A high-precision plate shape measurement calibration method and system

Abstract

The application provides a high-precision plate shape measurement calibration method and system, and belongs to the technical field of steel plate online measurement.The application aims at the problem that the measurement precision of the laser triangulation method is reduced due to vibration interference during the movement of the steel plate, and adopts a cross laser projection scheme: a horizontal laser line is used to measure the plate shape profile along the width direction, and a vertical laser line is used to obtain height information along the movement direction.A calibration mapping relationship combining a shape-preserving segmented cubic Hermite interpolation and a residual model is constructed during calibration.During measurement, the phase correlation method is used to estimate the sub-pixel level displacement of adjacent frames of vertical data, the vibration amount is extracted by the height difference average of the overlapping area after alignment, and the horizontal plate shape is compensated in real time.The application does not need accurate speed-frame rate matching, the system structure is compact, the rigid body vibration interference can be effectively inhibited, the plate shape measurement precision is significantly improved, and the application is suitable for high-precision online detection of moving steel plates such as hot-rolled steel strips.

Description

Technical Field

[0001] This invention belongs to the field of online steel plate measurement technology, specifically a high-precision plate shape measurement and calibration method and system. Background Technology

[0002] In the steel plate production process, accurate online measurement of the steel plate profile is a crucial step in quality control. Laser triangulation, due to its advantages of being non-contact, highly accurate, and having a fast response, is widely used in steel plate contour measurement. This method projects a laser line onto the steel plate surface, captures the laser line image with a camera, and calculates the height information of the steel plate surface based on the principle of triangulation. However, the measurement accuracy of traditional laser triangulation is limited by various factors such as installation errors or calibration accuracy.

[0003] Furthermore, vibration suppression during steel plate movement is another crucial factor limiting measurement accuracy. Specifically, vibration in the direction of movement is inevitable during steel plate movement. Existing technologies have attempted vibration compensation using a single camera and dual laser lines. This approach envisions that in two adjacent frames, the second laser line in the previous frame and the first laser line in the subsequent frame spatially overlap, eliminating vibration by comparing the height difference of the overlapping area. However, this method requires precise matching between the steel plate's movement speed and the camera's frame rate to ensure the two laser lines precisely overlap. In actual production, steel plate speed fluctuates, making precise speed-frame-rate matching difficult, thus hindering the practical application of this method.

[0004] Another technical approach is to add additional displacement sensors (such as laser displacement gauges, encoders, etc.) to the measurement system and compensate by synchronously acquiring vibration signals. However, this method increases system cost and installation complexity, and requires solving the problem of multi-sensor data synchronization, which is not conducive to its widespread application on existing production lines.

[0005] In summary, existing laser triangulation techniques for measuring steel plate profiles either have stringent requirements for velocity matching or are complex and costly, making it difficult to balance accuracy and practicality in online detection scenarios with limited installation space and frequent velocity fluctuations. Therefore, a technical solution is needed that is structurally simple, requires no precise velocity matching, can independently extract vibration information, and provides high-precision compensation for transverse plate profiles. Summary of the Invention

[0006] This invention addresses the technical problems of limited calibration accuracy in existing plate shape measurement methods, and the complexity of systems caused by reliance on velocity matching or the addition of extra sensors for vibration compensation. It proposes a high-precision plate shape measurement and calibration method and system. This invention simultaneously projects transverse and longitudinal laser lines, using the longitudinal laser line to independently measure vibration information, thereby compensating for transverse plate shape. This eliminates the need for additional measurement planes or sensors, making it suitable for online detection scenarios with limited installation space.

[0007] To achieve the above objectives, the present invention provides a high-precision plate shape measurement and calibration method, comprising the following steps:

[0008] Step S1, Calibration Stage: The horizontal and vertical laser lines are calibrated using a standard lifting platform to establish the mapping relationship between the pixel value at each pixel position and the actual height. The horizontal laser line is arranged along the width direction of the steel plate, and the vertical laser line is arranged along the movement direction of the steel plate.

[0009] Step S2, Actual Measurement Stage: Simultaneously project cross-shaped horizontal and vertical laser lines onto the surface of the moving steel plate, and continuously acquire images containing both laser lines;

[0010] Step S3, Height Calculation Stage: For each frame of image acquired, the sub-pixel positions of the horizontal and vertical laser lines are extracted respectively, and based on the mapping relationship established in step S1, the sub-pixel positions are converted into the corresponding actual height distribution, thereby obtaining the horizontal plate contour data and the vertical height data.

[0011] Step S4, Vibration Extraction Stage: Process the longitudinal height data sequence corresponding to two adjacent frames, use the phase correlation method to estimate the sub-pixel level longitudinal displacement between the two frames, and calculate the average height difference of the overlapping area after aligning the two frames based on the displacement. Use this average value as the steel plate vibration displacement of the current frame relative to the previous frame.

[0012] Step S5, Plate Shape Correction Stage: The calculated vibration displacement is used to compensate for the transverse plate shape profile data at the same moment, eliminating vibration interference and obtaining the true steel plate shape.

[0013] Furthermore, the specific process for establishing the mapping relationship between pixel values ​​and actual height in step S1 is as follows:

[0014] Step S11: For each laser line, within the calibration stroke, move the standard lifting platform at equal intervals, record the sub-pixel value of each pixel on the laser line at each calibration height, and form a calibration matrix;

[0015] Step S12: For each fixed pixel position on the laser line, construct an independent conformal piecewise cubic Hermite interpolation PCHIP function with its sub-pixel value corresponding to each calibrated height as the independent variable and the actual height as the dependent variable. Then, derive each sub-interval according to the Hermite interpolation conditions. The cubic polynomial expression on:

[0016]

[0017] in, For pixels The calculated height at that location, These are the polynomial coefficients;

[0018] Store the polynomial coefficients for each subinterval The calibration results are used for online height calculation;

[0019] Step S13: Using the calibration matrix, calculate the residual between the calculated height and the actual height of the PCHIP function at each calibration point, and establish a quadratic residual model with sub-pixel values ​​as independent variables:

[0020] ;

[0021] in, This is the residual compensation value. Subpixel values, These are the coefficients of the quadratic residual model;

[0022] Step S14: The PCHIP function and the quadratic residual model together constitute the comprehensive mapping relationship of the pixel position.

[0023] Furthermore, in the height calculation stage, converting sub-pixel positions into actual heights includes:

[0024] Step 31: Based on the sub-pixel value to be converted, locate the sub-interval where the PCHIP function is located at the corresponding pixel position, and substitute the sub-pixel value into the cubic polynomial of the sub-interval to calculate the initial height value.

[0025] Step S32: Substitute the subpixel value into the quadratic residual model suggested in step S13 to obtain the residual compensation value;

[0026] Step S33: Add the initial height value to the residual compensation value to obtain the final actual height.

[0027] Furthermore, step S4, which uses the phase correlation method to calculate the inter-frame sub-pixel displacement estimation, specifically includes...

[0028] Step S41: Preprocess the vertical height data of two adjacent frames, including smoothing filtering or median filtering, and subtract their respective mean values ​​to eliminate the overall height offset.

[0029] Step S42: Calculate the Fourier transform of the two preprocessed sequences, and then calculate the normalized cross power spectrum.

[0030] Step S43: Perform an inverse Fourier transform on the normalized cross-power spectrum to obtain the phase correlation function;

[0031] Step S44: Find the integer index position of the peak of the phase correlation function and use it as the integer pixel displacement;

[0032] Step S45: Perform quadratic polynomial fitting using the peak value and its left and right adjacent points to calculate the sub-pixel correction amount. Add the integer pixel displacement to the sub-pixel correction amount to obtain the final sub-pixel level vertical displacement. .

[0033] Furthermore, the sub-pixel correction value obtained from the quadratic polynomial fitting in step S45 Calculate according to the following formula:

[0034]

[0035] in These are the relevant values ​​for integer peak positions. and These are the relevant values ​​of its left and right adjacent points, respectively.

[0036] Furthermore, the calculation process for the vibration displacement of the steel plate in step S4 is as follows:

[0037] Let the vertical height data sequences of two adjacent frames be respectively and The obtained sub-pixel level longitudinal displacement is If subpixel-level displacement estimation Then the previous frame signal Translation The signal after and the next frame Alignment, i.e.

[0038] , n takes values ​​of 0, 1, ..., Nd-1;

[0039] Length of overlapping area after alignment N is the length of a single frame signal. Within the overlapping region, calculate the height difference between corresponding points. Take the average of all differences as the vibration displacement of the current frame relative to the previous frame: .

[0040] Furthermore, the plate shape correction stage specifically includes:

[0041] Let the horizontal plate shape measurement result of the current frame be... The vibration displacement of the current frame relative to the previous frame obtained from the vibration extraction stage is: The corrected true plate height The corrected actual plate height is used as the height coordinate of each point in the width direction of the steel plate. ;

[0042] Coordinates along the direction of steel plate movement Speed ​​of steel plate movement With running time The product is determined as follows: ;

[0043] Coordinates along the width of the steel plate Calculate using the following formula: ;

[0044] in, Current image center pixel coordinates, These are the initial image pixel coordinates. This is the imaging magnification in the width direction. Adjust the scaling factor for the field of view.

[0045] Another technical solution to be protected by this invention: a high-precision plate shape measurement and calibration system, used in the above-mentioned method, comprising:

[0046] The laser projection module is used to simultaneously project mutually perpendicular horizontal and vertical laser lines onto the surface of a moving steel plate.

[0047] The image acquisition module is used to continuously acquire images of the steel plate surface containing the transverse laser lines and the longitudinal laser lines;

[0048] The calibration module is used to establish and store the calibration mapping relationship between sub-pixel values ​​and actual height for the horizontal and vertical laser lines using a standard lifting platform;

[0049] The data processing module includes:

[0050] The height calculation unit is used to extract the sub-pixel position of the laser line from the acquired image and call the calibration mapping relationship to convert it into the actual height to obtain the horizontal plate shape contour data and the vertical height data;

[0051] The vibration extraction unit is used to perform phase correlation operation on the longitudinal height data sequence of adjacent frames, estimate the sub-pixel level longitudinal displacement, and calculate the average value of the height difference of the overlapping area as the vibration displacement of the steel plate after aligning the data based on the displacement.

[0052] The plate shape correction unit is used to compensate the transverse plate shape profile data using the vibration displacement and output the true plate shape data after eliminating vibration interference.

[0053] Preferably, the calibration module is specifically used to: construct a shape-preserving piecewise cubic Hermite interpolation function with sub-pixel values ​​as independent variables and actual height as dependent variables for each fixed pixel position on each laser line, calculate the cubic polynomial coefficients of each sub-interval, and establish a quadratic residual model with sub-pixel values ​​as independent variables. The PCHIP function and the residual model together form a comprehensive calibration mapping table.

[0054] The height calculation unit first calculates the initial height using the PCHIP function by looking up the integrated calibration mapping table, then calculates the compensation value using the residual model, and finally adds the two to obtain the final height.

[0055] Compared with the prior art, the innovations and advantages of this invention are reflected in the following aspects:

[0056] 1. High calibration accuracy: This invention uses the PCHIP interpolation method instead of traditional linear fitting or cubic spline interpolation, generating smooth curves while maintaining the monotonicity of the data and avoiding overshoot. Furthermore, a quadratic residual compensation model is introduced to systematically correct calibration errors, significantly improving the mapping accuracy from sub-pixel values ​​to true height.

[0057] 2. Vibration compensation does not depend on velocity matching: In the cross laser structure, the longitudinal laser line independently measures vibration information. The displacement is calculated by phase correlation matching of longitudinal data of adjacent frames, and then the vibration is calculated. There is no need for precise matching between the steel plate movement speed and the camera frame rate, which overcomes the velocity matching problem in actual production of the dual parallel laser line scheme.

[0058] 3. The system has a compact structure and low cost: it can simultaneously measure plate shape and vibration with only a single camera and a single cross laser, without the need for additional displacement sensors or multiple measurement planes. It has low installation space requirements and is easy to promote and apply on existing production lines.

[0059] 4. Subpixel-level displacement estimation: By using the phase correlation method combined with three-point quadratic fitting subpixel correction, subpixel-level accurate estimation of inter-frame displacement can be achieved, thereby obtaining high-precision vibration compensation effect.

[0060] 5. Effective use of the rigid body vibration assumption: By utilizing the characteristic that the vibration of the steel plate is the vibration of the whole rigid body, the average value of the height difference of each point in the overlapping area is used as the vibration quantity, which can effectively suppress local noise interference and improve the robustness of vibration compensation. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the overall measurement method of the present invention.

[0062] Figure 2 This is a schematic diagram illustrating the nonlinear monotonically increasing relationship between sub-pixel values ​​and actual height during the calibration phase of this invention.

[0063] Figure 3 This is a schematic diagram of the matching results for overlapping positions;

[0064] Figure 4 This is a schematic diagram of measurements taken at two adjacent time points according to the present invention.

[0065] Figure 5 This is a comparison diagram of the plate shape before and after vibration reduction according to the present invention. Detailed Implementation

[0066] The technical solutions in the embodiments of the present invention will now be clearly and completely described in conjunction with the accompanying drawings.

[0067] like Figure 1 As shown, a high-precision plate shape measurement and calibration method is performed according to the following steps:

[0068] Step S1, Calibration Stage

[0069] During calibration, the standard lifting platform is placed within the camera's field of view and adjusted to an initial height of 0 mm. It is then gradually raised at equal intervals of 'a' (in this embodiment, 'a' = 2 mm) until the maximum travel distance 'd' (in this embodiment, 'd' = 50 mm), for a total of m calibrations, where m = d / a + 1 = 26. Due to unavoidable errors in the system and installation, the laser is slightly tilted; therefore, these n values ​​monotonically increase along the laser's tilt direction. However, this error remains constant during calibration and measurement; therefore, all calibration values ​​are taken as the result. The matrix table (m is the number of calibrations).

[0070] Then, based on this matrix, a mapping relationship between the pixel value at each pixel location and the actual height is constructed. For a fixed pixel location k, data pairs are extracted from the calibration matrix:

[0071] (1)

[0072] That is, height Corresponding sub-pixel value These data constitute a one-dimensional interpolation problem, requiring the construction of an interpolation function. The entire process is equivalent to establishing n independent PCHIP interpolation functions:

[0073] (2)

[0074] Each of them satisfy .

[0075] Node and interval definitions:

[0076] Set nodes ,in Corresponding function value Interval length average slope In this embodiment, height samples are taken at equal intervals, with a difference of 2mm between adjacent calibration heights. Figure 2 As shown, the sub-pixel value monotonically increases with height, therefore all .

[0077] Construction of PCHIP interpolation function for a single pixel

[0078] In each interval The PCHIP interpolation function is as follows:

[0079] (3)

[0080] The basis functions are:

[0081] (4)

[0082] The aforementioned basis functions satisfy the boundary conditions of function value being 1 or 0 and derivative being 0 or 1 at the endpoints of the interval, thereby ensuring that the interpolation function accurately satisfies the function value and derivative conditions at the nodes.

[0083] The PCHIP method directly calculates the node slope from discrete calibration data and adaptively adjusts it based on local monotonicity characteristics, thus ensuring the shape preservation of the interpolation curve. In the calibration of this study, the height sampling sequence is equally spaced, the difference between adjacent calibration heights is constant at 2 mm, and the sub-pixel value monotonically increases with height; therefore, the slope... This holds true for all i, providing a monotonicity premise for subsequent calculations. For internal nodes... Its slope is determined by the average slope of two adjacent intervals. and Together, if the two slopes have opposite signs or one of them has a slope of zero, it indicates that a local extremum may exist at the node. In this case:

[0084] (5)

[0085] If both signs are the same, and the node is located within a monotonic interval, take the weighted harmonic average of the slopes of the two intervals:

[0086] (6)

[0087] Where ω is the weighting coefficient and is related to the interval length:

[0088] (7)

[0089] When adjacent intervals are of equal length (in this embodiment, the calibration step size is fixed, therefore...) The above equation degenerates into:

[0090] (8)

[0091] That is, the slope of the node is three times the harmonic average of the average slopes of the two intervals.

[0092] For the left and right endpoints of the interval, the slope is determined by the slope amplitude of the adjacent interval. According to the empirical criterion of "the slope does not exceed three times the average slope" in the conformal interpolation theory, the abnormal behavior of the interpolation curve at the boundary is effectively prevented from being too steep or too gentle. For the system of the present invention, it is ensured that the height near the boundary has a reasonable monotonic outward twitch.

[0093] Substituting the determined node values ​​and node slopes into the Hermite interpolation formula, the coefficients of the cubic polynomial for each interval can be derived. Formula (3) is then applied according to... Expanding and rearranging by powers yields information about... Standard polynomial form:

[0094] (9)

[0095] Expanding from the Hermite interpolation formula, we get:

[0096] (10)

[0097] (11)

[0098] (12)

[0099] (13)

[0100] Store the polynomial coefficients for each subinterval As a calibration result, it is used in the actual measurement stage to locate the interval according to the input subpixel value, and then substitute it into the standard cubic polynomial in formula (9) to calculate the corresponding calculated height.

[0101] Establishment of the residual model

[0102] The interpolation height at each calibration point is calculated using the PCHIP function described above. and the actual height By comparison, the residuals are obtained:

[0103]

[0104] For each fixed pixel position k, a quadratic residual model is established with the subpixel value p as the independent variable and the residual r as the dependent variable:

[0105] (15)

[0106] Using m points in this column of the calibration matrix Solving using the least squares method makes

[0107]

[0108] The coefficient that takes the minimum value The solution method involves constructing a system of normal equations or directly solving the matrix. Ultimately, the comprehensive mapping relationship for this pixel location is formed by the PCHIP function and the residual model:

[0109]

[0110] Steps S2 and S3, actual measurement stage + height calculation stage

[0111] like Figure 4 As shown, during actual measurement, the camera continuously acquires images of the steel plate containing the crosshair laser lines. The sub-pixel centers of the horizontal and vertical laser lines are extracted from each frame to obtain the measured sub-pixel value matrix (size: [missing information]). , (Number of measurements). For each pixel location. and measured subpixel values The height calculation steps are as follows:

[0112] First, positioning Sub-interval: found , making Calculate local offset Substitute the local offset into the cubic polynomial of this interval to solve for the original height:

[0113] (14)

[0114] Then Substituting into the residual model yields the compensation value. Then the final height Performing the above calculations on the transverse and longitudinal laser lines respectively yields the following results:

[0115] Lateral plate outline data: one height vector per frame (The length is equal to the number of pixels on the horizontal laser line).

[0116] Vertical height data: height vectors between two adjacent frames and (The length is equal to the number of pixels on the vertical laser line). To reduce random noise, each row of data can be individually smoothed using a window size of 9 (such as a moving average filter).

[0117] Step S4, Vibration Extraction Stage: Process the longitudinal height data sequence corresponding to two adjacent frames, use the phase correlation method to estimate the sub-pixel level longitudinal displacement between the two frames, and calculate the average height difference of the overlapping area after aligning the two frames based on the displacement. Use this average value as the steel plate vibration displacement of the current frame relative to the previous frame.

[0118] Let the longitudinal height data sequence measured at time T1 and time T2 be as follows:

[0119] (17)

[0120] Because the speed of the steel plate movement is not strictly matched with the camera frame rate, the longitudinal displacement of the steel plate between adjacent frames... Unknown. Meanwhile, the steel plate experiences vertical vibration, causing the same physical point to have different height values ​​in different frames. The method of this invention estimates displacement using the phase correlation method. This allows for the separation of vibration quantities.

[0121] Data preprocessing: Subtract the respective mean values ​​to eliminate overall height offset:

[0122] (18)

[0123] Perform a Fourier transform on the preprocessed discrete signal:

[0124] (19)

[0125] and , All are complex sequences;

[0126] (20)

[0127] The coefficient part is the amplitude spectrum and the exponent part is the phase spectrum;

[0128] set up It is the original signal. It is the translated signal. If we consider their Fourier transforms, then the translation relationship is:

[0129] (twenty one)

[0130] Calculation and normalization of cross-power spectrum:

[0131] (twenty two)

[0132] in for The conjugate complex number; substituting the translation relation, we get:

[0133] (twenty three)

[0134] Normalized cross-power spectrum: The amplitude of each frequency component is normalized to 1, retaining only the phase information.

[0135] (twenty four)

[0136] This step eliminates The effect of this makes the result independent of the specific shape of the signal, retaining only the phase term related to the displacement.

[0137] Calculate the phase correlation function:

[0138] right Performing the inverse Fourier transform yields the phase correlation function in the spatial domain. , where m is an integer pixel displacement used to find the vicinity of the precise value.

[0139] (25)

[0140] Normalized Substituting, we get:

[0141] (26)

[0142] The above formula is a geometric series summation, when hour: The summation result approaches 0.

[0143] Therefore, the phase correlation function is a unit impulse function:

[0144] (27)

[0145] That is, in There is a sharp peak value of 1 at one location, while other locations are close to 0.

[0146] The function of the phase correlation coefficient is to locate the peak value of sub-pixels. For the integer index of the subpixel, the integer bits of the subpixel can be found (by...). (Identify), and at the same time locate the range of precise values. Then, at the peak and the two points before and after, determine the decimal places of the subpixel by fitting a quadratic polynomial at three points. Finally, add the two to obtain the subpixel value of the displacement.

[0147] Since the measured values ​​are sub-pixel values, and d is not an integer, then... Instead of an ideal pulse, it is a Dirichlet kernel (the discrete counterpart of the periodic sin function in the discrete-time Fourier transform). The Dirichlet kernel is the sinc function of the discrete world, which determines the distribution of energy at adjacent integer points during subpixel displacement, allowing the subpixel displacement to be deduced from integer sample values.

[0148] (28)

[0149] Its amplitude is its coefficient part.

[0150] This function is The maximum value is obtained at [location]. Because [this is due to...]. For integer index, when When the true peak is located between two integer sampling points, it is manifested as energy spreading to the nearest integer points, forming a predictable distribution pattern; in order to obtain sub-pixel level displacement estimation, the local region near the peak of the phase correlation function (the peak and its left and right adjacent points) is fitted to construct a quadratic function.

[0151] If the peak position is Then the integer pixel displacement is: ;

[0152] Take three points and its related values Fitting a quadratic function:

[0153] By fitting a quadratic polynomial using three points, and the graph of this polynomial being a parabola, the maximum value of the parabola is obtained, which is the precise sub-pixel decimal point.

[0154] (29)

[0155] The least squares method is used to solve the fitting function, and the peak coordinates with sub-pixel accuracy are obtained by taking the derivative of the fitting function, thereby obtaining the sub-pixel estimate of the translation amount;

[0156] Set peak position , It is the integer peak position. This is the subpixel correction amount.

[0157] (30)

[0158] Final subpixel displacement:

[0159] If we convert the real distance ;in Spatial resolution (mm / pixel)

[0160] Vibration Calculation

[0161] Let the vertical height data sequences of two adjacent frames be respectively and The obtained sub-pixel level longitudinal displacement is If subpixel-level displacement estimation Then the previous frame signal Translation The signal after and the next frame Alignment, i.e.

[0162] ,

[0163] ,

[0164] n takes the value 0, 1, ..., Nd-1;

[0165] Length of overlapping area after alignment Where N is the length of a single frame signal, calculate the height difference between corresponding points within the overlapping region:

[0166] ;

[0167] Take the average of all differences as the vibration displacement of the current frame relative to the previous frame: .

[0168] Step S5, Plate Shape Correction Stage: The calculated vibration displacement is used to compensate for the transverse plate shape profile data at the same moment, eliminating vibration interference and obtaining the true steel plate shape. The plate shape correction stage specifically includes:

[0169] like Figure 5 As shown, let the horizontal plate shape measurement result of the current frame be... The vibration displacement of the current frame relative to the previous frame obtained from the vibration extraction stage is: The corrected true plate height The corrected actual plate height is used as the height coordinate of each point in the width direction of the steel plate. ;

[0170] Coordinates along the direction of steel plate movement Speed ​​of steel plate movement With running time The product is determined as follows: ;

[0171] Coordinates along the width of the steel plate Calculate using the following formula: ;

[0172] in, Current image center pixel coordinates, These are the initial image pixel coordinates. This is the imaging magnification in the width direction. Adjust the scaling factor for the field of view.

[0173] This embodiment employs a cross-shaped laser line for vibration measurement. The transverse laser line is arranged along the width of the steel plate to measure its cross-sectional profile; the longitudinal laser line is arranged along the direction of the steel plate's movement to acquire its longitudinal height information. Since the longitudinal laser line is parallel to the direction of the steel plate's movement, the data collected at different times on the longitudinal laser line correspond to different longitudinal positions on the steel plate's surface as the plate moves. By comparing the similarity of longitudinal laser line data from two adjacent frames, the longitudinal displacement of the steel plate between frames can be estimated. Based on this, comparing the height values ​​of the same physical point at different times yields the vibration information of the steel plate. Using this vibration information to compensate for the transverse laser line measurement results, the true plate shape after eliminating vibration interference can be obtained.

[0174] A high-precision plate shape measurement and calibration system for implementing the above method includes:

[0175] The laser projection module is used to simultaneously project mutually perpendicular horizontal and vertical laser lines onto the surface of a moving steel plate.

[0176] The image acquisition module is used to continuously acquire images of the steel plate surface containing the transverse laser lines and the longitudinal laser lines.

[0177] The calibration module is used to establish and store the calibration mapping relationship between sub-pixel values ​​and actual height for the horizontal and vertical laser lines using a standard lifting platform.

[0178] The calibration module is specifically used to: construct a shape-preserving piecewise cubic Hermite interpolation function with sub-pixel values ​​as independent variables and actual height as dependent variables for each fixed pixel position on each laser line, calculate the cubic polynomial coefficients of each sub-interval, and establish a quadratic residual model with sub-pixel values ​​as independent variables. The PCHIP function and the residual model together form a comprehensive calibration mapping table.

[0179] The data processing module includes:

[0180] The height calculation unit is used to extract the sub-pixel position of the laser line from the acquired image and call the calibration mapping relationship to convert it into the actual height to obtain the horizontal plate shape contour data and the vertical height data;

[0181] The vibration extraction unit is used to perform phase correlation operation on the longitudinal height data sequence of adjacent frames, estimate the sub-pixel level longitudinal displacement, and calculate the average value of the height difference of the overlapping area as the vibration displacement of the steel plate after aligning the data based on the displacement.

[0182] The plate shape correction unit is used to compensate the transverse plate shape profile data using the vibration displacement and output the true plate shape data after eliminating vibration interference.

[0183] Experimental verification:

[0184] Experimental platform setup: laser wavelength 450 nm, power 50 mW; camera resolution 2048×2048, frame rate 200fps; steel plate movement speed 0.5 m / s; calibration stroke 0~50 mm, step size 2 mm. After PCHIP+ residual compensation, the height measurement accuracy reached ±0.02 mm (RMS); the phase correlation method displacement estimation accuracy was 0.05 pixels (corresponding to an actual displacement of 0.005 mm). Figure 5 The comparison of the plate shape before and after vibration reduction is shown: the uncompensated plate shape is superimposed with vibration noise of ±0.3 mm amplitude, while the actual convexity of the compensated plate shape (about 0.15 mm) is clearly visible, and the repeatability is better than ±0.03 mm.

[0185] In the above embodiments, the residual model uses a quadratic polynomial, but a higher-order polynomial or neural network can also be used; PCHIP can be replaced by Akima interpolation; the sub-pixel fitting method can also use the parabolic centroid method. All technical solutions based on cross-lasers, utilizing longitudinal line phase correlation to extract vibration and compensate for transverse plate shape fall within the protection scope of this invention.

[0186] It is understood that the above specific description of the present invention is only for illustrating the present invention and is not limited to the technical solutions described in the embodiments of the present invention. Those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention to achieve the same technical effect; as long as the use needs are met, they are all within the protection scope of the present invention.

Claims

1. A high-precision plate shape measurement and calibration method, characterized in that, Includes the following steps: Step S1, Calibration Stage: The horizontal and vertical laser lines are calibrated using a standard lifting platform to establish the mapping relationship between the pixel value at each pixel position and the actual height. The horizontal laser line is arranged along the width direction of the steel plate, and the vertical laser line is arranged along the movement direction of the steel plate. Step S2, Actual Measurement Stage: Simultaneously project cross-shaped horizontal and vertical laser lines onto the surface of the moving steel plate, and continuously acquire images containing both laser lines; Step S3, Height Calculation Stage: For each frame of image acquired, the sub-pixel positions of the horizontal and vertical laser lines are extracted respectively, and based on the mapping relationship established in step S1, the sub-pixel positions are converted into the corresponding actual height distribution, thereby obtaining the horizontal plate contour data and the vertical height data. Step S4, Vibration Extraction Stage: Process the longitudinal height data sequence corresponding to two adjacent frames, use the phase correlation method to estimate the sub-pixel level longitudinal displacement between the two frames, and calculate the average height difference of the overlapping area after aligning the two frames based on the displacement. Use this average value as the steel plate vibration displacement of the current frame relative to the previous frame. Step S5, Plate Shape Correction Stage: The calculated vibration displacement is used to compensate for the transverse plate shape profile data at the same moment, eliminating vibration interference and obtaining the true steel plate shape.

2. The high-precision plate shape measurement and calibration method according to claim 1, characterized in that: The specific process for establishing the mapping relationship between pixel values ​​and actual height in step S1 is as follows: Step S11: For each laser line, within the calibration stroke, move the standard lifting platform at equal intervals, record the sub-pixel value of each pixel on the laser line at each calibration height, and form a calibration matrix; Step S12: For each fixed pixel position on the laser line, construct an independent conformal piecewise cubic Hermite interpolation PCHIP function with its subpixel value corresponding to each calibrated height as the independent variable and the actual height as the dependent variable. Then, derive each sub-interval according to the Hermite interpolation conditions. The cubic polynomial expression on: ； in, For pixels The calculated height at that location, These are the polynomial coefficients; Store the polynomial coefficients for each subinterval The calibration results are used for online height calculation; Step S13: Using the calibration matrix, calculate the residual between the calculated height and the actual height of the PCHIP function at each calibration point, and establish a quadratic residual model with sub-pixel values ​​as independent variables: ； in, This is the residual compensation value. Subpixel values, These are the coefficients of the quadratic residual model; Step S14: The PCHIP function and the quadratic residual model together constitute the comprehensive mapping relationship of the pixel position.

3. The high-precision plate shape measurement and calibration method according to claim 2, characterized in that: In the height calculation stage, converting sub-pixel positions into actual heights includes: Step 31: Based on the sub-pixel value to be converted, locate the sub-interval where the PCHIP function is located at the corresponding pixel position, and substitute the sub-pixel value into the cubic polynomial of the sub-interval to calculate the initial height value. Step S32: Substitute the subpixel value into the quadratic residual model suggested in step S13 to obtain the residual compensation value; Step S33: Add the initial height value to the residual compensation value to obtain the final actual height.

4. The high-precision plate shape measurement and calibration method according to claim 3, characterized in that: Step S4, which uses the phase correlation method to calculate the inter-frame subpixel displacement estimation, specifically includes... Step S41: Preprocess the vertical height data of two adjacent frames, including smoothing filtering or median filtering, and subtract their respective mean values ​​to eliminate the overall height offset. Step S42: Calculate the Fourier transform of the two preprocessed sequences, and then calculate the normalized cross power spectrum. Step S43: Perform an inverse Fourier transform on the normalized cross-power spectrum to obtain the phase correlation function; Step S44: Find the integer index position of the peak of the phase correlation function, and use it as the integer pixel displacement; Step S45: Perform quadratic polynomial fitting using the peak value and its left and right adjacent points to calculate the sub-pixel correction amount. Add the integer pixel displacement to the sub-pixel correction amount to obtain the final sub-pixel level vertical displacement. .

5. The high-precision plate shape measurement and calibration method according to claim 4, characterized in that: Sub-pixel correction values ​​for quadratic polynomial fitting in step S45 Calculate according to the following formula: ； in These are the relevant values ​​for integer peak positions. and These are the relevant values ​​of its left and right adjacent points, respectively.

6. The high-precision plate shape measurement and calibration method according to claim 1, characterized in that: The calculation process for the vibration displacement of the steel plate in step S4 is as follows: Let the vertical height data sequences of two adjacent frames be respectively and The obtained sub-pixel level longitudinal displacement is If subpixel-level displacement estimation Then the previous frame signal Translation The signal after and the next frame Alignment, i.e. , n takes values ​​of 0, 1, ..., Nd-1; Length of overlapping area after alignment N is the length of a single frame signal. Within the overlapping region, calculate the height difference between corresponding points. Take the average of all differences as the vibration displacement of the current frame relative to the previous frame: .

7. The high-precision plate shape measurement and calibration method according to claim 1, characterized in that: The plate shape correction stage specifically includes: Let the horizontal plate shape measurement result of the current frame be... The vibration displacement of the current frame relative to the previous frame obtained from the vibration extraction stage is: The corrected true plate height The corrected actual plate height is used as the height coordinate of each point in the width direction of the steel plate. ; Coordinates along the direction of steel plate movement Speed ​​of steel plate movement With running time The product is determined as follows: ; Coordinates along the width of the steel plate Calculate using the following formula: ; in, Current image center pixel coordinates, These are the initial image pixel coordinates. The imaging magnification in the width direction. Adjust the scaling factor for the field of view.

8. A high-precision plate shape measurement and calibration system for implementing the method according to any one of claims 1-7, characterized in that, include: The laser projection module is used to simultaneously project mutually perpendicular horizontal and vertical laser lines onto the surface of a moving steel plate. The image acquisition module is used to continuously acquire images of the steel plate surface containing the transverse laser lines and the longitudinal laser lines; The calibration module is used to establish and store the calibration mapping relationship between sub-pixel values ​​and actual height for the horizontal and vertical laser lines using a standard lifting platform; The data processing module includes: The height calculation unit is used to extract the sub-pixel position of the laser line from the acquired image and call the calibration mapping relationship to convert it into the actual height to obtain the horizontal plate shape contour data and the vertical height data; The vibration extraction unit is used to perform phase correlation operation on the longitudinal height data sequence of adjacent frames, estimate the sub-pixel level longitudinal displacement, and calculate the average value of the height difference of the overlapping area as the vibration displacement of the steel plate after aligning the data based on the displacement. The plate shape correction unit is used to compensate the transverse plate shape profile data using the vibration displacement and output the true plate shape data after eliminating vibration interference.

9. A high-precision plate shape measurement and calibration system according to claim 8, characterized in that: The calibration module is specifically used for: For each fixed pixel position on each laser line, a shape-preserving piecewise cubic Hermite interpolation function is constructed with subpixel values ​​as independent variables and actual height as dependent variables. The cubic polynomial coefficients of each sub-interval are calculated. At the same time, a quadratic residual model with subpixel values ​​as independent variables is established. The PCHIP function and the residual model together form a comprehensive calibration mapping table. The height calculation unit first calculates the initial height using the PCHIP function by looking up the integrated calibration mapping table, then calculates the compensation value using the residual model, and finally adds the two to obtain the final height.

Images