes. Therefore, ensuring the safety and stability of aircraft necessitates precise measurement and analysis of the thermal deformation of aircraft materials and components.

[0003] Traditional contact measurement methods are not well-suited for the aforementioned scenarios. Currently, non-contact measurement methods are widely used across various industries. Among these, laser triangulation, eddy current methods, ultrasonic measurement, and machine vision measurement are commonly employed. Binocular vision measurement technology is a crucial component of machine vision measurement. This technology mimics the imaging principles of the human eye to perform 3D measurement, processing the image to obtain a disparity map, and then calculating the 3D coordinates of the object's surface using system parameters. However, this method has relatively low reconstruction accuracy. To obtain higher-precision 3D models, structured light is often used for 3D reconstruction. This method emits an coded light beam onto a smooth, featureless target surface, and then uses an algorithm to calculate the distortion of the returned coded pattern to obtain the object's position and depth information, thus achieving 3D model reconstruction.

[0004] However, traditional binocular structured light measurement systems can only obtain depth information of objects during 3D reconstruction, meaning they only contain the 3D shape information of the object's surface. Because they lack the ability to measure temperature information under high-temperature conditions, the reconstructed 3D model of a high-temperature object does not include the temperature information at the time of capture; it is merely a 3D spatial coordinate point with geometric position information. Temperature information, as a crucial characteristic under high-temperature conditions, is of great significance in the measurement of high-temperature objects. In the measurement of high-temperature objects, high-quality 3D point clouds with temperature information, compared to single coordinate point cloud data, can better reflect the current temperature distribution on the object's surface. The temperature distribution provides a direct understanding of the material's current state. A uniform, undeformed object should have a uniform surface temperature distribution after being heated. If the temperature distribution is uneven, or even has abnormally convex areas, it indicates that the material has deformed or even been damaged, resulting in structural damage. Meanwhile, when acquiring images of the object under test in a high-temperature environment, the acquired images will generate a lot of noise, and the grayscale difference between the background area and the test piece itself is very small. This will lead to a poor fusion effect and low fusion efficiency when fusion of temperature information. Therefore, it is necessary to process the image first to remove the interference of background and noise before fusion of temperature information. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a high-precision point cloud reconstruction method for high-temperature objects based on low-resolution temperature images in high-temperature environments. This method enables non-contact measurement of high-temperature objects in high-temperature environments while integrating the surface temperature information of the high-temperature objects, providing a more intuitive understanding of their surface morphology and corresponding position temperature information.

[0006] To achieve these objectives and other advantages according to the present invention, a method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target is provided, comprising the following steps:

[0007] Step 1: Acquire color images during the heating process. Calib and its corresponding temperature data T; when heated to the specified test temperature, acquire a color image I under non-illuminated conditions. Color High-resolution grayscale image under non-illuminated conditions and Binocular structured light grayscale image under active illumination and

[0008] Step 2: Obtain the 3D point cloud of the high-temperature object surface using binocular structured light grayscale images. w ;

[0009] Step 3: Select a high-resolution grayscale image captured by the grayscale camera on the same side as the color camera under non-light conditions. A contrast-limited adaptive histogram equalization algorithm is used to process the high-resolution grayscale left image. Contrast enhancement is performed to obtain an enhanced image.

[0010] Step 4: After contrast enhancement of the grayscale image under non-illuminated conditions, the grayscale difference between the background and the high-temperature object is amplified. Next, a filtering threshold θ is set. thr To separate the background and the high-temperature object, a thresholding process is applied to the enhanced image to obtain the thresholded image.

[0011] Step 5: Process the high-resolution grayscale left image. and color image I Color Perform edge detection to obtain a contour map. and color images

[0012] Step 6: Calculate the contour image To contour image The transformation matrix H g ;

[0013] Step 7: Transform the matrix H g For color image I Color The transformation is performed, and a new high-resolution color image is generated using an image interpolation algorithm.

[0014] Step 8: Utilize the color images I acquired at different times during the heating process in Step 1. Calib Based on its temperature data T, colorimetric thermometry calibration is completed;

[0015] Step 9: Using the calibrated colorimetric temperature correspondence, convert the high-resolution color image... Convert to temperature image

[0016] Step 10: Generate 3D point cloud data of the surface of the high-temperature object. w The surface temperature information of the point cloud is assigned to obtain the point cloud data of the high-temperature object. Temp .

[0017] Preferably, the specific method for acquiring image data of the surface of a high-temperature object in step one includes:

[0018] A data acquisition system is constructed, which includes two high-resolution grayscale industrial cameras, one color industrial camera, and one non-contact pyrometer.

[0019] The object under test is heated, and as the temperature rises to the specified test temperature, the temperature change T at a fixed location on the object's surface is monitored using a non-contact pyrometer. Simultaneously, multiple color images I at different temperature intervals are captured using a color industrial camera. Calib Record and store different temperatures T and their corresponding color images I Calib Used for subsequent colorimetric temperature measurement calibration;

[0020] After the object under test is heated to the specified test temperature, structured light is actively projected onto the surface of the high-temperature object using the optomechanical system of the binocular structured light system. Simultaneously, two high-resolution grayscale cameras acquire grayscale images of the binocular structured light under active illumination, which are used for subsequent calculation of the 3D point cloud. Once the optomechanical system stops projecting light, a grayscale camera positioned on the same side as the color camera acquires a high-resolution grayscale image of the high-temperature object. Gray Color images of high-temperature objects under non-illuminated conditions were acquired using a color camera. Color .

[0021] Preferably, in step two, a three-dimensional point cloud of the high-temperature object surface is acquired. w and images of the surface temperature of high-temperature objects Specific methods include:

[0022] A high-resolution grayscale left camera in a binocular structured light system is used to acquire a grayscale image of the target under illumination. The pixel value of a pixel is represented as (x LD ,y LD () indicates a high-resolution grayscale left image The xth LD line, y LD Column coordinates, M LD and N LD This refers to the image size of the high-resolution grayscale left image, i.e., the high-resolution grayscale left image. The total number of pixels is M LD ×N LD The system uses a high-resolution grayscale right camera to acquire a grayscale image of the target under illumination. The pixel value of a pixel is represented as (x RD ,y RD () indicates a high-resolution grayscale image. The xth RD line, y RD Column coordinates, M RD and N RDThis refers to the image size of the high-resolution grayscale right image, i.e., the high-resolution grayscale right image. The total number of pixels is M RD ×N RD Finally, based on the principle of active structured light binocular vision, a 3D point cloud of the surface of a high-temperature object was obtained. w In a 3D point cloud, each point is represented as a PointCloud. w (xS,yS,z S S = 0, 1, ..., PN-1, (x s ,y s ,z s ) represents the three-dimensional coordinates of each point, s represents the index value of that point, and PN represents the number of points in the three-dimensional point cloud.

[0023] Preferably, in step three, a high-resolution grayscale image under non-illuminated conditions is selected, and a contrast-limited adaptive histogram equalization algorithm is used to enhance the contrast of the high-resolution grayscale left image to obtain an enhanced image. Specific methods include:

[0024] High-resolution grayscale images under non-illuminated conditions and Select the grayscale image from the same side as the color camera. If the color camera is located on the left, then select the grayscale image captured by the left-side grayscale camera. left image with high resolution grayscale For example, the pixel value is represented as (x L ,y L () indicates a high-resolution grayscale left image The xth L line, y L Column coordinates, M L and N L This refers to the image size of the high-resolution grayscale left image, i.e., the high-resolution grayscale left image. The total number of pixels is M L ×N L ;

[0025] S3.1, Enhance the high-resolution grayscale image to be enhanced. The image is divided into M equal parts by manually setting the number of parts. p ×N p Divide the row into M small blocks, that is, divide the row into M equal parts. p Divide the column into N equal parts. p If a row or column contains elements that are not divisible by M, then... p and N pIn this case, the rows and columns are padded below and to the right of the original image to be divisible by M. p and N p The pixel values ​​of the supplemented pixels are obtained by mirroring the boundaries of the original image; the padded image is denoted as... The pixel value of its pixel is represented as M Com N represents the number of rows in the image after pixel padding. Com This represents the number of columns in the image after pixel padding; each pixel block is denoted as... R represents in the image In the image, the row index of the pixel block is C, where C represents the row index of the pixel block. In the table, the column index of the pixel block is used, and the pixel values ​​within the pixel block are denoted as... Each pixel block has M Com N Com / M p N p Each pixel is a separate pixel block that does not overlap with each other.

[0026] S3.2 Calculate each pixel block grayscale histogram Gray-scale histogram The horizontal axis represents the grayscale value, ranging from [0, 255], and the vertical axis represents the number of pixels at the current grayscale value. The combination of the horizontal and vertical axes is called a bin, or grayscale histogram. There are 256 bins, and the size of the corresponding y-axis represents the size of that bin. The size of each bin in the grayscale histogram of each pixel block is denoted as... The index representing the bin;

[0027] S3.3, Manually set the grayscale limit threshold δ, using each pixel block The average number of pixels per bin is obtained by dividing the number of pixels in the grayscale histogram by the number of bins (256). If the grayscale level in the region is relatively uniform, the number of pixels corresponding to the grayscale level in each bin should be equal. Therefore, the portion that needs to be cropped is the portion exceeding the grayscale level in each bin. The size of the portion exceeding the limit is denoted as . Will All excess portions are summed, and the sum is denoted as Sum. RC Sum RC Distribute evenly among each bin, with a value of Sum. RC / 256, generating a new grayscale histogram. The size of each bin in this histogram is denoted as . The index representing the bin;

[0028] S3.4, For each pixel block The resulting new grayscale histogram Histogram equalization is performed using the mapping function in Formula 1 to obtain the mapped value corresponding to each grayscale value, which is then used to replace the original grayscale value; where... This represents the new value obtained after mapping the current grayscale value through this function, where N represents the number of pixel blocks. The number of pixels, N=M Com N Com / M p N p ;

[0029]

[0030] S3.5, Process each pixel block through steps S3.1-S3.4. After processing, if each pixel block If pixels in a pixel block are transformed only through the mapping function within that block, it will result in a blocky effect in the final image, with obvious gaps between pixel blocks. Therefore, to solve this problem, when calculating the pixel value of each pixel in each block, the pixel value is obtained through a bilinear interpolation algorithm, ultimately resulting in an enhanced image.

[0031] For pixel point O, let its pixel value be O(m,n). Subtract half of the pixel block height and half of the pixel block width from the x and y coordinates of this point to obtain the mapped pixel point OC, whose coordinates are... Pixel OC may not be within the image range; assuming the pixel block containing pixel OC is... At this point, the ranges of R and C are respectively R∈[-1,M] p -1],C∈[-1,N p -1], find its three adjacent pixel blocks, and denote them as follows: If R = -1 or C = -1, it means the current pixel block does not exist; it is a virtual pixel block and cannot be used as an associated pixel block. Therefore, values ​​less than 0 are set to 0. When R = -1 and C = -1, the four associated pixel blocks are: and only If the first pixel is an actual existing pixel block, and the portions of the other three pixel blocks that are less than 0 are all set to 0, then we have: Therefore, only one pixel block actually affects the pixel value of the current pixel point O; then, the mapping function of the associated pixel block is used to map the pixel value O(m,n) to obtain four mapped pixel values. and

[0032] After obtaining the four mapped pixel values ​​of pixel value O(m,n) in the associated pixel block, the mapped pixel values ​​need to be weighted according to a certain ratio to obtain the final pixel value OF(m,n). The weighting coefficients are determined by bilinear interpolation. First, the distances from the mapped pixel point OC to the top, bottom, left, and right boundaries of the pixel block are calculated, denoted as y1, y2, x1, and x2, respectively. Then, these four values ​​are divided by the height and width of the current pixel block to obtain the weighting coefficients. in The final pixel value OF(m,n) is obtained by weighting the pixel values ​​using the following formula;

[0033]

[0034] Preferably, in step four, a filtering threshold θ is set. thr To separate the background and the high-temperature object, the following formula is used to determine whether the grayscale value meets the requirements:

[0035]

[0036] This preserves the specimen while removing the background, resulting in a thresholded image. The pixel values ​​of an image can be represented as The image resolution remains unchanged from before processing.

[0037] Preferably, in step six, the contour image is calculated. To contour image The transformation matrix H g Specific methods include:

[0038] S6.1. Use the SURF algorithm to process the contour map. and Feature points are extracted and described to obtain feature point sets. and the corresponding feature descriptor subset The number of elements in different feature point sets may not be the same, but the number of elements in the feature descriptor subset corresponding to each feature point set remains the same.

[0039]

[0040]

[0041]

[0042]

[0043] in, Representative outline diagram The extracted feature points, i = 0, ..., M fg -1, M fg Representative outline diagram The number of feature points extracted; Representative outline diagram The extracted feature points, j = 0,...,M ft -1, M ft Representative outline diagram The number of feature points extracted; Representative outline diagram The feature descriptors corresponding to the feature points extracted; Representative outline diagram The feature descriptor corresponding to the feature point extracted is a 128-dimensional feature vector.

[0044] S6.2, By analyzing the obtained feature descriptor subset and Perform matching to obtain the set of matched feature point pairs Mate. rough ;

[0045] For contour images and contour images Feature descriptors in and The similarity between feature vectors is measured using the Euclidean distance; the smaller the Euclidean distance, the more similar the two vectors. The feature descriptor is calculated using the following formula. and feature descriptors The Euclidean distance between two vectors is calculated using the expression Dist().

[0046]

[0047] Find the contour image Each feature descriptor With contour image Feature descriptor subset All feature descriptors Euclidean distance, taken from With minimum Euclidean distance As its matching point; however, a one-to-many problem arises when calculating the matching point, for feature descriptors In the feature descriptor subset We can find multiple descriptors with the smallest Euclidean distance that are equal to each other. Therefore, we sort the calculated Euclidean distances from smallest to largest, and select the feature points corresponding to the two feature descriptors with the first and second highest Euclidean distances as the feature points to be matched. The feature point with the highest Euclidean distance is called the nearest neighbor feature point. With feature descriptors The distance is represented as dist Near The feature point ranked second is called the second nearest neighbor feature point. With feature descriptors The distance is represented as dist NNear Where Near represents the nearest neighbor feature point; In the set of feature points In the index, NNear represents the second nearest neighbor feature point. In the set of feature points The index in the middle; the nearest neighbor feature point is determined by the ratio of the nearest neighbor Euclidean distance to the second nearest neighbor Euclidean distance. Whether it is a matching feature point; if the ratio equals 1, it indicates a mismatch and is directly excluded because the matching points of this feature point are not unique; a manually set ratio, ratio∈(0,1), if dist Near / dist NNear If the ratio is less than the feature point, then the feature point is considered to be... These are matching feature points that meet the conditions. Pair these matching points... Add to the matching point pair set Mate rough ; for feature descriptor subsets Perform the above operation on all descriptors to obtain the final set of matching point pairs:

[0048]

[0049] Where 'a' represents the contour image. The index of the feature point in the feature point set, where b represents the contour image. The index of the feature point pair in the feature point set;

[0050] S6.3. Manually assist in filtering the set of matching point pairs, and use the RANSAC algorithm to eliminate mismatches to obtain the final transformation matrix H. g ;

[0051] For the obtained set of matching point pairs Mate rough If the transformation matrix is ​​calculated directly, the result will have a large error due to the presence of many mismatched feature point pairs. Therefore, manual assistance is needed in the experiment to filter feature point pairs and obtain a more accurate set of matching point pairs, Mate. arti :

[0052]

[0053] Where k represents the contour image feature point set The index of the feature point, v represents the contour image. feature point set The index of the feature point is determined; after manual filtering, since manual filtering also has errors, the final transformation matrix H needs to be accurately estimated using the Random Consensus Algorithm (RANSAC). g ;

[0054] Transformation matrix H g It is a 3×3 matrix, which can be represented as:

[0055]

[0056] Each matched feature point pair should, in principle, satisfy the following equation:

[0057]

[0058] in Mate represents the set of matching point pairs. arti Mid-contour image Corresponding feature points x-coordinate Mate represents the set of matching point pairs. arti Mid-contour image Corresponding feature points The ordinate; Mate represents the set of matching point pairs. arti Mid-contour image Corresponding feature points x-coordinate Mate represents the set of matching point pairs. arti Mid-contour image Corresponding feature points The ordinate, ω, represents the scale parameter;

[0059] For matrix H g In this context, it has 8 unknown parameters and requires at least 8 linear equations to solve. Therefore, at least 4 matching point pairs are needed for calculation at a time. The RANSAC algorithm is used here to calculate the optimal transformation matrix model H using multiple matching point pairs. g Specifically, it includes:

[0060] S6.3.1, First, start with the matching point pair set Mate arti Four feature point pairs are randomly selected from the data to calculate the matrix model, denoted as... Where i = 0, 1, ..., iter-1, i represents the (i+1)th iteration, iter represents the total number of iterations, and the initial value of iter is set manually;

[0061] S6.3.2 Then, calculate the cost value of all matching point pairs in the current iteration number according to the cost function of the following formula:

[0062]

[0063] Compared to a manually set cost threshold thre In comparison, if Then add the matching point pair to the current inner point set Inner. i Otherwise, add the matching point pair to the current outer point set Outer. i ;

[0064] S6.3.3, Statistical Inner i The number of interior points is denoted as Number. i With the optimal inner set of points Inner best Number of interior points best In comparison, if Number i >Number best Then update the optimal set of inliers to the current set of inliers. best =Inner i Number best =Number i Update the optimal model H g For the currently calculated model, Simultaneously update i and the iteration count iter, i = i + 1. Where w i represents the proportion of interior points to total points in the (i+1)th iteration, m is the minimum number of samples required to calculate the model, and p is the confidence level, where m = 4 and p = 0.995;

[0065] S6.3.4 Finally, iterate through steps S631-S634 until i > iter, at which point the iteration ends; at this point, the optimal inner point set Inner is determined. best The matching point pairs in the set are the final set of precise feature point pairs, denoted as Mate. accurate H calculated in this way g The optimal transformation matrix is:

[0066]

[0067] Where c represents the contour image after error matching is corrected. feature point set The index of the feature point is 'b', where 'b' represents the contour image after error correction. feature point set The index of the feature point.

[0068] Preferably, in step seven, the transformation matrix H is used... g For color image I Color The transformation is performed, and a new high-resolution color image is generated using an image interpolation algorithm. Specific methods include:

[0069] S7.1 If color image I Color The resolution is lower than that of the grayscale left image The resolution, for color images I Color and high-resolution grayscale left image In other words, it can be achieved through the transformation matrix H g Color image I Color Mapped to high-resolution grayscale left image In the coordinate system in which it is located, therefore the color image I Color The coordinates before and after the transformation can be represented as:

[0070]

[0071] Among them, u new and v new Indicates the left image in high-resolution grayscale. Color image I in the coordinate system Color The corresponding coordinates, ω = h 31 u t +h 32 v t +1;

[0072] As can be seen from the above formula, there is a one-to-one correspondence between the two coordinates. (Low-resolution color image I) Color M t ×N t Each pixel corresponds to a high-resolution grayscale left image. There should also be an M in the coordinate system in which it is located. t ×N t Each pixel represents a single point cloud on the surface of a high-temperature object. However, when assigning temperature information later, the reconstructed point cloud data is obtained from a high-resolution grayscale image. The result obtained through forward mapping is only M. t ×N t To ensure a one-to-one correspondence between the reconstructed point cloud data and the temperature information, the temperature information of each pixel needs to be obtained from the low-resolution color image I. Color and high-resolution grayscale left image The resolution is uniformly set to M.L ×N L This yields a high-resolution color image, which can then be used to modify the low-resolution color image I through inverse mapping and interpolation. Color Process it;

[0073] Through formula We can obtain:

[0074]

[0075] Simultaneously, left multiplication get:

[0076]

[0077] make:

[0078]

[0079] Expanding, we get the inverse mapping formula:

[0080]

[0081] The above formula yields the inverse mapping from high-resolution to low-resolution images, where (u h ,v h () represents the pixel coordinates of the high-resolution image. The coordinate value must be an integer. l ,v l The coordinates represent the corresponding values ​​of the high-resolution image coordinates in the low-resolution image coordinate system. These coordinate values ​​may be floating-point numbers. Although this formula is also a one-to-one correspondence, when each pixel of the high-resolution image is mapped to the low-resolution image, an out-of-bounds check can be performed. If the pixel exceeds the bounds, its pixel value is padded to (0,0,0), and finally a high-resolution image can be obtained. For the case where the coordinates of the mapped point are floating-point numbers, their pixel values ​​can be obtained through interpolation algorithms.

[0082] Therefore, first create a blank image with the same size as the high-resolution grayscale left image. The pixel value of a pixel is represented as (x Reg ,y Reg ) represents a blank image The xth Reg line, y Reg The column coordinates are obtained using the following formula:

[0083]

[0084] Among them, (x New ,y New () represents a blank image The coordinates are transformed in the low-resolution color image Coordinates in a coordinate system may go out of bounds and cannot be used in low-resolution temperature images. Find the corresponding value above, and you need to manually set the pixel value. Also, the coordinate value of this coordinate may be a floating-point value, so further interpolation is needed to obtain the corresponding pixel value.

[0085] S7.2 Next, determine whether the converted image coordinates are out of bounds. If they are, fill the pixel value at that position with (0,0,0); otherwise, use bicubic interpolation to fill the blank image. Assign values ​​to RGB color information to generate a high-resolution color image. Specifically, it includes:

[0086] S7.2.1, regarding coordinates (x) New ,y New Perform an out-of-bounds check. If it is out of bounds, set the pixel value at that position to (0,0,0); otherwise, use bicubic interpolation to obtain the pixel value at that coordinate point.

[0087] S7.2.2, will (x New ,y New The 16 pixels within a 4×4 neighborhood are numbered according to their row and column, and each coordinate is related to (x) New ,y New By calculating the distances in the rows and columns respectively, we can construct separate row and column distance vectors:

[0088] R = [1 + uu + 1 - u + 2 - u]

[0089] C = [1 + v + v + 1 - v + 2 - v]

[0090] S7.2.3, In order to obtain the pixel value of each pixel in the 4×4 neighborhood, I Color (x New ,y New The influence weights are calculated by weighting the row and column distances of adjacent pixels using a cubic polynomial W(x), respectively, to obtain the relationship between adjacent pixels and I. Color (x New ,y New The row and column distance weight vector W R and W C , where a = -0.5, and [] indicates rounding down to the nearest integer;

[0091]

[0092] W R =[W(1+x New -[x New ]) W(xNew -[x New ]) W(1-x New +[x New ]) W(2-x New +[x New ])]

[0093] W C =[W(1+y New -[y New ]) W(y New -[y New ]) W(1-y New +[y New ]) W(2-y New +[y New ])]

[0094] S7.2.4, Obtain the pixel value at this coordinate by weighting. Among them W R W represents a row-weighted vector. C S represents a column-weighted vector. RC A pixel value matrix representing 16 pixels within a 4×4 neighborhood:

[0095]

[0096] Preferably, in step eight, the different temperatures T and their corresponding color images I obtained in step one are used. Calib The specific methods for conducting colorimetric temperature calibration include:

[0097] Using the color images I corresponding to different temperatures T obtained in step one Calib I was obtained through the principle of colorimetric thermometry. Calib The colorimetric thermometry method corresponds to the red and green channels of the pixels at the measurement points of the pyrometer. Based on the functional relationship between the ratio of the spectral radiance of an object at two wavelengths and the measured temperature, the temperature value is calculated through formula transformation and derivation. The colorimetric thermometry formula is as follows:

[0098]

[0099] Where C2 is the second radiation constant, λ g λ represents the wavelength corresponding to green light. r Where λ represents the wavelength corresponding to red light, R is the output value of the R channel of the color camera, G is the output value of the G channel of the color camera, and K is the spectral response coefficient of the camera. K g K represents the response coefficient of the G primary color channel of a color camera. r The R primary color channel response coefficient of the camera;

[0100] By transforming the above formula, we obtain the derived formula for the spectral response coefficient K to be calibrated:

[0101]

[0102] Based on the above formula, we substitute different temperatures T and color images I. Calib The measurement position corresponds to the pixel point Numerical values ​​can be used to calculate the corresponding K values ​​for different temperatures T. To find a more realistic relationship between the data, a cubic B-spline algorithm is used to fit the obtained data, yielding the calibration parameter K and the corresponding values ​​at the measurement point temperature T. The curve showing the fitting relationship between the numerical values.

[0103] Preferably, the full-field temperature image of the high-temperature object surface is obtained using the calibrated fitting curve. In step nine, the specific methods include:

[0104] By using the fitted relationship curve, each pixel point is obtained. The calibration parameter K corresponding to the value can be used to calculate the mapped temperature of the pixel through the derivation formula of the spectral response coefficient K. By traversing the entire image, the temperature matrix corresponding to the entire color image can be obtained.

[0105] After calculating the temperature corresponding to each pixel, it is assigned to the color image I. Color Each pixel is analyzed to obtain a temperature image containing temperature information. Where the pixel value is represented as (u t ,v t ),u t =0,...,M t -1,v t =0,...,N t -1, (u t ,v t () represents a temperature image The uth t line, v t Column coordinates, M t and N t The image size represents the temperature image. The total number of pixels is M t ×N t .

[0106] Preferably, in step ten, the three-dimensional point cloud data of the high-temperature object surface is processed. w The surface temperature information of the point cloud is assigned to obtain the point cloud data of the high-temperature object.Temp Specific methods include:

[0107] Based on the imaging model of a stereo camera, the 3D point cloud data of the high-temperature object surface is generated. w The point coordinates in the high-resolution grayscale left image The pixel coordinates are mapped one-to-one, by using high-resolution temperature images The temperature information corresponding to each pixel is assigned to the 3D point cloud data of the high-temperature object surface (PointCloud). w PointCloud data of high-temperature objects was obtained. Temp In a 3D point cloud, each point is represented as a PointCloud. Temp (x T ,y T ,z T ,r T ,g T ,b T ),T=0,1,...,PN-1,(x T ,y T ,z T (r) represents the three-dimensional spatial coordinates of each point. T ,g T ,b T ) represents the temperature information carried by each point, T represents the index value of that point, and PN represents the number of points in the 3D point cloud.

[0108] The present invention has at least the following beneficial effects:

[0109] (1) This invention uses a contrast enhancement algorithm to enhance high-resolution grayscale images acquired under high-temperature conditions. Processing is performed. To preserve image detail and suppress noise, a contrast-limited adaptive histogram equalization algorithm is selected. Enhancement was performed to address the problem of difficulty in separating the background and specimen due to small grayscale differences in the image. The grayscale difference between the two was increased, resulting in an enhanced image. Using a thresholding algorithm to enhance images The background and specimen are separated to obtain a thresholded image containing only the specimen. This reduces the impact of cluttered backgrounds on subsequent edge extraction.

[0110] (2) This invention uses an edge extraction algorithm to process low-resolution temperature images. and thresholding images Edge extraction was performed. To obtain more accurate edge features, the Canny edge extraction algorithm was selected to extract the edges of the specimen from the two images, resulting in a contour image. and By contour image and Feature extraction and matching are performed on the above, which improves the efficiency and accuracy of subsequent feature point extraction and matching.

[0111] (3) This invention uses a binocular structured light system to obtain a three-dimensional point cloud on the surface of a high-temperature object. w Low-resolution temperature images of object surfaces under high-temperature conditions are acquired using a temperature camera. PointCloud data of high-temperature objects is obtained by combining the three-dimensional shape information and temperature information of objects under high-temperature conditions. Temp This invention can better reflect the morphological information and current temperature distribution of high-temperature objects, and intuitively observe abnormal areas on the object's surface. It makes up for the shortcomings of traditional high-temperature 3D reconstruction, which cannot simultaneously obtain the 3D morphological information and temperature information of high-temperature objects, and is difficult to intuitively judge material anomalies.

[0112] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0113] Figure 1 This is a flowchart of a specific implementation of the method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target according to the present invention;

[0114] Figure 2 This is a schematic diagram for acquiring image data of the surface of a high-temperature object;

[0115] Figure 3 This is a result image of the point cloud data acquired on the surface of a high-temperature object;

[0116] Figure 4 It is a high-resolution grayscale image obtained by the high-resolution grayscale left camera in a binocular structured light system;

[0117] Figure 5 It is a low-resolution temperature image acquired by a temperature camera;

[0118] Figure 6 This is a diagram showing the contrast enhancement result of a high-resolution grayscale image;

[0119] Figure 7 This is a diagram showing the results of thresholding a high-resolution grayscale image;

[0120] Figure 8 This is the image contour extraction result of the thresholding process;

[0121] Figure 9 This is a low-resolution temperature image contour extraction result image;

[0122] Figure 10 This is a high-resolution temperature image result after registration and interpolation;

[0123] Figure 11 It is the final point cloud data of a high-temperature object obtained by fusing temperature information into the point cloud data of the surface of the high-temperature object; Detailed Implementation

[0124] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0125] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0126] like Figure 1 As shown: A method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target, comprising the following steps:

[0127] Step 1: Acquire color images during the heating process. Calib and its corresponding temperature data T; when heated to the specified test temperature, acquire a color image I under non-illuminated conditions. Color High-resolution grayscale image under non-illuminated conditions and Binocular structured light grayscale image under active illumination and

[0128] Step 2: Obtain the 3D point cloud of the high-temperature object surface using binocular structured light grayscale images. w ;

[0129] Step 3: Select a high-resolution grayscale image captured by the grayscale camera on the same side as the color camera under non-light conditions. A contrast-limited adaptive histogram equalization algorithm is used to process the high-resolution grayscale left image. Contrast enhancement is performed to obtain an enhanced image.

[0130] Step 4: After contrast enhancement of the grayscale image under non-illuminated conditions, the grayscale difference between the background and the high-temperature object is amplified. Next, a filtering threshold θ is set. thr To separate the background and the high-temperature object, a thresholding process is applied to the enhanced image to obtain the thresholded image.

[0131] Step 5: Process the high-resolution grayscale left image. and color image I Color Perform edge detection to obtain a contour map. and color images

[0132] Step 6: Calculate the contour image To contour image The transformation matrix H g ;

[0133] Step 7: Transform the matrix H g For color image I Color The transformation is performed, and a new high-resolution color image is generated using an image interpolation algorithm.

[0134] Step 8: Utilize the color images I acquired at different times during the heating process in Step 1. Calib Based on its temperature data T, colorimetric thermometry calibration is completed;

[0135] Step 9: Using the calibrated colorimetric temperature correspondence, convert the high-resolution color image... Convert to temperature image

[0136] Step 10: Generate 3D point cloud data of the surface of the high-temperature object. w The surface temperature information of the point cloud is assigned to obtain the point cloud data of the high-temperature object. Temp .

[0137] In the above technical solution, the specific method for obtaining high-temperature object surface image data in step one includes:

[0138] A data acquisition system is constructed, which includes two high-resolution grayscale industrial cameras, one color industrial camera, and one non-contact pyrometer.

[0139] The object under test is heated, and as the temperature rises to the specified test temperature, the temperature change T at a fixed location on the object's surface is monitored using a non-contact pyrometer. Simultaneously, multiple color images I at different temperature intervals are captured using a color industrial camera. Calib Record and store different temperatures T and their corresponding color images I Calib Used for subsequent colorimetric temperature measurement calibration;

[0140] After the object under test is heated to the specified test temperature, structured light is actively projected onto the surface of the high-temperature object using the optomechanical system of the binocular structured light system. Simultaneously, two high-resolution grayscale cameras acquire grayscale images of the binocular structured light under active illumination, which are used for subsequent calculation of the 3D point cloud. Once the optomechanical system stops projecting light, a grayscale camera positioned on the same side as the color camera acquires a high-resolution grayscale image of the high-temperature object. GrayColor images of high-temperature objects under non-illuminated conditions were acquired using a color camera. Color .

[0141] In the above technical solution, step two involves acquiring a three-dimensional point cloud of the high-temperature object surface. w and images of the surface temperature of high-temperature objects Specific methods include:

[0142] A high-resolution grayscale left camera in a binocular structured light system is used to acquire a grayscale image of the target under illumination. The pixel value of a pixel is represented as (x LD ,y LD () indicates a high-resolution grayscale left image The xth LD line, y LD Column coordinates, M LD and N LD This refers to the image size of the high-resolution grayscale left image, i.e., the high-resolution grayscale left image. The total number of pixels is M LD ×N LD The system uses a high-resolution grayscale right camera to acquire a grayscale image of the target under illumination. The pixel value of a pixel is represented as (x RD ,y RD () indicates a high-resolution grayscale image. The xth RD line, y RD Column coordinates, M RD and N RD This refers to the image size of the high-resolution grayscale right image, i.e., the high-resolution grayscale right image. The total number of pixels is M RD ×N RD Finally, based on the principle of active structured light binocular vision, a 3D point cloud of the surface of a high-temperature object was obtained. w In a 3D point cloud, each point is represented as a PointCloud. w (x S ,y S ,z S S = 0, 1, ..., PN-1, (x s ,y s ,z s ) represents the three-dimensional coordinates of each point, s represents the index value of that point, and PN represents the number of points in the three-dimensional point cloud.

[0143] In the above technical solution, in step three, a high-resolution grayscale image under non-illuminated conditions is selected, and a contrast-limited adaptive histogram equalization algorithm is used to enhance the contrast of the high-resolution grayscale left image to obtain an enhanced image. Specific methods include:

[0144] High-resolution grayscale images under non-illuminated conditions and Select the grayscale image from the same side as the color camera. If the color camera is located on the left, then select the grayscale image captured by the left-side grayscale camera. left image with high resolution grayscale For example, the pixel value is represented as (x L ,y L () indicates a high-resolution grayscale left image The xth L line, y L Column coordinates, M L and N L This refers to the image size of the high-resolution grayscale left image, i.e., the high-resolution grayscale left image. The total number of pixels is M L ×N L ;

[0145] S3.1, Enhance the high-resolution grayscale image to be enhanced. The image is divided into M equal parts by manually setting the number of parts. p ×N p Divide the row into M small blocks, that is, divide the row into M equal parts. p Divide the column into N equal parts. p If a row or column contains elements that are not divisible by M, then... p and N p In this case, the rows and columns are padded below and to the right of the original image to be divisible by M. p and N p The pixel values ​​of the supplemented pixels are obtained by mirroring the boundaries of the original image; the padded image is denoted as... The pixel value of its pixel is represented as M Com N represents the number of rows in the image after pixel padding. Com This represents the number of columns in the image after pixel padding; each pixel block is denoted as... R represents in the image In the image, the row index of the pixel block is C, where C represents the row index of the pixel block. In the table, the column index of the pixel block is used, and the pixel values ​​within the pixel block are denoted as... Each pixel block has MCom N Com / M p N p Each pixel is a separate pixel block that does not overlap with each other.

[0146] S3.2 Calculate each pixel block grayscale histogram Gray-scale histogram The horizontal axis represents the grayscale value, ranging from [0, 255], and the vertical axis represents the number of pixels at the current grayscale value. The combination of the horizontal and vertical axes is called a bin, or grayscale histogram. There are 256 bins, and the size of the corresponding y-axis represents the size of that bin. The size of each bin in the grayscale histogram of each pixel block is denoted as... The index representing the bin;

[0147] S3.3, Manually set the grayscale limit threshold δ, using each pixel block The average number of pixels per bin is obtained by dividing the number of pixels in the grayscale histogram by the number of bins (256). If the grayscale level in the region is relatively uniform, the number of pixels corresponding to the grayscale level in each bin should be equal. Therefore, the portion that needs to be cropped is the portion exceeding the grayscale level in each bin. The size of the portion exceeding the limit is denoted as . Will All excess portions are summed, and the sum is denoted as Sum. RC Sum RC Distribute evenly among each bin, with a value of Sum. RC / 256, generating a new grayscale histogram. The size of each bin in this histogram is denoted as . num = 0, 1, ..., 255 represents the index of this bin;

[0148] S3.4, For each pixel block The resulting new grayscale histogram Histogram equalization is performed using the mapping function in Formula 1 to obtain the mapped value corresponding to each grayscale value, which is then used to replace the original grayscale value; where... This represents the new value obtained after mapping the current grayscale value through this function, where N represents the number of pixel blocks. The number of pixels, N=M Com N Com / M p N p ;

[0149]

[0150] S3.5, Process each pixel block through steps S3.1-S3.4. After processing, if each pixel block If pixels in a pixel block are transformed only through the mapping function within that block, it will result in a blocky effect in the final image, with obvious gaps between pixel blocks. Therefore, to solve this problem, when calculating the pixel value of each pixel in each block, the pixel value is obtained through a bilinear interpolation algorithm, ultimately resulting in an enhanced image.

[0151] For pixel point O, let its pixel value be O(m,n). Subtract half of the pixel block height and half of the pixel block width from the x and y coordinates of this point to obtain the mapped pixel point OC, whose coordinates are... Pixel OC may not be within the image range; assuming the pixel block containing pixel OC is... At this point, the ranges of R and C are respectively R∈[-1,M] p -1],C∈[-1,N p -1], find its three adjacent pixel blocks, and denote them as follows: If R = -1 or C = -1, it means the current pixel block does not exist; it is a virtual pixel block and cannot be used as an associated pixel block. Therefore, values ​​less than 0 are set to 0. When R = -1 and C = -1, the four associated pixel blocks are: and only If the first pixel is an actual existing pixel block, and the portions of the other three pixel blocks that are less than 0 are all set to 0, then we have: Therefore, only one pixel block actually affects the pixel value of the current pixel point O; then, the mapping function of the associated pixel block is used to map the pixel value O(m,n) to obtain four mapped pixel values. and

[0152] After obtaining the four mapped pixel values ​​of pixel value O(m,n) in the associated pixel block, the mapped pixel values ​​need to be weighted according to a certain ratio to obtain the final pixel value OF(m,n). The weighting coefficients are determined by bilinear interpolation. First, the distances from the mapped pixel point OC to the top, bottom, left, and right boundaries of the pixel block are calculated, denoted as y1, y2, x1, and x2, respectively. Then, these four values ​​are divided by the height and width of the current pixel block to obtain the weighting coefficients. in The final pixel value OF(m,n) is obtained by weighting the pixel values ​​using the following formula;

[0153]

[0154] In the above technical solution, in step four, a filtering threshold θ is set. thr To separate the background and the high-temperature object, the following formula is used to determine whether the grayscale value meets the requirements:

[0155]

[0156] This preserves the specimen while removing the background, resulting in a thresholded image. The pixel values ​​of an image can be represented as The image resolution remains unchanged from before processing.

[0157] In the above technical solution, in step six, the contour image is calculated. To contour image The transformation matrix H g Specific methods include:

[0158] S6.1. Use the SURF algorithm to process the contour map. and Feature points are extracted and described to obtain feature point sets. and the corresponding feature descriptor subset The number of elements in different feature point sets may not be the same, but the number of elements in the feature descriptor subset corresponding to each feature point set remains the same.

[0159]

[0160]

[0161]

[0162]

[0163] in, Representative outline diagram The extracted feature points, i = 0, ..., M fg -1, M fg Representative outline diagram The number of feature points extracted; Representative outline diagram The extracted feature points, j = 0,...,M ft -1, M ft Representative outline diagram The number of feature points extracted; Representative outline diagram The feature descriptors corresponding to the feature points extracted; Representative outline diagram The feature descriptor corresponding to the feature point extracted is a 128-dimensional feature vector.

[0164] S6.2, By analyzing the obtained feature descriptor subset and Perform matching to obtain the set of matched feature point pairs Mate. rough ;

[0165] For contour images and contour images Feature descriptors in and The similarity between feature vectors is measured using the Euclidean distance; the smaller the Euclidean distance, the more similar the two vectors. The feature descriptor is calculated using the following formula. and feature descriptors The Euclidean distance between two vectors is calculated using the expression Dist().

[0166]

[0167] Find the contour image Each feature descriptor With contour image Feature descriptor subset All feature descriptors Euclidean distance, taken from With minimum Euclidean distance As its matching point; however, a one-to-many problem arises when calculating the matching point, for feature descriptors In the feature descriptor subset We can find multiple descriptors with the smallest Euclidean distance that are equal to each other. Therefore, we sort the calculated Euclidean distances from smallest to largest, and select the feature points corresponding to the two feature descriptors with the first and second highest Euclidean distances as the feature points to be matched. The feature point with the highest Euclidean distance is called the nearest neighbor feature point. With feature descriptors The distance is represented as dist Near The feature point ranked second is called the second nearest neighbor feature point. With feature descriptors The distance is represented as dist NNear Where Near represents the nearest neighbor feature point; In the set of feature points In the index, NNear represents the second nearest neighbor feature point. In the set of feature points The index in the middle; the nearest neighbor feature point is determined by the ratio of the nearest neighbor Euclidean distance to the second nearest neighbor Euclidean distance. Whether it is a matching feature point; if the ratio equals 1, it indicates a mismatch and is directly excluded because the matching points of this feature point are not unique; a manually set ratio, ratio∈(0,1), if dist Near / dist NNear If the ratio is less than the feature point, then the feature point is considered to be... These are matching feature points that meet the conditions. Pair these matching points... Add to the matching point pair set Mate rough ; for feature descriptor subsets Perform the above operation on all descriptors to obtain the final set of matching point pairs:

[0168]

[0169] Where 'a' represents the contour image. The index of the feature point in the feature point set, where b represents the contour image. The index of the feature point pair in the feature point set;

[0170] S6.3. Manually assist in filtering the set of matching point pairs, and use the RANSAC algorithm to eliminate mismatches to obtain the final transformation matrix H. g ;

[0171] For the obtained set of matching point pairs Mate rough If the transformation matrix is ​​calculated directly, the result will have a large error due to the presence of many mismatched feature point pairs. Therefore, manual assistance is needed in the experiment to filter feature point pairs and obtain a more accurate set of matching point pairs, Mate. arti :

[0172]

[0173] Where k represents the contour image feature point set The index of the feature point, v represents the contour image. feature point set The index of the feature point is determined; after manual filtering, since manual filtering also has errors, the final transformation matrix H needs to be accurately estimated using the Random Consensus Algorithm (RANSAC). g ;

[0174] Transformation matrix H g It is a 3×3 matrix, which can be represented as:

[0175]

[0176] Each matched feature point pair should, in principle, satisfy the following equation:

[0177]

[0178] in Mate represents the set of matching point pairs. arti Mid-contour image Corresponding feature points x-coordinate Mate represents the set of matching point pairs. arti Mid-contour image Corresponding feature points The ordinate; Mate represents the set of matching point pairs. arti Mid-contour image Corresponding feature points x-coordinate Mate represents the set of matching point pairs. arti Mid-contour image Corresponding feature points The ordinate, ω, represents the scale parameter;

[0179] For matrix H g In this context, it has 8 unknown parameters and requires at least 8 linear equations to solve. Therefore, at least 4 matching point pairs are needed for calculation at a time. The RANSAC algorithm is used here to calculate the optimal transformation matrix model H using multiple matching point pairs. g Specifically, it includes:

[0180] S6.3.1, First, start with the matching point pair set Mate arti Four feature point pairs are randomly selected from the data to calculate the matrix model, denoted as... Where i = 0, 1, ..., iter-1, i represents the (i+1)th iteration, iter represents the total number of iterations, and the initial value of iter is set manually;

[0181] S6.3.2 Then, calculate the cost value of all matching point pairs in the current iteration number according to the cost function of the following formula:

[0182]

[0183] Compared to a manually set cost threshold thre In comparison, if Then add the matching point pair to the current inner point set Inner. i Otherwise, add the matching point pair to the current outer point set Outer. i ;

[0184] S6.3.3, Statistical Inner iThe number of interior points is denoted as Number. i With the optimal inner set of points Inner best Number of interior points best In comparison, if Number i >Number best Then update the optimal set of inliers to the current set of inliers. best =Inner i Number best =Number i Update the optimal model H g For the currently calculated model, Simultaneously update i and the iteration count iter, i = i + 1. Where w i represents the proportion of interior points to total points in the (i+1)th iteration, m is the minimum number of samples required to calculate the model, and p is the confidence level, where m = 4 and p = 0.995;

[0185] S6.3.4 Finally, iterate through steps S631-S634 until i > iter, at which point the iteration ends; at this point, the optimal inner point set Inner is determined. best The matching point pairs in the set are the final set of precise feature point pairs, denoted as Mate. accurate H calculated in this way g The optimal transformation matrix is:

[0186]

[0187] Where c represents the contour image after error matching is corrected. feature point set The index of the feature point is 'b', where 'b' represents the contour image after error correction. feature point set The index of the feature point.

[0188] In the above technical solution, in step seven, the transformation matrix H is used. g For color image I Color The transformation is performed, and a new high-resolution color image is generated using an image interpolation algorithm. Specific methods include:

[0189] S7.1 If color image I Color The resolution is lower than that of the grayscale left image The resolution, for color images I Color and high-resolution grayscale left image In other words, it can be achieved through the transformation matrix H g Color image I ColorMapped to high-resolution grayscale left image In the coordinate system in which it is located, therefore the color image I Color The coordinates before and after the transformation can be represented as:

[0190]

[0191] Among them, u new and v new Indicates the left image in high-resolution grayscale. Color image I in the coordinate system Color The corresponding coordinates, ω = h 31 u t +h 32 v t +1;

[0192] As can be seen from the above formula, there is a one-to-one correspondence between the two coordinates. (Low-resolution color image I) Color M t ×N t Each pixel corresponds to a high-resolution grayscale left image. There should also be an M in the coordinate system in which it is located. t ×N t Each pixel represents a single point cloud on the surface of a high-temperature object. However, when assigning temperature information later, the reconstructed point cloud data is obtained from a high-resolution grayscale image. The result obtained through forward mapping is only M. t ×N t To ensure a one-to-one correspondence between the reconstructed point cloud data and the temperature information, the temperature information of each pixel needs to be obtained from the low-resolution color image I. Color and high-resolution grayscale left image The resolution is uniformly set to M. L ×N L This yields a high-resolution color image, which can then be used to modify the low-resolution color image I through inverse mapping and interpolation. Color Process it;

[0193] Through formula We can obtain:

[0194]

[0195] Simultaneously, left multiplication get:

[0196]

[0197] make:

[0198]

[0199] Expanding, we get the inverse mapping formula:

[0200]

[0201] The above formula yields the inverse mapping from high-resolution to low-resolution images, where (u h ,v h () represents the pixel coordinates of the high-resolution image. The coordinate value must be an integer. l ,v l The coordinates represent the corresponding values ​​of the high-resolution image coordinates in the low-resolution image coordinate system. These coordinate values ​​may be floating-point numbers. Although this formula is also a one-to-one correspondence, when each pixel of the high-resolution image is mapped to the low-resolution image, an out-of-bounds check can be performed. If the pixel exceeds the bounds, its pixel value is padded to (0,0,0), and finally a high-resolution image can be obtained. For the case where the coordinates of the mapped point are floating-point numbers, their pixel values ​​can be obtained through interpolation algorithms.

[0202] Therefore, first create a blank image with the same size as the high-resolution grayscale left image. The pixel value of a pixel is represented as (x Reg ,y Reg ) represents a blank image The xth Reg line, y Reg The column coordinates are obtained using the following formula:

[0203]

[0204] Among them, (x New ,y New () represents a blank image The coordinates are transformed in the low-resolution color image Coordinates in a coordinate system may go out of bounds and cannot be used in low-resolution temperature images. Find the corresponding value above, and you need to manually set the pixel value. Also, the coordinate value of this coordinate may be a floating-point value, so further interpolation is needed to obtain the corresponding pixel value.

[0205] S7.2 Next, determine whether the converted image coordinates are out of bounds. If they are, fill the pixel value at that position with (0,0,0); otherwise, use bicubic interpolation to fill the blank image. Assign values ​​to RGB color information to generate a high-resolution color image. Specifically, it includes:

[0206] S7.2.1, regarding coordinates (x) New ,y NewPerform an out-of-bounds check. If it is out of bounds, set the pixel value at that position to (0,0,0); otherwise, use bicubic interpolation to obtain the pixel value at that coordinate point.

[0207] S7.2.2, will (x New ,y New The 16 pixels within a 4×4 neighborhood are numbered according to their row and column, and each coordinate is related to (x) New ,y New By calculating the distances in the rows and columns respectively, we can construct separate row and column distance vectors:

[0208] R = [1 + uu + 1 - u + 2 - u]

[0209] C = [1 + v + v + 1 - v + 2 - v]

[0210] S7.2.3, In order to obtain the pixel value of each pixel in the 4×4 neighborhood, I Color (x New ,y New The influence weights are calculated by weighting the row and column distances of adjacent pixels using a cubic polynomial W(x), respectively, to obtain the relationship between adjacent pixels and I. Color (x New ,y New The row and column distance weight vector W R and W C , where a = -0.5, and [] indicates rounding down to the nearest integer;

[0211]

[0212] W R =[W(1+x New -[x New ]) W(x New -[x New ]) W(1-x New +[x New ]) W(2-x New +[x New ])]

[0213] W C =[W(1+y New -[y New ]) W(y New -[y New ]) W(1-y New +[y New ]) W(2-y New +[y New ])]

[0214] S7.2.4, Obtain the pixel value at this coordinate by weighting. Among them W R W represents a row-weighted vector. C S represents a column-weighted vector. RC A pixel value matrix representing 16 pixels within a 4×4 neighborhood:

[0215]

[0216] In the above technical solution, in step eight, the different temperatures T and their corresponding color images I obtained in step one are utilized. Calib The specific methods for conducting colorimetric temperature calibration include:

[0217] Using the color images I corresponding to different temperatures T obtained in step one Calib I was obtained through the principle of colorimetric thermometry. Calib The colorimetric thermometry method corresponds to the red and green channels of the pixels at the measurement points of the pyrometer. Based on the functional relationship between the ratio of the spectral radiance of an object at two wavelengths and the measured temperature, the temperature value is calculated through formula transformation and derivation. The colorimetric thermometry formula is as follows:

[0218]

[0219] Where C2 is the second radiation constant, λ g λ represents the wavelength corresponding to green light. r Where λ represents the wavelength corresponding to red light, R is the output value of the R channel of the color camera, G is the output value of the G channel of the color camera, and K is the spectral response coefficient of the camera. K g K represents the response coefficient of the G primary color channel of a color camera. r The R primary color channel response coefficient of the camera;

[0220] By transforming the above formula, we obtain the derived formula for the spectral response coefficient K to be calibrated:

[0221]

[0222] Based on the above formula, we substitute different temperatures T and color images I. Calib The measurement position corresponds to the pixel point Numerical values ​​can be used to calculate the corresponding K values ​​for different temperatures T. To find a more realistic relationship between the data, a cubic B-spline algorithm is used to fit the obtained data, yielding the calibration parameter K and the corresponding values ​​at the measurement point temperature T. The curve showing the fitting relationship between the numerical values.

[0223] In the above technical solution, the full-field temperature image of the surface of a high-temperature object is obtained using the calibrated fitting relationship curve. In step nine, the specific methods include:

[0224] By using the fitted relationship curve, each pixel point is obtained. The calibration parameter K corresponding to the value can be used to calculate the mapped temperature of the pixel through the derivation formula of the spectral response coefficient K. By traversing the entire image, the temperature matrix corresponding to the entire color image can be obtained.

[0225] After calculating the temperature corresponding to each pixel, it is assigned to the color image I. Color Each pixel is analyzed to obtain a temperature image containing temperature information. Where the pixel value is represented as (u t ,v t ),u t =0,...,M t -1,v t =0,...,N t -1, (u t ,v t () represents a temperature image The uth t line, v t Column coordinates, M t and N t The image size represents the temperature image. The total number of pixels is M t ×N t .

[0226] In the above technical solution, in step ten, the three-dimensional point cloud data PointCloud of the high-temperature object surface is processed. w The surface temperature information of the point cloud in d is assigned to obtain the point cloud data of the high-temperature object. Temp Specific methods include:

[0227] Based on the imaging model of a stereo camera, the 3D point cloud data of the high-temperature object surface is generated. w The point coordinates in the high-resolution grayscale left image The pixel coordinates are mapped one-to-one, by using high-resolution temperature images The temperature information corresponding to each pixel is assigned to the 3D point cloud data of the high-temperature object surface (PointCloud). w PointCloud data of high-temperature objects was obtained. Temp In a 3D point cloud, each point is represented as a PointCloud. Temp (x T ,y T ,z T ,r T ,g T ,bT ),T=0,1,...,PN-1,(x T ,y T ,z T (r) represents the three-dimensional spatial coordinates of each point. T ,g T ,b T ) represents the temperature information carried by each point, T represents the index value of that point, and PN represents the number of points in the 3D point cloud.

[0228] In this example, the target of the 3D reconstruction is an engine blade. The target is under high temperature, appearing reddish-brown and emitting a bright white light. Using... Figure 1 The invention described provides a comprehensive measurement method for the three-dimensional morphology and temperature field of high-temperature targets, enabling high-precision reconstruction of the three-dimensional point cloud and fusion of temperature information on engine blades. It utilizes a binocular grayscale camera and a color camera to acquire grayscale structured light images under illumination, grayscale images under non-illumination conditions, and grayscale images under non-illumination conditions, as described in the acquisition method below. Figure 2 As shown. The three-dimensional point cloud data of the surface of a high-temperature object acquired using a high-precision binocular structured light system is as follows. Figure 3 As shown. Image acquired by the high-resolution left camera in a binocular structured light system in a non-illuminated state. like Figure 4 As shown, its resolution is 3000×4096, with a total of 12,288,000 pixels; the low-resolution color image I acquired by the low-resolution temperature camera. Color like Figure 5 As shown, its resolution is 2048×2448, with a total of 5,013,504 pixels. For the high-resolution grayscale left image... Image obtained after contrast enhancement like Figure 6 As shown, contrast enhancement amplifies the grayscale difference between the specimen and the background. This is applied to the enhanced image. Image obtained after thresholding like Figure 7 As shown, thresholding was used to separate the specimen from the background. The Canny edge extraction algorithm was employed to analyze the image. and Image I Color Image obtained after edge extraction and images Each as Figure 8 Figure 9 As shown. For the image and images The SURF algorithm is used to extract and describe feature points. Based on the obtained feature descriptor subsets, the Euclidean distance between feature vectors is used to determine whether feature points match, resulting in a feature point pair matching set Mate.rough The RANSAC algorithm is used to match the obtained feature point pairs to the set Mate. rough Error correction and manual filtering of matching points are performed to obtain the final accurate feature point pair matching set Mate. accurate Based on this, the transformation matrix H is calculated. g Create a high-resolution grayscale left image. Blank images of uniform size Through images With Image I Color The mapping relationship between them, combined with the bicubic interpolation algorithm, gives high-resolution blank images By assigning values ​​to the pixels, a new high-resolution temperature image is obtained. like Figure 10 As shown. Finally, the high-resolution temperature image... The temperature information contained within is mapped onto the 3D point cloud data of the high-temperature object surface using rules from a binocular vision system. w The above two methods are combined to obtain PointCloud data of high-temperature objects with temperature information. Temp like Figure 11 As shown.

[0229] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.

[0230] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target, characterized in that, Includes the following steps: Step 1: Acquire color images during the heating process I Calib and its corresponding temperature data T ; When heated to the specified test temperature, acquire color images under non-illuminated conditions. I Color High-resolution grayscale image under non-illuminated conditions I L Gray and I R Gray Binocular structured light grayscale image under active illumination. I LD Gray and I RD Gray ; Step 2: Obtain the 3D point cloud of the high-temperature object surface using binocular structured light grayscale images through calculation. PointCloud w ; Step 3: Select a high-resolution grayscale image captured by the grayscale camera on the same side as the color camera under non-light conditions. I L Gray A contrast-limited adaptive histogram equalization algorithm is used to process the high-resolution grayscale left image. I L Gray Contrast enhancement is performed to obtain an enhanced image. I New Gray ; Step 4: After enhancing the contrast of the grayscale image in the non-illuminated state, the grayscale difference between the background and the high-temperature object is amplified. Next, a filtering threshold is set. θ thr To separate the background and the high-temperature object, a thresholding process is applied to the enhanced image to obtain the thresholded image. I New Gray The following formula is used to determine whether the grayscale value meets the requirements: This preserves the specimen while removing the background, resulting in a thresholded image. I Thre Gray The pixel values ​​of the image are represented as I Thre Gray ( x L , y L ), x L =0,…, M L-1 , y L =0,…, N L-1 The image resolution remains unchanged from before processing; Step 5: Process the high-resolution grayscale left image. I Thre Gray and color images I Color Perform edge detection to obtain a contour map. I Edge Gray and color images I Edge Color ; Step 6: Calculate the contour image I Edge Color To contour image I Edge Gray Transformation matrix H g ; Step 7: Transform the matrix H g For color images I Color The transformation is performed, and a new high-resolution color image is generated using an image interpolation algorithm. I Reg Color ; Step 8: Utilize the color images captured at different times during the heating process in Step 1. I Calib and its temperature data T Complete the colorimetric temperature measurement calibration; Step 9: Using the calibrated colorimetric temperature correspondence, convert the high-resolution color image... I Reg Color Convert to temperature image I Reg Temp ; Step 10: Analyze the 3D point cloud data of the high-temperature object surface. PointCloud w The surface temperature information of the point cloud is assigned to obtain the point cloud data of the high-temperature object. PointCloud temp .

2. The method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target as described in claim 1, characterized in that, In step one, the specific method for obtaining image data of the surface of a high-temperature object includes: A data acquisition system is constructed, which includes two high-resolution grayscale industrial cameras, one color industrial camera, and one non-contact pyrometer. The object to be tested is heated, and as the temperature rises to the specified test temperature, the temperature change at a fixed location on the object's surface is monitored using a non-contact pyrometer. T Simultaneously, multiple color images at different temperature intervals were captured using a color industrial camera. I Calib Record and store different temperatures T and its corresponding color image I Calib Used for subsequent colorimetric temperature measurement calibration; Once the object under test is heated to the specified test temperature, structured light is actively projected onto the surface of the high-temperature object using the optomechanical system of the binocular structured light system. Simultaneously, two high-resolution grayscale cameras acquire binocular structured light grayscale images under active illumination for subsequent 3D point cloud calculations. After the optomechanical system stops projecting light, a grayscale camera positioned on the same side as the color camera acquires a high-resolution grayscale image of the high-temperature object. I Gray Color images of high-temperature objects under non-light-shielded conditions were captured using a color camera. I Color .

3. The method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target as described in claim 1, characterized in that, In step two, a three-dimensional point cloud of the surface of the high-temperature object is obtained. PointCloud w and images of the surface temperature of high-temperature objects I C Temp Specific methods include: A high-resolution grayscale left camera in a binocular structured light system is used to acquire a grayscale image of the target under illumination. I LD Gray The pixel value of a pixel is represented as I LD Gray ( x LD , y LD ), x LD =0,..., M LD -1, y LD =0,..., N LD -1, (x LD , y LD ) represents the first digit in the high-resolution grayscale left image. x LD line, number y LD Column coordinates M LD and N LD This refers to the image size of the high-resolution grayscale left image, i.e., the high-resolution grayscale left image. I LD Gray The total number of pixels is M LD × N LD The system uses a high-resolution grayscale right camera to acquire a grayscale image of the target under illumination. I RD Gray The pixel value of a pixel is represented as I RD Gray ( x RD , y RD ), x RD =0,..., M RD -1, y RD =0,..., N RD -1, ( x RD , y RD () indicates a high-resolution grayscale image. I RD Gray The first in x RD line, number y RD Column coordinates M RD and N RD This refers to the image size of the high-resolution grayscale right image, i.e., the high-resolution grayscale right image. I RD Gray The total number of pixels is M RD × N RD Finally, based on the principle of active structured light binocular vision, three-dimensional point clouds of the surface of high-temperature objects were obtained. PointCloud w In a 3D point cloud, each point is represented as... PointCloud w ( x S ,y S ,z S ), S =0,1,..., PN -1, ( x S ,y S ,z S () represents the three-dimensional coordinates of each point. S This represents the index value of each point. PN This indicates the number of points in a 3D point cloud.

4. The method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target as described in claim 1, characterized in that, In step three, a high-resolution grayscale image under non-illuminated conditions is selected, and a contrast-limited adaptive histogram equalization algorithm is used to enhance the contrast of the high-resolution grayscale left image to obtain the enhanced image. I New Gray Specific methods include: High-resolution grayscale images under non-illuminated conditions I L Gray and I R Gray Select the grayscale image from the same side as the color camera. If the color camera is located on the left, then select the grayscale image captured by the left-side grayscale camera. I L Gray left image with high resolution grayscale I L Gray For example, the pixel value is represented as I L Gray ( x L , y L ), x L =0,..., M L -1, y L =0,..., N L -1, ( x L , y L () indicates a high-resolution grayscale left image I L Gray The first in x L , line, number y L Column coordinates M L and N L This refers to the image size of the high-resolution grayscale left image, i.e., the high-resolution grayscale left image. I L Gray The total number of pixels is M L × N L ; S3.1, Enhance the high-resolution grayscale image to be enhanced. I L Gray The image is divided into blocks by manually setting the number of blocks. M p × N p Divide the row into small blocks, that is, divide the row into equal parts. M p Divide the column into equal parts. N p If a row or column is not divisible by another row or column. M p and N p In this case, the rows and columns are padded below and to the right of the original image to ensure they are divisible by the original image. M p and N p The pixel values ​​of the supplemented pixels are obtained by mirroring the boundaries of the original image; the padded image is denoted as... I Com Gray The pixel value of its pixel is represented as I Com Gray ( x Com , y Com ), x Com =0,..., M Com -1, y Com =0,..., N Com -1, M Com This represents the number of rows in the image after pixel padding. N Com This represents the number of columns in the image after pixel padding; each pixel block is denoted as... I RC Gray , R =0,..., M p -1, C =0,..., N p -1, R Represent in image I Com Gray In the text, the row index of the pixel block. C Represent in image I Com Gray In the diagram, the column index of the pixel block is used, and the pixel value within the pixel block is denoted as... I RC Gray ( x RC , y RC ), x RC =0,..., M Com / M p -1, y RC =0,..., N Com / N p -1, each pixel block has M Com N Com / M p N p Each pixel is a separate pixel block that does not overlap with each other. S3.2 Calculate each pixel block I RC Gray grayscale histogram I RC HO grayscale histogram I RC HO The horizontal axis represents the grayscale value, ranging from [0, 255], and the vertical axis represents the number of pixels at the current grayscale value. The combination of the horizontal and vertical axes is called a bin, or grayscale histogram. I RC HO There are 256 bins, and the size of the corresponding y-axis represents the bin size. The size of each bin in the grayscale histogram of each pixel block is denoted as... Bin num RC num=0,1,…,255 represents the index of bin; S3.3, Manually setting grayscale limit thresholds δ Using each pixel block I RC Gray The average number of pixels per bin is obtained by dividing the number of pixels in the grayscale histogram by the number of bins (256). Bin aver RC , If the grayscale is relatively uniform, the number of pixels corresponding to the grayscale level in each bin is equal. Therefore, the part that needs to be cropped is the portion exceeding the grayscale level in each bin. δ Bin aver RC The size of the portion exceeding the limit is denoted as . OB num RC , OB num RC = Bin num RC - δBin aver RC ;Will I RC H All excess portions are summed, and the sum is denoted as . Sum RC ,Will Sum RC Distribute evenly among each bin, with the assigned value being... Sum RC / 256, generating a new grayscale histogram. I RC HN The size of each bin in the histogram is denoted as . NBin num RC , num =0,1,…,255, representing the index of bin; S3.4, For each pixel block I RC Gray The generated new grayscale histogram I RC HN Histogram equalization is performed using the following mapping function to obtain the mapped value corresponding to each grayscale value, which is then used to replace the original grayscale value; where P RC m This represents the new value obtained after mapping the current grayscale value using a function. N Represents pixel blocks I RC Gray The number of pixels, N= M Com N Com / M p N p ; S3.5, Process each pixel block through steps S3.1-S3.

4. I RC Gray After processing, if each pixel block I RC Gray If pixels are transformed only through a mapping function within a pixel block, the final image will exhibit a blocky effect, with noticeable separation between pixel blocks. Therefore, to address this issue, the pixel values ​​of each pixel in each block are calculated using a bilinear interpolation algorithm, ultimately resulting in an enhanced image. I New Gray ; For pixels O Record its pixel value O ( m , n The mapped pixel is obtained by subtracting half the height and half the width of the pixel block from the x and y coordinates of the pixel. OC Its coordinates are pixel OC It may not be within the image range; assuming a pixel. OC The pixel block is I RC Gray ,at this time R and C The ranges are respectively R ∈[-1, M p -1], C ∈[-1, N p -1], find its three adjacent pixel blocks, and denote them as follows: I (R+1)C Gray , I R(C+1) Gray , I (R+1)(C+1) Gray ;if R =-1 or C =-1 indicates that the current pixel block does not exist; it is a virtual pixel block and cannot be used as an associated pixel block. Therefore, values ​​less than 0 are set to 0. R =-1 and C =-1, the four associated pixel blocks are: I (-1)(-1) Gray , I (0)(-1) Gray , I (-1)(0) Gray and I (0)(0) Gray ,only I (0)(0) Gray If the first pixel is an actual existing pixel block, and the portions of the other three pixel blocks that are less than 0 are all set to 0, then we have: I (-1)(-1) Gray = I (0)(0) Gray , I (0)(-1) Gray = I (0)(0) Gray , I (-1)(0) Gray = I (0)(0) Gray Therefore, only one pixel block actually affects the pixel value of the current pixel; then, the mapping function of the associated pixel block is used to adjust the pixel value. O ( m , n The mapping is performed to obtain four mapped pixel values. P RC O(m,n) , P (R+1)C O(m,n) , P R(C+1) O(m,n) and P (R+1)(C+1) O(m,n) ; Obtain pixel values ​​in associated pixel blocks O ( m,n After obtaining the four mapped pixel values, the mapped pixel values ​​need to be weighted according to a certain ratio to obtain the final pixel value. OF ( m,n The weighting coefficients are determined by bilinear interpolation. First, the mapped pixels are calculated. OC The distance to the top, bottom, left, and right boundaries of the corresponding pixel block is denoted as . y 1. y 2. x 1 and x 2. Then, divide the above four values ​​by the height and width of the current pixel block respectively to obtain the weighting coefficients: , , , ,in , The final pixel value is obtained by weighting the pixel values ​​using the following formula. OF ( m, n ); 。 5. The method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target as described in claim 1, characterized in that, In step six, the contour image is calculated. I Edge Color To contour image I Edge Gray Transformation matrix H g Specific methods include: S6.

1. Use the SURF algorithm to process the contour map. I Edge Gray and I Edge Color Feature points are extracted and described to obtain feature point sets. Feature Edge Gray , Feature Edge Color and the corresponding feature descriptor subset FD Edge Gray , FD Edge Color The number of elements in different feature point sets may not be the same, but the number of elements in the feature descriptor subset corresponding to the feature point set remains the same. in, feature i Gray Representative outline diagram I Edge Gray The feature points extracted from them i =0,…, M fg -1, M fg Representative outline diagram I Edge Gray The number of feature points extracted; feature j Color Representative outline diagram I Edge Color The feature points extracted from them j =0,…, M ft -1, M ft Representative outline diagram I Edge Color The number of feature points extracted; FD i Gray Representative outline diagram I Edge Gray The feature descriptors corresponding to the feature points extracted; FD j Color Representative outline diagram I Edge Color The feature descriptor corresponding to the feature point extracted is a 128-dimensional feature vector. S6.2, By analyzing the obtained feature descriptor subset FD Edge Gray and FD Edge Color Perform matching to obtain a set of matched feature point pairs. Mate rough ; For contour images I Edge Gray and contour images I Edge Color Feature descriptors in FD i Gray and FD j Color The similarity between feature vectors is measured using the Euclidean distance; the smaller the Euclidean distance, the more similar the two vectors. The feature descriptor is calculated using the following formula. FD i Gray and feature descriptors FD j Color The Euclidean distance, of which Dist () indicates the Euclidean distance between two vectors: dist ij =Dist ( FD i Gray , FD j Color ) Find the contour image I Edge Gray Each feature descriptor FD i Gray With contour image I Edge Color Feature descriptor subset FD Edge Color All feature descriptors FD j Color Euclidean distance, taken from FD i Gray With minimum Euclidean distance FD j Color As its matching point; however, a one-to-many problem arises when calculating the matching point, for feature descriptors FD i Gray In the feature description subset FD Edge Color We can find multiple descriptors with the smallest Euclidean distance that are equal to each other. Therefore, we sort the calculated Euclidean distances from smallest to largest, and select the feature points corresponding to the two feature descriptors with the first and second highest Euclidean distances as the feature points to be matched. The feature point with the highest Euclidean distance is called the nearest neighbor feature point. feature Near Color , and feature descriptors FD i Gray The distance is expressed as dist Near The feature point ranked second is called the second nearest neighbor feature point. feature Color NNear , and feature descriptors FD i Gray The distance is expressed as dist NNear ;in Near Represents the nearest neighbor feature points feature Color Near In the set of feature points Feature Edge Color Index in NNear Representing the second nearest neighbor feature points Feature Color NNear In the set of feature points Feature Edge Color The index in the middle; the nearest neighbor feature point is determined by the ratio of the nearest neighbor Euclidean distance to the second nearest neighbor Euclidean distance. feature Color Near Whether a feature point is a match or not, if the ratio equals 1, it indicates a false match and is directly excluded because matching points for feature points are not unique; this is manually set. ratio , ratio ∈(0,1), if dist Near / dist NNear < ratio Then we consider the feature points feature Color Near These are the matching feature points that meet the conditions. This matching point pair ( feature i Gray , feature Color Near Add to the set of matching points Mate rough ; for feature descriptor subsets FD Edge Gray Perform the above operation on all descriptors to obtain the final set of matching point pairs: Mate rough =｛( feature Gray a , feature Color b )｝ in a Representative contour image I Edge Gray The index of a feature point in a feature point set. b Representative contour image I Edge Color Indices of feature point pairs in the feature point set; S6.

3. Manually assist in filtering the set of matching point pairs, and use the RANSAC algorithm to eliminate mismatches to obtain the final transformation matrix. H g ; For the obtained set of matching point pairs Mate rough If the transformation matrix is ​​calculated directly, the result will have a large error due to the presence of many mismatched feature point pairs. Therefore, manual assistance is needed in the experiment to filter feature point pairs and obtain a more accurate set of matching pairs. Mate arti : Mate arti =｛( feature Gray k , feature Color v )｝ in k Representative contour image I Edge Gray feature point set Feature Edge Gray Index of feature points in the middle, v Representative contour image I Edge Color feature point set Feature Edge Color The index of feature points; after manual selection, since manual selection also has errors, the Random Consensus Algorithm (RANSAC) is needed to accurately estimate the final transformation matrix. H g ; Transformation matrix H g It is a 3×3 matrix, represented as: Each matched feature point pair should, in principle, satisfy the following equation: in feature Color v ( x ) represents the set of matching point pairs Mate arti Mid-contour image I Edge Color Corresponding feature points feature Color v x-coordinate feature Color v ( y ) represents the set of matching point pairs Mate arti Mid-contour image I Edge Color Corresponding feature points feature Color v The ordinate; feature Gray k ( x ) represents the set of matching point pairs Mate arti Mid-contour image I Edge Gray Corresponding feature points feature Gray k x-coordinate feature Gray k ( y ) represents the set of matching point pairs Mate arti Mid-contour image I Edge Gray Corresponding feature points feature Gray k The ordinate, ω Represents the scale parameter; For matrix H g In this context, it has 8 unknown parameters and requires at least 8 linear equations to solve. Therefore, at least 4 matching point pairs are needed for calculation at a time. The RANSAC algorithm is used here to calculate the optimal transformation matrix model using multiple matching point pairs. H g Specifically, it includes: S6.3.1, First, start with the set of matching points. Mate arti Four feature point pairs are randomly selected from the data to calculate the matrix model, denoted as... H g i ,in i =0,1,…, iter -1, i Representing the i +1 iterations, iter This represents the total number of iterations. iter The initial value is set manually; S6.3.2 Then, calculate the cost value of all matching point pairs in the current iteration number according to the cost function of the following formula: Compared to manually set cost thresholds Cost thre In comparison, if Cost i kv < Cost thre Then add the matching point pair to the current inner point set. Inner i Otherwise, add the matching point pair to the current outer point set. Outer i ; S6.3.3, Statistics Inner i The number of interior points is denoted as . Number i With the optimal interior point set Inner best Number of interior points Number best In comparison, if Number i > Number best Then update the optimal set of interior points to the current set of interior points. Inner best = Inner i , Number i >= Number best Update the optimal model H g For the currently calculated model, H g = H i g Simultaneously updated i and number of iterations iter , i = i +1, ,in w i Representing the i The proportion of interior points to the total number of points in +1 iterations m To calculate the minimum number of samples required for the model, p For confidence level, here m =4, p =0.995; S6.3.4 Finally, iterate through steps S631-S634 until... i > iter The iteration ends; at this point, the optimal set of interior points is... Inner best The matching point pairs in the set are the final set of precise feature point pairs, denoted as . Mate accurate Calculated in this way H g The optimal transformation matrix is: Mate accurate =｛( feature c Gray , feature d Color )｝ in c Represents the contour image after error matching removal I Edge Gray feature point set Feature Edge Gray Index of feature points in the middle, b Represents the contour image after error matching removal I Reg Color feature point set Feature Edge Color Index of feature points.

6. The method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target as described in claim 1, characterized in that, In step seven, the transformation matrix is ​​used. H g For color images I Color The transformation is performed, and a new high-resolution color image is generated using an image interpolation algorithm. I Reg Color Specific methods include: S7.1 If color image I Color The resolution is lower than that of the grayscale left image I L Gray The resolution, for color images I Color and high-resolution grayscale left image I L Gray In other words, by transforming the matrix H g Color image I Color Mapped to high-resolution grayscale left image I L Gray Therefore, the color image is in the coordinate system in which it is located. I Color The coordinates before and after the transformation are represented as follows: in, u new and v new Indicates the left image in high-resolution grayscale. I L Gray Color image in the coordinate system I Color Corresponding coordinate values ，ω= h 31 u t + h 32 v t + 1; As can be seen from the above formula, there is a one-to-one correspondence between the two coordinates, for low-resolution color images. I Color of M t × N t Each pixel corresponds to a high-resolution grayscale left image. I L Gray In the coordinate system where M t × N t Each pixel represents a single point cloud on the surface of a high-temperature object. However, when assigning temperature information later, the reconstructed point cloud data is obtained from a high-resolution grayscale image. I L Gray What we get, through forward mapping, is only... M t × N t To ensure a one-to-one correspondence between the reconstructed point cloud data and the temperature information, the temperature information of each pixel needs to be obtained from the low-resolution color image. I Color and high-resolution grayscale left image I L Gray The resolution is unified as M L × N L This yields a high-resolution color image, so the low-resolution color image is then processed through inverse mapping and interpolation. I Color Process it; Through formula get: Simultaneously, left multiplication H -1 g get: make: Expanding, we get the inverse mapping formula: The above formula yields the inverse mapping from high-resolution image to low-resolution image, where ( u h , v h () represents the pixel coordinates of a high-resolution image. The pixel coordinate values ​​here must be integers. u l , v l The coordinates represent the corresponding values ​​of the high-resolution image coordinates in the low-resolution image coordinate system. The coordinate values ​​of the pixels may be floating-point numbers. Although the reverse mapping formula is also one-to-one, when each pixel of the high-resolution image is mapped to the low-resolution image, an out-of-bounds check is performed. If it is out of bounds, the pixel value of the pixel is padded to (0,0,0), and finally a high-resolution image can be obtained. For the case where the coordinate values ​​of the mapped points are floating-point numbers, the pixel values ​​are obtained through interpolation algorithms. Therefore, first create a blank image with the same size as the high-resolution grayscale left image. I Reg Color The pixel value of a pixel is represented as I Reg Color ( x Reg , y Reg ), x Reg =0,..., M L -1, y Reg =0,..., N L -1, ( x Reg , y Reg ) represents a blank image I Reg Color The first in x Reg line, number y Reg The column coordinates are obtained using the following formula: in,( x New , y New () represents a blank image I Reg Color The coordinates are transformed in the low-resolution color image I C Color Coordinates in a coordinate system, ( x New , y New (This may result in out-of-range situations, making it impossible to obtain low-resolution temperature images.) I C Color Find the corresponding value online; you need to manually set the pixel value. (At the same time...) x New , y New The coordinates of the image may be floating-point values, so further interpolation is needed to obtain the corresponding pixel values. S7.2 Next, determine whether the coordinates of the transformed image are out of bounds. If they are, fill the pixel values ​​to (0,0,0); otherwise, use bicubic interpolation to fill the blank image. I Reg Color Assign values ​​to RGB color information to generate a high-resolution color image. I Reg Color Specifically, it includes: S7.2.1, Regarding coordinates ( x New , y New Perform an out-of-bounds check; if an out-of-bounds error occurs, set the pixel value to (0,0,0); otherwise, use bicubic interpolation to obtain the result. x New , y New The pixel value of ); S7.2.2, will ( x New , y New The 16 pixels within a 4×4 neighborhood are numbered according to their row and column, and each coordinate is related to ( x New , y New Calculate the distances in the row and column respectively, and construct the row and column distance vectors respectively: R =[1+ uu 1- u 2- u ] C =[1+ vv 1- v 2- v ] S7.2.3, In order to obtain the pixel value pair of each pixel in the 4×4 neighborhood. I Color ( x New , y New The influence weights are calculated using a cubic polynomial. W ( x The row and column distances of adjacent pixels are weighted separately to obtain the distances between adjacent pixels. I Color ( x New , y New Row and column distance weight vectors W R and W C ,in a =-0.5, [] indicates rounding down to the nearest integer; W R =[ W (1+ x New -[ x New ]) W ( x New -[ x New ]) W (1- x New +[ x New ]) W (2- x New +[ x New ])] W C =[ W (1+ y New -[ y New ]) W ( y New -[ y New ]) W (1- y New +[ y New ]) W (2- y New +[ y New ])] S7.2.4, weighted average yields ( x New , y New ) pixel values I Reg Color ( x Reg , y Reg )= I Color ( x New , y New )= W R S RC W T C ,in W R Represents a row-weighted vector. W C Represents a column-weighted vector. S RC A pixel value matrix representing 16 pixels within a 4×4 neighborhood: 。 7. The method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target as described in claim 1, characterized in that, In step eight, the different temperatures obtained in step one are utilized. T Colorimetric temperature calibration is performed using the corresponding color images, and the specific methods include: Using the different temperatures obtained in step one T Corresponding color image I Calib Based on the principle of colorimetric thermometry, the following was obtained: I Calib The colorimetric thermometry method corresponds to the red and green channels of the pixels at the measurement points of the pyrometer. Based on the functional relationship between the ratio of the spectral radiance of an object at two wavelengths and the measured temperature, the temperature value is calculated through formula transformation and derivation. The colorimetric thermometry formula is as follows: in, C 2 is the second radiation constant. λ g The wavelength corresponding to green light. λ r The wavelength corresponding to red light. R For color cameras R The output value of the channel, G For color cameras G The output value of the channel, K For the camera's spectral response coefficient, K = ln( K g / K r ), K g For color cameras G Primary color channel response coefficients K r For camera R Primary color channel response coefficients; Transforming the above formula yields the spectral response coefficients to be calibrated. K The derivation formula: Based on the above formula, substituting different temperatures... T and color images I Calib ln(m) of the pixel corresponding to the measurement position R / G Numerical values ​​were calculated to obtain different temperatures. T Correspondingly K To find a more realistic relationship between the data, a cubic B-spline algorithm was used to fit the obtained data to obtain calibration parameters. K Temperature at the measurement point T and corresponding ln( R / G The fitting curve of the numerical values.

8. The method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target as described in claim 1, characterized in that, The full-field temperature image of the surface of a high-temperature object is obtained using the calibrated fitting curve. I C Temp In step nine, the specific method includes: Using the fitted relationship curve, the ln( ) of each pixel is obtained. R / G The calibration parameters corresponding to the value) K Through spectral response coefficient K The derived formula can be used to calculate the mapped temperature of a pixel. By traversing the entire image, the temperature matrix corresponding to the entire color image can be obtained. I Reg Temp ; After calculating the temperature corresponding to each pixel, it is assigned to the color image. I Color Each pixel is used to obtain a temperature image containing temperature information. I C Temp , where the pixel value is represented as I C Temp ( u t , v t ), u t =0,..., M t - 1, v t =0,..., N t - 1, ( u t , v t () represents a temperature image I C Temp The first in u t line, number v t Column coordinates M t and N t The image size represents the temperature image. I C Temp The total number of pixels is M t × N t .

9. The method for comprehensively measuring the three-dimensional morphology and temperature field of a high-temperature target as described in claim 1, characterized in that, In step ten, the three-dimensional point cloud data of the high-temperature object surface is processed. PointCloud w The surface temperature information of the point cloud is assigned to obtain the point cloud data of the high-temperature object. PointCloud Temp Specific methods include: Based on the imaging model of a stereo camera, the three-dimensional point cloud data of the surface of a high-temperature object is... PointCloud w The point coordinates in the high-resolution grayscale left image I L Gray The pixel coordinates are mapped one-to-one, by using high-resolution temperature images I C Temp The temperature information corresponding to each pixel is assigned to the 3D point cloud data of the high-temperature object surface. PointCloud w The point cloud data of the high-temperature object was obtained. PointCloud Temp In a 3D point cloud, each point is represented as... PointCloud Temp ( x T , y T , z T , r T , g T , b T ), T =0,1,…, PN -1, ( x T , y T , z T ) represents the three-dimensional spatial coordinates of each point, ( r T , g T , b T This represents the temperature information carried by each point. T This represents the index value of each point. PN This indicates the number of points in a 3D point cloud.

Images