A global three-dimensional morphology and temperature field reconstruction method for high-temperature objects

Abstract

The present invention discloses a method for reconstructing the global three-dimensional morphology and temperature field of a high-temperature object. The three-dimensional point cloud reconstruction system uses a projection surface structured light to project onto the surface of the high-temperature object, calculates the surface phase of the object by a phase shift method, and calculates the three-dimensional point cloud of the measured high-temperature object with the phase as the medium. The point cloud is colored to obtain a colored point cloud containing temperature information. A fixed marker sphere is arranged around the three-dimensional point cloud reconstruction system, and the three-dimensional coordinates of the center of the marker sphere are fitted according to the three-dimensional coordinates of the marker point. Through the three-dimensional point cloud matching of the marker sphere, the rigid body transformation relationship R and T of each point cloud obtained by the front binocular system is calculated, and the point cloud is spliced ​​to obtain an overall colored point cloud. The present invention uses a binocular vision positioning and tracking system to calculate the positional relationship between the point clouds obtained by the three-dimensional point cloud reconstruction system. The number of matching points is small, the amount of calculation is small, and the accuracy is high. It overcomes the problem that the point cloud features of the measured object are not obvious and the matching is difficult due to the excessive number of point clouds, and greatly improves the measurement speed and accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of high-temperature three-dimensional morphology measurement, and more specifically, relates to a global three-dimensional morphology and temperature field reconstruction method for a high-temperature object. Background Art

[0002] In modern industrial production, high-temperature 3D shape measurement technology is increasingly showing its irreplaceable importance. With the continuous advancement of science and technology and the continuous improvement of production needs, the requirements for material performance, product form and quality control are also getting higher and higher. For example, aerospace engines need to withstand a high temperature of 2000℃ during operation, and metallurgical forging can reach 1000℃. We all need to accurately measure the shape of objects under high temperature conditions.

[0003] In the measurement of high-temperature objects, accurate measurement of their morphology and temperature is crucial. Morphology reflects the geometric structure and shape characteristics of the surface of an object. By monitoring the three-dimensional morphology of an object at high temperature, its performance and function can be evaluated. Temperature information, as an important physical property of an object in a high-temperature environment, can reveal the thermodynamic properties of the object and provide an important basis for theoretical analysis and engineering design. For example, in steel manufacturing, for an object with a uniform texture, its surface temperature should also be evenly distributed. Then a sudden change in temperature indicates that the object is likely to be deformed or mixed with slag. Therefore, in the measurement of high-temperature objects, we must consider both morphology and temperature.

[0004] At present, binocular structured light system technology is relatively mature. By simulating the human eye and projecting structured light, it can achieve high-precision 3D point cloud reconstruction of the object to be measured, and is widely used in industrial measurement and manufacturing. In addition to this system, in order to measure the temperature of the object, it is necessary to add an additional color camera to capture temperature information, and use it to render the point cloud data to obtain a 3D color point cloud model that integrates temperature information.

[0005] In addition, for surface structured light technology, due to the limitations of light intensity and accuracy, only a certain range of areas can be projected, scanned and reconstructed at one time. For large-sized high-temperature objects such as large forgings, a single scan cannot cover the entire area of ​​the object. At this time, it is necessary to reconstruct the three-dimensional point cloud of different areas of the high-temperature object at different positions, and then combine them through point cloud stitching technology to obtain a complete three-dimensional point cloud model. At this stage, point cloud stitching mostly uses the method of feature matching of the object itself or projecting feature points. For high-temperature objects, the high-temperature environment eliminates the method of setting marker points on the surface, and the feature information itself is less, and the curvature, normal vector, texture and other feature information available are less, so the matching relationship error calculated in this way is large. Moreover, for point clouds with a large amount of data, global search and calculation of feature information for matching requires a lot of calculations, which slows down the operation speed of the system. For the method of projecting feature points, because the high-temperature object is always in the process of ablation, there will be deformation, and the feature points projected and photographed at different positions will produce position errors due to deformation, so the matching relationship error calculated in this way is large. Summary of the invention

[0006] The purpose of the present invention is to overcome the problems that the point cloud features of the high-temperature object being measured are not obvious, the matching is difficult due to the excessive number of point clouds, and the matching error is large due to deformation, and to provide a global three-dimensional morphology and temperature field reconstruction method for high-temperature objects to improve the speed and accuracy of feature point recognition and matching, thereby improving the speed and accuracy of measurement.

[0007] To achieve the above-mentioned purpose, the present invention provides a method for reconstructing the global three-dimensional morphology and temperature field of a high-temperature object, which is characterized by comprising:

[0008] (1) Image acquisition

[0009] 1.1) Construct a 3D point cloud reconstruction system: install two high-resolution grayscale cameras, i.e., a high-resolution grayscale left camera and a high-resolution grayscale right camera, at the left and right ends of the horizontal device and keep a certain distance between them, install the optical machine in the middle of the two high-resolution grayscale cameras, and install the low-resolution color camera above the optical machine. At the same time, design a marker point frame, wherein the two high-resolution grayscale cameras, the optical machine and the low-resolution color camera are surrounded by the marker point frame, and the marker point frame includes a plurality of marker balls, and the triangles formed by the three different marker balls on the marker point frame are different and unique, and the marker ball includes a plurality of rhombus surfaces, and each rhombus surface has a circular marker point;

[0010] Construct a binocular visual positioning and tracking system: install two other high-resolution grayscale cameras at both ends of the horizontal fixed device and keep a certain distance between them as a binocular visual positioning and tracking system;

[0011] 1.2) Ensure that the optical function in the 3D point cloud reconstruction system can fully and clearly project structured light onto the surface of the high-temperature object being measured, and that the two high-resolution grayscale cameras and low-resolution color cameras can normally, completely and clearly capture the high-temperature object being measured, and ensure that the two high-resolution grayscale cameras in the binocular vision positioning and tracking system can normally, completely and clearly capture the marker point framework;

[0012] 1.3) First, adjust the position and posture of the calibration plate and use Zhang's calibration method to complete the calibration of the camera in the 3D point cloud reconstruction system and the binocular vision positioning and tracking system;

[0013] 1.4) For the 3D point cloud reconstruction system, take photos of the high-temperature object at J positions:

[0014] For the jth position, j = 1, 2, ..., J, first, a low-resolution color camera is used to shoot the high-temperature object to be measured, and the obtained image is recorded as image The high-temperature object to be measured is photographed with a high-resolution grayscale left camera, and the obtained image is recorded as image Then, an optical machine is used to project N phase-shift code patterns and K complementary Gray code patterns onto the surface of the high-temperature object to be measured. For each projected structured light pattern, the high-temperature object to be measured is photographed by left and right high-resolution grayscale cameras, and the obtained images are recorded as images and images The image and images The first N images of each are phase-shift code patterns, and the last K images are Gray code patterns;

[0015] For the binocular vision positioning and tracking system, the 3D point cloud reconstruction system is tracked and photographed at a fixed position. Every time the 3D point cloud reconstruction system adjusts a position, the two high-resolution grayscale cameras in the binocular vision positioning and tracking system, i.e., the high-resolution grayscale left and right cameras, take a frame image of the marked point, recorded as image and

[0016] (2) At each location, the 3D point cloud reconstruction system uses images and Calculate the 3D blank point cloud PC of the high temperature object being measured j

[0017] 2.1) Images taken at various locations and Computational phase image and

[0018]

[0019]

[0020] in, Respectively, images The pixel value at position (u,v), Phase image and The pixel value at position (u,v);

[0021] The phase image and Expand into absolute phase images

[0022]

[0023]

[0024] in, is the stripe level:

[0025]

[0026]

[0027] in, and The first K-1 frames and the K frames are complementary Gray code images respectively The code value obtained by converting Gray code to decimal is: and The first K-1 frames and the K frames are complementary Gray code images respectively The code value obtained by converting Gray code to decimal;

[0028] 2.2) Left and right absolute phase images and Perform epipolar correction to obtain the corrected image and The reprojection matrix Q is calculated and returned synchronously during the epipolar correction process j :

[0029]

[0030] Among them, the focal lengths of the calibrated high-resolution grayscale left and right cameras are the same, both f, c v is the coordinate of the principal point of the high-resolution grayscale left camera in the image, c v is the coordinate of the high-resolution grayscale right camera principal point in the image, T u is the baseline length;

[0031] 2.3) Phase matching generates disparity and calculates the coordinates of 3D points through disparity

[0032] For pixels By comparing the numerical value method in the image Find its corresponding position in , if it satisfies the judgment formula:

[0033]

[0034] Then it means (u l ,v l ) and (u r ,v r ) is the matching point;

[0035] The 3D point cloud reconstruction system is a binocular vision system, with a high-resolution grayscale left camera as the reference, and the disparity map size and image The disparity value is the number of columns v of matching points of the high-resolution grayscale left camera. l Subtract the number of columns v of high-resolution grayscale right camera matching points r :

[0036] disp(u l ,v l )=v l -v r

[0037] According to the disparity and reprojection matrix Q j The formula for calculating the coordinates of a 3D point is:

[0038]

[0039] The three-dimensional coordinates are (X / W, Y / W, Z / W), for the image and Calculate the three-dimensional coordinates of all points to get the three-dimensional blank point cloud PC j ;

[0040] (3) In the 3D point cloud reconstruction system, images are used and 3D blank point cloud PC j Temperature information fusion

[0041] 3.1) For images at each position and Calculate the optimal transformation matrix H jg Make the image Convert to Image In the coordinate system

[0042] 3.1.1) Use ORB algorithm to analyze the image and Extract and describe feature points to obtain feature point sets and the corresponding feature descriptor set

[0043] 3.1.2) The obtained feature descriptor set and Perform rough matching to obtain a set of matching feature point pairs And filter it to get a more accurate set of matching point pairs For feature points, the Hamming matrix between corresponding descriptors is used as a measure of similarity. The smaller the Hamming distance, the more similar the feature points are. The matching points with too large distance are filtered out to obtain a set of matching point pairs.

[0044] 3.1.3) Calculate the optimal transformation matrix H by using the RANSAC algorithm to remove mismatches jg : Randomly select 4 points to calculate the current transformation matrix model And calculate the cost function to update the good internal point data set, and iterate until the predetermined number of iterations is met, so as to calculate the optimal transformation matrix H jg ;

[0045] 3.1.4) Add correct matching points by manual selection;

[0046] 3.1.5) Use the optimal transformation matrix H jg Convert low-resolution color images The coordinate system is transformed to the high-resolution grayscale left image In the coordinate system of ;

[0047] 3.2) While performing coordinate system transformation, interpolation improves low-resolution color images Resolution, get the left image in high resolution grayscale High-resolution color images in coordinate systems

[0048] For each point of the high-resolution color image Perform reverse mapping and interpolation to obtain the corresponding points of the low-resolution color image Among them, M c ,N c Represent low-resolution color images The width and height of Represent the high-resolution grayscale left image Width and height of high-resolution color images and high resolution grayscale left image The same size:

[0049]

[0050]

[0051] in, Represents a high-resolution color image The horizontal and vertical coordinates of a point in the c ,v c ) represents a low-resolution color image The horizontal and vertical coordinates of the corresponding points in the

[0052] If the converted point exceeds the low-resolution color image The image boundary is The corresponding points on the low-resolution color image are discarded and assigned a value of 0. The image boundaries will correspond to the horizontal and vertical coordinates (u c ,v c ) is assigned to the corresponding point in the high-resolution color image. For the horizontal and vertical coordinates The calculated horizontal and vertical coordinates (u c ,v c ) is a floating point number, which is processed according to the bicubic interpolation algorithm to obtain the corresponding pixel value and assign it to the corresponding point of the high-resolution color image This results in a high-resolution color image

[0053] 3.3) For high-resolution color images Perform epipolar correction to obtain the corrected image

[0054] 3.4) Image The temperature information image is obtained using the colorimetric temperature measurement method, recorded as image

[0055] 3.5) For blank 3D point cloud PC j Each point (x, y, z) in the image is converted from a three-dimensional point to a pixel coordinate system to obtain the corresponding Coordinates on (u,v):

[0056]

[0057]

[0058] Among them, Z is the coefficient, int() means rounding;

[0059] In high-resolution color images Get the value of [RGB] at the coordinate (u,v) for the blank 3D point cloud PC j The three channels of the point (x, y, z) in are assigned values ​​respectively;

[0060] In this way, the blank 3D point cloud PC j Assign values ​​to the three channels of all points in the color point cloud Thus, we can get all the colored point clouds at J positions.

[0061] (4) Binocular vision positioning and tracking system through image and Calculate the 3D coordinates of the marker points in the 3D point cloud reconstruction system to obtain the 3D point cloud data of the marker points

[0062] 4.1) Extracting images and Coordinates of the circle marking points

[0063] 4.1.1) In the binocular vision positioning and tracking system, the image and Binarization is performed according to the threshold value. Pixel values ​​greater than the threshold value are set to 1, and those less than the threshold value are set to 0 to obtain a black and white image. and

[0064] 4.1.2) Image and Select a 3×3 or 5×5 filter template and perform median filtering or Gaussian filtering to obtain an image and

[0065] 4.1.3) Image and Perform edge detection and contour extraction to obtain the edge contour of each object in the image, and judge the edge contour image according to the following criteria:

[0066] Size guide: S min <S r <S max , where S r is the number of edge pixels in the edge contour image, S min and S max They are the minimum and maximum number of edge pixels set respectively;

[0067] Roundness criterion: R 1 <C=T 2 / S<R 2 , T is the perimeter of the edge point in the edge contour image, S is the area, C is the roundness, R1 and R 2 The parameters to be set;

[0068] If it is satisfied, the edge point in the edge contour image is the marked circular edge point, and the algebraic distance from the edge point in the edge contour image to the fitted ellipse is taken as the error ε, and the points that meet the error criterion are retained:

[0069]

[0070] in, represents the average error of the contour points, ε max Indicates the maximum allowable error, ε mean is the allowable average error;

[0071] Calculate the center of the circular edge contour of each marking point, that is, the center of the marking point;

[0072] So for the image and Get the point set consisting of the center of the marked point and Among them, i represents the mark point number, Respectively represent images and The coordinates of the center of the i-th marked point;

[0073] 4.2) Point Set and Match the center of the circle with the marked point in the middle to get the matching point pair set Mate jTw :

[0074] If the point set The center of the i-th marked point in The center of the i′th marked point in satisfies the limit constraint:

[0075]

[0076] Then it is a matching point pair, and the matching point pair set Mate is obtained jTw ,in, To set the threshold;

[0077] 4.3) According to the matching point set Mate jTw According to the principle of binocular vision measurement, the three-dimensional coordinates of the center of the marked point are calculated to obtain a three-dimensional point cloud.

[0078] (5) Through 3D point cloud Fitting the 3D point cloud of the center point of the marker sphere

[0079] Using clustering algorithm to determine the 3D point cloud The points in belong to the same marked ball, and the point set contained in the kth marked ball is obtained. Through point set Fit the 3D coordinates of the sphere center to form a 3D point cloud of the center point of the marked sphere

[0080] For other locations and The same operation is performed on the image to obtain the three-dimensional point cloud of the center point of the marker ball at all positions

[0081] (6) 3D point cloud of the center point of the marker ball at adjacent positions The three-dimensional marker points are roughly matched by Euclidean distance to obtain a set of three-dimensional matching point pairs. Then, we can obtain the set of precise matching point pairs through the constraints of congruent triangles.

[0082] (7) Rigid body transformation to obtain a complete color point cloud

[0083] 7.1) Through the precise matching point pair set Solve the rigid body transformation matrix R j_j+1 ,T j_j+1 ;

[0084] 7.2) Color point cloud obtained from adjacent positions Between two through R j_j+1 ,T j_j+1 Perform a rigid transformation:

[0085]

[0086] is the point cloud at position j+1 The point cloud point set converted to the coordinate system of position j is continuously fused forward in pairs. The point cloud at position j+1, j=1,…, J-1 is finally converted to the coordinate system of the point cloud at position 1 to obtain a fused complete color point cloud.

[0087] The object of the invention of the present invention is achieved in this way:

[0088] The present invention is directed to a method for reconstructing the global three-dimensional morphology and temperature field of a high-temperature object using two sets of binocular camera systems for three-dimensional point cloud reconstruction and point cloud stitching. The three-dimensional point cloud reconstruction system uses a projection surface structured light to project onto the surface of the high-temperature object, calculates the surface phase of the object by a phase shift method, and calculates the three-dimensional point cloud of the measured high-temperature object using the phase as an intermediary. An additional color camera is added to obtain a low-resolution color image, and color temperature information is obtained by operations such as image registration, interpolation, and colorimetric temperature measurement, so as to colorize the point cloud to obtain a color point cloud containing temperature information. A fixed marker sphere is arranged around the three-dimensional point cloud reconstruction system, and the sphere contains circular marker points. The binocular vision positioning and tracking system shoots the marker point frame of the front binocular system, i.e., the three-dimensional point cloud reconstruction system, at a fixed point, and calculates the three-dimensional coordinates of the marker points by matching the left and right image marker points. The three-dimensional coordinates of the center of the marker sphere are fitted according to the three-dimensional coordinates of the marker points. The three-dimensional point cloud reconstruction system changes its own position, shoots the measured high-temperature object at different positions, and calculates the three-dimensional point cloud of the measured high-temperature object at different positions. At the same time, the binocular vision positioning and tracking system calculates the 3D point cloud of the marker ball on the 3D point cloud reconstruction system at each position. By matching the 3D point cloud of the marker ball, the rigid body transformation relationship R and T of each point cloud obtained by the front binocular system is calculated to achieve point cloud splicing and obtain the overall color point cloud.

[0089] The present invention uses a binocular vision positioning and tracking system to calculate the positional relationship between point clouds obtained by a three-dimensional point cloud reconstruction system. The number of matching points is small, the amount of calculation is small, and the accuracy is high. It overcomes the problem that the point cloud features of the measured object are not obvious and the matching is difficult due to the excessive number of point clouds, and greatly improves the speed and accuracy of measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 It is a flowchart of a specific implementation method of the global three-dimensional morphology and temperature field reconstruction method of a high-temperature object according to the present invention;

[0091] Figure 2 It is a schematic diagram of the image acquisition principle of a specific implementation method of the global three-dimensional morphology and temperature field reconstruction method for a high-temperature object of the present invention;

[0092] Figure 3 is a specific example diagram of a phase image;

[0093] Figure 4 is a specific example diagram of an absolute phase image;

[0094] Figure 5 It is a specific example of a three-dimensional point cloud of the surface morphology of a high-temperature object;

[0095] Figure 6 is a specific example diagram of a color image acquired by a low-resolution color camera;

[0096] Figure 7 It is a specific example diagram of a high-resolution color image;

[0097] Figure 8 It is a specific example of a color point cloud;

[0098] Fig. 9 It is a schematic diagram of point cloud distance consistency;

[0099] Fig.10 This is a schematic diagram of the precise matching points obtained by constraining congruent triangles. DETAILED DESCRIPTION

[0100] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0101] In the face of the difficulty in matching the morphology and problem measurement of high-temperature objects, as well as the problem of large matching errors caused by deformation, we need to find a method in the imaging system other than the high-temperature object itself. Since it is difficult to directly use the point cloud of the high-temperature object for stitching, we can consider adding additional markers to the binocular structured light system as a three-dimensional point cloud reconstruction system, and then use another binocular vision system as a binocular vision positioning and tracking system to track and calculate the markers on the three-dimensional point cloud reconstruction system. In this way, the position of the binocular structured light system is used to indirectly replace the position of the high-temperature point cloud, and the relationship between the different positions of the three-dimensional point cloud reconstruction system is also the matching relationship between the high-temperature point clouds. Moreover, the method of using markers for point cloud stitching has higher accuracy in recognition and matching, and the markers are less than the high-temperature point cloud data, so the calculation speed is fast. By improving the binocular vision system in this way, the accuracy and speed of point cloud stitching can be guaranteed. The complete high-temperature point cloud data obtained can help to fully understand the behavior and performance of the object and provide effective support for engineering practice and scientific research.

[0102] Figure 1 It is a flow chart of a specific implementation method of the global three-dimensional morphology and temperature field reconstruction method of a high-temperature object according to the present invention.

[0103] In this embodiment, if Figure 1 As shown, the method for reconstructing the global three-dimensional morphology and temperature field of a high-temperature object of the present invention comprises the following steps:

[0104] Step S1: Image acquisition

[0105] Step S1.1: Construct 3D point cloud reconstruction system and binocular vision positioning and tracking system

[0106] In this embodiment, if Figure 2As shown, a three-dimensional point cloud reconstruction system is constructed: two high-resolution grayscale cameras, namely a high-resolution grayscale left camera and a high-resolution grayscale right camera, are installed at the left and right ends of the horizontal device and a certain distance is kept, the optical machine is installed in the middle of the two high-resolution grayscale cameras, and the low-resolution color camera is installed above the optical machine. At the same time, a marker point frame is designed, and the two high-resolution grayscale cameras, the optical machine and the low-resolution color camera are surrounded by the marker point frame. The marker point frame contains multiple marker balls, and the triangles formed by three different marker balls on the marker point frame are different and unique. The marker ball contains multiple rhombus surfaces, and each rhombus surface has a circular marker point.

[0107] Build a binocular vision positioning and tracking system: Install two other high-resolution grayscale cameras at both ends of the horizontal fixed device and keep a certain distance between them as a binocular vision positioning and tracking system.

[0108] Step S1.2: Ensure that the optical machine in the 3D point cloud reconstruction system can fully and clearly project structured light onto the surface of the high-temperature object being measured, and that the two high-resolution grayscale cameras and low-resolution color cameras can normally, completely and clearly capture the high-temperature object being measured. Ensure that the two high-resolution grayscale cameras in the binocular vision positioning and tracking system can normally, completely and clearly capture the marker point framework.

[0109] Step S1.3: First, adjust the posture of the calibration plate and use Zhang’s calibration method to complete the calibration of the camera in the 3D point cloud reconstruction system and the binocular vision positioning and tracking system.

[0110] Step S1.4: For the 3D point cloud reconstruction system, photograph the high temperature object to be measured at J positions:

[0111] For the jth position, j = 1, 2, ..., J, first, a low-resolution color camera is used to shoot the high-temperature object to be measured, and the obtained image is recorded as image The high-temperature object to be measured is photographed with a high-resolution grayscale left camera, and the obtained image is recorded as image Then, an optical machine is used to project N phase-shift code patterns and K complementary Gray code patterns onto the surface of the high-temperature object to be measured. For each projected structured light pattern, the high-temperature object to be measured is photographed by left and right high-resolution grayscale cameras, and the obtained images are recorded as images and images The image and images The first N images of each are phase shift code patterns, and the last K images are Gray code patterns.

[0112] For the binocular vision positioning and tracking system, the 3D point cloud reconstruction system is tracked and photographed at a fixed position. Every time the 3D point cloud reconstruction system adjusts a position, the two high-resolution grayscale cameras in the binocular vision positioning and tracking system, i.e., the high-resolution grayscale left and right cameras, take a frame image of the marked point, recorded as image and

[0113] like Figure 2 As shown in the figure, the main body of the 3D point cloud reconstruction system is a binocular structured light system, and a low-resolution color camera is used to capture temperature and color information. In order to facilitate the binocular vision positioning and tracking system to identify the 3D point cloud reconstruction system, a marker ball is set around the binocular structured light system to form a frame. This system, as a front-end system, uses structured light technology to generate a point cloud of the measured high-temperature object. The main body of the binocular vision positioning and tracking system is a high-precision binocular grayscale camera, which measures the 3D coordinates of the marker ball for point cloud stitching.

[0114] The measurement principle of this system is: the 3D point cloud reconstruction system projects surface structured light to scan the surface of the object to be measured, calculates the 3D coordinates of the surface of the high-temperature object to be measured by shooting with a binocular grayscale camera, obtains color temperature information by shooting with a color camera, and fuses them to obtain a 3D color point cloud model. At the same time, the binocular vision positioning and tracking system synchronously shoots the 3D point cloud reconstruction system, uses the marker point as the transformation intermediary of the 3D point cloud splicing coordinate system, and transforms the 3D point cloud model of each position to the reference coordinate system of the initial position. The surface structured light traverses and scans the entire high-temperature object to be measured, and the 3D point cloud model information of the entire high-temperature object to be measured can be obtained. We select the left camera coordinate system of the binocular vision positioning and tracking system and the 3D point cloud reconstruction system as the reference coordinate system.

[0115] Step S2: Calculate the three-dimensional blank point cloud PC of the high-temperature object being tested j

[0116] At each location, the 3D point cloud reconstruction system uses images and Calculate the 3D blank point cloud PC of the high temperature object being measured j :

[0117] Step S2.1: Images taken at each position and Computational phase image and

[0118]

[0119]

[0120] in, Respectively, images The pixel value at position (u,v), Phase image and The pixel value at position (u,v).

[0121] To avoid the one-to-many matching phenomenon, the phase image and Expand into absolute phase images

[0122]

[0123]

[0124] in, is the stripe level:

[0125]

[0126]

[0127] in, and The first K-1 frames and the K frames are complementary Gray code images respectively The code value obtained by converting Gray code to decimal is: and The first K-1 frames and the K frames are complementary Gray code images respectively The code value obtained by converting Gray code to decimal.

[0128] The 3D point cloud reconstruction system projects structured light onto the surface of the object at each position and uses the phase shift method to obtain the relative phase. Figure 3 In the figure, we can see that the relative phase has stripes from deep to shallow, indicating that it is a periodic function along the x-axis, with a range of (-π,π). Periodic functions have the same value in different periods. To avoid ambiguity, the relative phase is expanded into the absolute phase of a single period, such as Figure 4 As shown. Figure 4 It can be seen that in the direction of the x-axis, the absolute phase is a pattern of deep and shallow, indicating that it is an increasing function of a single period as a whole, and the value is unique, avoiding one-to-many mismatching in phase matching.

[0129] Step S2.2: Left and right absolute phase images and Perform epipolar correction to obtain the corrected image and The reprojection matrix Q is calculated and returned synchronously during the epipolar correction process j :

[0130]

[0131] Among them, the focal lengths of the calibrated high-resolution grayscale left and right cameras are the same, both f, c v is the coordinate of the principal point of the high-resolution grayscale left camera in the image, c v is the coordinate of the high-resolution grayscale right camera principal point in the image, T u is the baseline length.

[0132] Step S2.3: Phase matching generates disparity and calculates the three-dimensional point coordinates through disparity

[0133] For pixels By comparing the numerical method in the image Find its corresponding position in , if it satisfies the judgment formula:

[0134]

[0135] Then it means (u l ,v l ) and (u r ,v r ) is the matching point.

[0136] The 3D point cloud reconstruction system is a binocular vision system, with a high-resolution grayscale left camera as the reference, and the disparity map size and image The disparity value is the number of columns v of matching points of the high-resolution grayscale left camera. l Subtract the number of columns v of high-resolution grayscale right camera matching points r :

[0137] disp(u l ,v l )=v l -v r

[0138] According to the disparity and reprojection matrix Q j The formula for calculating the coordinates of a 3D point is:

[0139]

[0140] The three-dimensional coordinates are (X / W, Y / W, Z / W), for the image and Calculate the three-dimensional coordinates of all points to get the three-dimensional blank point cloud PC j .

[0141] With absolute phase as the medium, the disparity map is obtained according to the principle of epipolar correction and phase matching, and the three-dimensional point cloud of the surface morphology of the high-temperature object is calculated based on it, such as Figure 5 The reconstructed 3D point cloud model can clearly show the appearance characteristics of the specimen and meet the requirements of the theoretical method.

[0142] Step S3: Reconstructing the 3D point cloud in the image and 3D blank point cloud PC j Temperature information fusion

[0143] Step S3.1: For images at each position and Calculate the optimal transformation matrix H jg Make the image Convert to Image In the coordinate system

[0144] Step S3.1.1: Use ORB algorithm to process the image and Extract and describe feature points to obtain feature point sets and the corresponding feature descriptor set

[0145] Step S3.1.2: Get the feature descriptor set and Perform rough matching to obtain a set of matching feature point pairs And filter it to get a more accurate set of matching point pairs For feature points, the Hamming matrix between corresponding descriptors is used as a measure of similarity. The smaller the Hamming distance, the more similar the feature points are. The matching points with too large distance are filtered out to obtain a set of matching point pairs.

[0146] Step S3.1.3: Calculate the optimal transformation matrix H by using the RANSAC algorithm to remove mismatches jg : Randomly select 4 points to calculate the current transformation matrix model And calculate the cost function to update the good internal point data set, and iterate until the predetermined number of iterations is met, so as to calculate the optimal transformation matrix H jg .

[0147] Step S3.1.4: For the high-temperature objects with unclear texture and complex background, the ORB algorithm cannot accurately identify the correct feature points. The correct matching points are added by manual point selection to improve the optimal transformation matrix H. jg Accuracy of calculations.

[0148] Step S3.1.5: Optimal transformation matrix H jg Low-resolution color images can be The coordinate system is transformed to the high-resolution grayscale left image 's coordinate system.

[0149] Step S3.2: While performing coordinate system conversion, interpolation is performed to improve the low-resolution color image Resolution, get the left image in high resolution grayscale High-resolution color images in coordinate systems

[0150] For each point of the high-resolution color image Perform reverse mapping and interpolation to obtain the corresponding points of the low-resolution color image Among them, M c ,N c Represent low-resolution color images The width and height of Represent the high-resolution grayscale left image Width and height of high-resolution color images and high resolution grayscale left image The same size:

[0151]

[0152]

[0153] in, Represents a high-resolution color image The horizontal and vertical coordinates of a point in the c ,v c ) represents a low-resolution color image The horizontal and vertical coordinates of the corresponding points in .

[0154] If the converted point exceeds the low-resolution color image The image boundary is The corresponding points on the low-resolution color image are discarded and assigned a value of 0. The image boundaries will correspond to the horizontal and vertical coordinates (u c ,v c ) is assigned to the corresponding point in the high-resolution color image. For the horizontal and vertical coordinates The calculated horizontal and vertical coordinates (u c ,v c ) is a floating point number, which is processed according to the bicubic interpolation algorithm to obtain the corresponding pixel value and assign it to the corresponding point of the high-resolution color image This results in a high-resolution color image

[0155] Step S3.3: High-resolution color image Perform epipolar correction to obtain the corrected image

[0156] Step S3.4: Image The temperature information image is obtained using the colorimetric temperature measurement method, recorded as image

[0157] Step S3.5: PC for blank 3D point cloud j Each point (x, y, z) in the image is converted from a three-dimensional point to a pixel coordinate system to obtain the corresponding Coordinates on (u,v):

[0158]

[0159]

[0160] Among them, Z is the coefficient and int() means rounding.

[0161] In high-resolution color images Get the value of [RGB] at the coordinate (u,v) for the blank 3D point cloud PC j The three channels of the point (x, y, z) in are assigned values ​​respectively.

[0162] In this way, the blank 3D point cloud PC j Assign values ​​to the three channels of all points in the color point cloud Thus, we can get all the colored point clouds at J positions.

[0163] In addition, a low-resolution color camera is used to obtain color images such as Figure 6 As shown in Figure 1, compared with the high-precision grayscale left image, its size is smaller and the coordinate system does not meet the theoretical requirements. According to the image registration method, after unifying the image coordinate system and resolution, the high-resolution color image is obtained as shown in Figure 1. Figure 7 As shown in the figure. The size and coordinate system of the high-resolution color image are consistent with the high-precision grayscale left image. The temperature information of the high-temperature object surface is calculated by the colorimetric temperature measurement method, and it is fused with the three-dimensional point cloud according to the mapping relationship to obtain the color temperature point cloud, as shown in the figure. Figure 8 shown.

[0164] Step S4: The binocular vision positioning and tracking system uses images and Calculate the 3D coordinates of the marker points in the 3D point cloud reconstruction system to obtain the 3D point cloud data of the marker points

[0165] Step S4.1: Extracting images and Coordinates of the circle marking points

[0166] Step S4.1.1: In the binocular vision positioning and tracking system, and Binarization is performed according to the threshold value. Pixel values ​​greater than the threshold value are set to 1, and those less than the threshold value are set to 0 to obtain a black and white image. and Binarization is:

[0167]

[0168] Among them, thd is the threshold, p(u,v) is the pixel value of the image before binarization, and B(u,v) is the pixel value of the image after binarization.

[0169] Step S4.1.2: Image and Select a 3×3 or 5×5 filter template and perform median filtering or Gaussian filtering to obtain an image and

[0170] Step S4.1.3: Image and Perform edge detection and contour extraction to obtain the edge contour of each object in the image, and judge the edge contour image according to the following criteria:

[0171] Size guide: S min <S r <S max , where S r is the number of edge pixels in the edge contour image, S min and S max They are the minimum and maximum number of edge pixels set respectively;

[0172] Roundness criterion: R 1 <C=T 2 / S<R 2 , T is the perimeter of the edge point in the edge contour image, S is the area, C is the roundness, R 1 and R 2 The parameters to be set;

[0173] If it is satisfied, the edge point in the edge contour image is the marked circular edge point, and the algebraic distance from the edge point in the edge contour image to the fitted ellipse is taken as the error ε, and the points that meet the error criterion are retained:

[0174]

[0175] in, represents the average error of the contour points, ε max Indicates the maximum allowable error, ε mean is the average allowable error.

[0176] The center of the circular edge contour of each marking point is calculated, that is, the center of the marking point.

[0177] So for the image and Get the point set consisting of the center of the marked point and Among them, i represents the mark point number, Respectively represent images and The coordinates of the center of the i-th marker point.

[0178] Step S4.2: Point Set and Match the center of the circle with the marked point in the middle to get the matching point pair set Mate jTw :

[0179] If the point set The center of the i-th marked point in The center of the i′th marked point in satisfies the limit constraint:

[0180]

[0181] Then it is a matching point pair, and the matching point pair set Mate is obtained jTw ,in, To set the threshold.

[0182] Step S4.2: According to the matching point set Mate jTw According to the principle of binocular vision measurement, the three-dimensional coordinates of the center of the marked point are calculated to obtain a three-dimensional point cloud.

[0183] Step S5: Through the 3D point cloud Fitting the 3D point cloud of the center point of the marker sphere

[0184] Using clustering algorithm to determine the 3D point cloud The points in belong to the same marked ball, and the point set contained in the kth marked ball is obtained. Through point set Fit the 3D coordinates of the sphere center to form a 3D point cloud of the center point of the marked sphere

[0185] For other locations and The same operation is performed on the image to obtain the three-dimensional point cloud of the center point of the marker ball at all positions

[0186] Step S6: 3D point cloud of the center points of the marker balls at adjacent positions The three-dimensional marker points are roughly matched by Euclidean distance to obtain a set of three-dimensional matching point pairs. Then, we can obtain the set of precise matching point pairs through the constraints of congruent triangles.

[0187] Step S6.1: Due to the invariance of spatial geometric features, the distances between the overlapping area marker points of the high-resolution point clouds of adjacent sub-regions remain unchanged. Therefore, the rough matching is:

[0188] First, based on this property, we construct a point cloud set at position j The distance matrix D G :

[0189]

[0190] in, Represents the 3D point cloud of the center point of the marker sphere The kth point in jD , the distance matrix D G The element is The Euclidean distance between all three-dimensional points in . The distance matrix D G The kth row in represents With point cloud collection The distance vectors of the remaining marked points. The corresponding distance matrix D is also constructed for the adjacent position j+1 H .

[0191] Then, traverse the distance matrix D G The k-th row element and the distance matrix D G The distance similarity between the elements in the k′th row is used to determine the number of elements that meet the distance threshold ε. If count>λ, where λ is the number threshold, then it means and is a pair of matching points. Perform the above judgment operation on all points to achieve rough matching of three-dimensional points and obtain a set of matching point pairs. d j_j+1 represents the number of matching point pairs between position j and position j+1, and k″ represents the sequence number of the matching point pair.

[0192] like Fig. 9As shown in the figure, it is a schematic diagram of 3D point cloud matching based on distance consistency. Adjacent point clouds contain overlapping areas, and the marker points in the overlapping areas are matched one by one. In 3D space, there is a feature of spatial geometric invariance, and the distance between the marker points in the overlapping areas of adjacent point clouds remains unchanged. As can be seen from the figure, and are one-to-one equal, and and The distance consistency relationship is not satisfied.

[0193] Step S6.2: Rough matching based on Euclidean distance may result in a one-to-many situation, which can be further matched using the congruent triangle constraint. Perform precise matching to obtain a more accurate set of matching points for the marked points

[0194] There are multiple marker balls around the 3D point cloud reconstruction system, and the triangles formed by the three different marker balls on the frame are different and unique. In step S6.1, a part of the matching points have been selected by the distance constraint. Here, we will exhaustively list the triangles formed by all the marker balls with known 3D coordinates at this position. Because the 3D points will not have image affine changes and will not be distorted, the triangles at different positions and perspectives should satisfy the congruence relationship. Those that do not satisfy the congruence relationship will be deleted from the set of matching point pairs.

[0195] Based on the set of matching point pairs obtained in the previous section For all matching point pairs, the set is precisely matched with congruent triangle constraints. The specific algorithm process is as follows:

[0196] Step 1: Set the distance threshold α and quantity threshold β;

[0197] Step 2: For The k″th pair of marked points in When k″≠d j_j+1 When , select the l=k″+1th pair of marking point matching pairs, l≤d j_j+1 ; When k″=d j_j+1 When , select l = 1, that is, the first pair of marking points; if Match the marked points to and Constitute a benchmark matching pair and go to step 3; otherwise, let l = l + 1. If the loop iteration continues until l = d j_j+1 If no benchmark matching pair is found, it means that the point set There is no matching pair with the marked point The correctly matched marker point pair reselects a new marker point matching pair for benchmark matching pair judgment;

[0198] Step 3: Set the number of congruent triangles c = 0, and for the remaining d j_j+1 -2 pairs of marked point matching pairs are verified in turn to see whether they and the reference matching pairs form congruent triangle constraints within the error tolerance. If they meet the judgment criteria, then c = c + 1. After verifying the remaining d j_j+1 After 2 pairs of marked points are matched, if c ≥ β, all the marked point matching pairs that meet the congruent triangle constraint under this reference matching point are saved as the final correct marked point set. Otherwise, return to step 2 and reselect the reference matching pair for calculation.

[0199] In the specific process of calculating the constraints of congruent triangles, the reference matching point pair and As point set The remaining markers in the match pair It complies with the following formula:

[0200]

[0201] This means that these three points satisfy the congruent triangle constraint

[0202] Fig.10 The congruent triangle principle is further used to remove the one-to-many mismatched marker points. In the adjacent point clouds with rigid body transformation, the two triangles formed by the three correctly matched marker points in the two point clouds should be congruent. Fig.10 As can be seen in is the reference matching point, the correct matching pair and will form congruent triangles with the reference matching pair, while the wrong matching pair and The correspondence relationship of congruent triangles is not satisfied, and these two points are removed from the point cloud matching.

[0203] Filtering of precise matching algorithm of marked points based on congruent triangle constraints The wrong matching points in the result are used to obtain a more accurate set of matching point pairs.

[0204] Step S7: Rigid body transformation to obtain a complete color point cloud

[0205] Step S7.1: Using a set of precise matching point pairs Solve the rigid body transformation matrix R j_j+1 ,T j_j+1 .

[0206] Step S7.2: Color point cloud obtained for adjacent positions Between two through R j_j+1,T j_j+1 Perform a rigid transformation:

[0207]

[0208] is the point cloud at position j+1 The point cloud point set converted to the coordinate system of position j is continuously fused forward in pairs. The point cloud at position j+1, j=1,…, J-1 is finally converted to the coordinate system of the point cloud at position 1 to obtain a fused complete color point cloud.

[0209] Although the above describes an illustrative specific embodiment of the present invention to facilitate the understanding of the present invention by those skilled in the art, it should be clear that the present invention is not limited to the scope of the specific embodiment. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

Claims

1. A method for reconstructing the global three-dimensional morphology and temperature field of a high-temperature object, comprising: (1) Image acquisition 1.1) Construct a 3D point cloud reconstruction system: install two high-resolution grayscale cameras, i.e., a high-resolution grayscale left camera and a high-resolution grayscale right camera, at the left and right ends of the horizontal device and keep a certain distance between them, install the optical machine in the middle of the two high-resolution grayscale cameras, and install the low-resolution color camera above the optical machine. At the same time, design a marker point frame, wherein the two high-resolution grayscale cameras, the optical machine and the low-resolution color camera are surrounded by the marker point frame, and the marker point frame includes a plurality of marker balls, and the triangles formed by the three different marker balls on the marker point frame are different and unique, and the marker ball includes a plurality of rhombus surfaces, and each rhombus surface has a circular marker point; Construct a binocular visual positioning and tracking system: install two other high-resolution grayscale cameras at both ends of the horizontal fixed device and keep a certain distance between them as a binocular visual positioning and tracking system; 1.2) Ensure that the optical function in the 3D point cloud reconstruction system can fully and clearly project structured light onto the surface of the high-temperature object being measured, and that the two high-resolution grayscale cameras and low-resolution color cameras can normally, completely and clearly capture the high-temperature object being measured, and ensure that the two high-resolution grayscale cameras in the binocular vision positioning and tracking system can normally, completely and clearly capture the marker point framework; 1.3) First, adjust the position and posture of the calibration plate and use Zhang's calibration method to complete the calibration of the camera in the 3D point cloud reconstruction system and the binocular vision positioning and tracking system; 1.4) For the 3D point cloud reconstruction system, take photos of the high-temperature object at J positions: For the jth position, j = 1, 2, ..., J, first, a low-resolution color camera is used to shoot the high-temperature object to be measured, and the obtained image is recorded as image The high-temperature object to be measured is photographed with a high-resolution grayscale left camera, and the obtained image is recorded as image Then, an optical machine is used to project N phase-shift code patterns and K complementary Gray code patterns onto the surface of the high-temperature object to be measured. For each projected structured light pattern, the high-temperature object to be measured is photographed by left and right high-resolution grayscale cameras, and the obtained images are recorded as images and images The image and images The first N images of each are phase-shift code patterns, and the last K images are Gray code patterns; For the binocular vision positioning and tracking system, the 3D point cloud reconstruction system is tracked and photographed at a fixed position. Every time the 3D point cloud reconstruction system adjusts a position, the two high-resolution grayscale cameras in the binocular vision positioning and tracking system, i.e., the high-resolution grayscale left and right cameras, take a frame image of the marked point, recorded as image and (2) At each location, the 3D point cloud reconstruction system uses images and Calculate the 3D blank point cloud PC of the high temperature object being measured j 2.1) Images taken at various locations and Computational phase image and in, Respectively, images The pixel value at position (u,v), Phase image and The pixel value at position (u,v); The phase image and Expand into absolute phase images in, is the stripe level: in, and The first K-1 frames and the K frames of complementary Gray code images are respectively The code value obtained by converting Gray code to decimal is: and The first K-1 frames and the K frames of complementary Gray code images are respectively The code value obtained by converting Gray code to decimal; 2.2) Left and right absolute phase images and Perform epipolar correction to obtain the corrected image and The reprojection matrix Q is calculated and returned synchronously during the epipolar correction process j : Among them, the focal lengths of the calibrated high-resolution grayscale left and right cameras are the same, both f, c v is the coordinate of the principal point of the high-resolution grayscale left camera in the image, c v is the coordinate of the high-resolution grayscale right camera principal point in the image, T u is the baseline length; 2.3) Phase matching generates disparity and calculates the coordinates of 3D points through disparity For pixels By comparing the numerical value method in the image Find its corresponding position in , if it satisfies the judgment formula: Then it means (u l ,v l ) and (u r ,v r ) is the matching point; The 3D point cloud reconstruction system is a binocular vision system, with a high-resolution grayscale left camera as the reference, and the disparity map size and image The disparity value is the number of columns v of matching points of the high-resolution grayscale left camera. l Subtract the number of columns v of high-resolution grayscale right camera matching points r : disp(u l ,v l )=v l -v r According to the disparity and reprojection matrix Q j The formula for calculating the coordinates of a 3D point is: The three-dimensional coordinates are (X / W, Y / W, Z / W), for the image and Calculate the three-dimensional coordinates of all points to get the three-dimensional blank point cloud PC j ; (3) In the 3D point cloud reconstruction system, images are used and 3D blank point cloud PC j Temperature information fusion 3.1) For images at each position and Calculate the optimal transformation matrix H jg Make the image Convert to Image In the coordinate system 3.1.1) Use ORB algorithm to analyze the image and Extract and describe feature points to obtain feature point sets and the corresponding feature descriptor set 3.1.2) The obtained feature descriptor set and Perform rough matching to obtain a set of matching feature point pairs And filter it to get a more accurate set of matching point pairs For feature points, the Hamming matrix between corresponding descriptors is used as a measure of similarity. The smaller the Hamming distance, the more similar the feature points are. The matching points with too large distance are filtered out to obtain a set of matching point pairs. 3.1.3) Calculate the optimal transformation matrix H by using the RANSAC algorithm to remove mismatches jg : Randomly select 4 points to calculate the current transformation matrix model And calculate the cost function to update the good internal point data set, and iterate until the predetermined number of iterations is met, so as to calculate the optimal transformation matrix H jg ; 3.1.4) Add correct matching points by manual selection; 3.1.5) Use the optimal transformation matrix H jg Convert low-resolution color images The coordinate system is transformed to the high-resolution grayscale left image In the coordinate system of ; 3.2) While performing coordinate system transformation, interpolation improves low-resolution color images Resolution, get the left image in high resolution grayscale High-resolution color images in coordinate systems For each point of the high-resolution color image Perform reverse mapping and interpolation to obtain the corresponding points of the low-resolution color image in, M c ,N c Represent low-resolution color images The width and height of Represent the high-resolution grayscale left image Width and height of high-resolution color images and high resolution grayscale left image The same size: in, Represents a high-resolution color image The horizontal and vertical coordinates of a point in the c ,v c ) represents a low-resolution color image The horizontal and vertical coordinates of the corresponding points in the If the converted point exceeds the low-resolution color image The image boundary of The corresponding points on the low-resolution color image are discarded and assigned a value of 0. The image boundaries will correspond to the horizontal and vertical coordinates (u c ,v c ) is assigned to the corresponding point in the high-resolution color image. For the horizontal and vertical coordinates The calculated horizontal and vertical coordinates (u c ,v c ) is a floating point number, which is processed according to the bicubic interpolation algorithm to obtain the corresponding pixel value and assign it to the corresponding point of the high-resolution color image This results in a high-resolution color image 3.3) For high-resolution color images Perform epipolar correction to obtain the corrected image 3.4) Image The temperature information image is obtained using the colorimetric temperature measurement method, recorded as image 3.5) For blank 3D point cloud PC j Each point (x, y, z) in the image is converted from a three-dimensional point to a pixel coordinate system to obtain the corresponding Coordinates on (u,v): Among them, Z is the coefficient, int() means rounding; In high-resolution color images Get the value of [RGB] at the coordinate (u,v) for the blank 3D point cloud PC j The three channels of the point (x, y, z) in are assigned values ​​respectively; In this way, the blank 3D point cloud PC j Assign values ​​to the three channels of all points in the color point cloud Thus, we can get all the colored point clouds at J positions. (4) Binocular vision positioning and tracking system through image and Calculate the 3D coordinates of the marker points in the 3D point cloud reconstruction system to obtain the 3D point cloud data of the marker points 4.1) Extracting images and Coordinates of the circle marking points 4.1.1) In the binocular vision positioning and tracking system, the image and Binarization is performed according to the threshold value. Pixel values ​​greater than the threshold value are set to 1, and those less than the threshold value are set to 0 to obtain a black and white image. and 4.1.2) Image and Select a 3×3 or 5×5 filter template and perform median filtering or Gaussian filtering to obtain an image and 4.1.3) Image and Perform edge detection and contour extraction to obtain the edge contour of each object in the image, and judge the edge contour image according to the following criteria: Size guide: S min <S r <S max ,in, S r is the number of edge pixels in the edge contour image, S min and S max They are the minimum and maximum number of edge pixels set respectively; Roundness criterion: R1<C=T 2 / S<R2, T is the perimeter of the edge point in the edge contour image, S is the area, C is the roundness, R1 and R2 are the set parameters; If it is satisfied, the edge point in the edge contour image is the marked circular edge point, and the algebraic distance from the edge point in the edge contour image to the fitted ellipse is taken as the error ε, and the points that meet the error criterion are retained: in, represents the average error of the contour points, ε max Indicates the maximum allowable error, ε mean is the allowable average error; Calculate the center of the circular edge contour of each marking point, that is, the center of the marking point; So for the image and Get the point set consisting of the center of the marked point and Among them, i represents the mark point number, Respectively represent images and The coordinates of the center of the i-th marked point; 4.2) Point Set and Match the center of the circle with the marked point in the middle to get the matching point pair set Mate jTw : If the point set The center of the i-th marked point in The center of the i′th marked point in satisfies the limit constraint: Then it is a matching point pair, and the matching point pair set Mate is obtained jTw ,in, To set the threshold; 4.3) According to the matching point set Mate jTw According to the principle of binocular vision measurement, the three-dimensional coordinates of the center of the marked point are calculated to obtain a three-dimensional point cloud. (5) Through 3D point cloud Fitting the 3D point cloud of the center point of the marker sphere Using clustering algorithm to determine the 3D point cloud The points in belong to the same marked ball, and the point set contained in the kth marked ball is obtained. Through point set Fit the 3D coordinates of the sphere center to form a 3D point cloud of the center point of the marked sphere For other locations and The same operation is performed on the image to obtain the three-dimensional point cloud of the center point of the marker ball at all positions (6) 3D point cloud of the center point of the marker ball at adjacent positions The three-dimensional marker points are roughly matched by Euclidean distance to obtain a set of three-dimensional matching point pairs. Then, we can obtain the set of precise matching point pairs through the constraints of congruent triangles. (7) Rigid body transformation to obtain a complete color point cloud 7.1) Through the precise matching point pair set Solve the rigid body transformation matrix R j_j+1 ,T j_j+1 ; 7.2) Color point cloud obtained from adjacent positions Between two through R j_j+1 ,T j_j+1 Perform a rigid transformation: is the point cloud at position j+1 The point cloud point set converted to the coordinate system of position j is continuously fused forward in pairs. The point cloud at position j+1, j=1,…, J-1 is finally converted to the coordinate system of the point cloud at position 1 to obtain a fused complete color point cloud.

2. The method for reconstructing the global three-dimensional morphology and temperature field of a high-temperature object according to claim 1, characterized in that: The rough matching described in step (6) is: First, construct a point cloud set at position j The distance matrix D G : in, Represents the 3D point cloud of the center point of the marker ball The kth point in jD , the distance matrix D G The middle element is The Euclidean distance between all three-dimensional points in the distance matrix D G The kth row in represents With point cloud collection The distance vectors of the remaining marked points and the corresponding distance matrix D of the adjacent position j+1 are also constructed H ; Then, traverse the distance matrix D G The k-th row element and the distance matrix D G The distance similarity between the elements in the k′th row is used to determine the number of elements that meet the distance threshold ε. If count>λ, where λ is the number threshold, then it means and is a pair of matching points, a point cloud set Perform the above judgment operation on all points to achieve rough matching of three-dimensional points and obtain a set of matching point pairs. d j_j+1 represents the number of matching point pairs between position j and position j+1, and k″ represents the sequence number of the matching point pair.

3. The method for reconstructing the global three-dimensional morphology and temperature field of a high-temperature object according to claim 1, characterized in that: The congruent triangle constraints described in step (6) obtain the set of exact matching point pairs for: Step 1: Set the distance threshold α and quantity threshold β; Step 2: For The k″th pair of marked points in When k″≠d j_j+1 When , select the l=k″+1th pair of marking point matching pairs, l≤d j_j+1 ; When k″=d j_j+1 When , select l = 1, that is, the first pair of marking points; if Match the marked points to and Construct a benchmark matching pair and proceed to step 3; Otherwise, let l = l + 1. If the loop continues to iterate until l = d j_j+1 If no benchmark matching pair is found, it means that the point set There is no matching pair with the marked point The correctly matched marker point pair reselects a new marker point matching pair for benchmark matching pair judgment; Step 3: Set the number of congruent triangles c = 0, and for the remaining d j_j+1 -2 pairs of marked point matching pairs are verified in turn to see whether they and the reference matching pairs form congruent triangle constraints within the error tolerance. If they meet the judgment criteria, then c = c + 1. After verifying the remaining d j_j+1 After 2 pairs of marked points are matched, if c ≥ β, all the marked point matching pairs that meet the congruent triangle constraint under this reference matching point are saved as the final correct marked point set. Otherwise, return to step 2 and reselect the reference matching pair for calculation.

Images