An image processing method and a measurement method for large-size flexible parts.

Abstract

This invention discloses an image processing method and a measurement method for a large-size flexible component. The image processing method corrects a first image of the flexible component acquired by a camera. The flexible component is tensioned at both ends along a first direction by rollers, exhibiting a state of being tight in the middle and loose at the ends, causing the edges of the flexible component to periodically undulate up and down. The method includes the following steps: processing the region in the first image corresponding to the middle position of the flexible component using a quadratic polynomial; processing the region in the first image corresponding to the two ends of the flexible component using a cubic spline curve; and obtaining a corrected second image. This results in a corrected image that more closely approximates the actual size, facilitating subsequent inspection.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology, specifically to an image processing method and a measurement method for large-size flexible parts. Background Technology

[0002] During the production of automotive airbags, it is necessary to measure the dimensions of the black yarn and airbag pattern on the fabric rolls online. Due to the flexible nature of the airbag fabric rolls, rigid calibration plates cannot be used to calibrate and correct distortion in the camera. Therefore, a high-precision flexible calibration plate is required. This plate is attached to the fabric roll and then scanned and photographed by a camera. However, due to the limited space of the production line layout, i.e., adding a testing station to the existing production line, the space is compact. Furthermore, during the transport of the airbag, the two ends of the fabric roll to be tested are tensioned by rollers. Due to the flexible nature of the fabric, the two parallel rollers that tension the fabric roll cannot fully tighten the airbag fabric, mainly exhibiting a state of tightness in the middle and looseness at both ends. This causes the fabric roll to exhibit periodic up-and-down undulations at both ends during actual operation. When calculating the feature point coordinates of the calibration plate, this results in varying degrees of undulation at both ends of the flexible calibration plate, leading to a deviation between the corrected image and the actual image. Summary of the Invention

[0003] The purpose of this invention is to overcome the aforementioned defects or problems in the prior art and to provide an image processing method and a measurement method for large-size flexible parts.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] Option 1: An image processing method for correcting a first image of a flexible component acquired by a camera. The flexible component is tensioned at both ends along a first direction by two parallel rollers, exhibiting a state of being tight in the middle and loose at the edges, causing the flexible component to periodically undulate up and down along its two ends perpendicular to the first direction. The method includes the following steps:

[0006] The region in the first image corresponding to the middle position of the flexible component is processed using a quadratic polynomial;

[0007] The regions on the first image corresponding to the positions of the two ends of the flexible component are processed using cubic spline curves;

[0008] To obtain the corrected second image.

[0009] Option 2, based on Option 1, calibrates the second image. In the horizontal direction, Zhang Zhengyou's calibration method is used to calibrate the camera's horizontal distortion and pixel scale scaling relationship. In the vertical direction, a linear calibration method is used to unify the pixel scale scaling relationship to be consistent with the horizontal direction.

[0010] Option 3: A method for measuring a large-size flexible component, used to measure the size of the object to be measured, wherein two parallel rollers are provided at both ends of the large-size flexible component to drive the large-size flexible component along a first direction; the large-size flexible component includes a plurality of objects to be measured arranged equidistantly along the first direction, with two objects to be measured separated by marking lines, and the marking lines being equally spaced, including the following steps:

[0011] To obtain the theoretical dimensional information required for the object under test;

[0012] The line scan camera captures images of the large flexible component moving along a first direction, and segments the image at two adjacent marker lines to obtain the original image of the object under test between the two marker lines.

[0013] The noise in the original image is eliminated using frequency domain processing to obtain the first image;

[0014] The image to be tested is obtained by processing the first image using an image processing method as described in Scheme 2.

[0015] Extract the contour data of the object to be tested from the image to be tested; measure the actual size information of the object to be tested based on the contour data;

[0016] The measurement result is obtained by comparing the actual size information of the object under test with the theoretical size information.

[0017] Option 4, based on Option 3, involves obtaining the theoretical dimensional information required for the object under test, including the following steps:

[0018] The CAD drawing of the object to be measured is used as the matching template for the measurement profile, and theoretical dimension information is set according to the CAD drawing. The theoretical dimension information includes dimension items, measurement type, measurement tolerance and corresponding geometric elements.

[0019] Option 5, based on Option 3, uses frequency domain processing to eliminate noise in the original image to obtain the first image, including the following steps:

[0020] After performing a Fourier transform on the original image, the grayscale features are converted to frequency domain features, and the frequencies corresponding to the noise features are eliminated in the image spectrum. Finally, an inverse Fourier transform is performed on the noise-eliminated image to obtain the first image.

[0021] Option 6, based on Option 3, involves enhancing the texture of the image under test before extracting the contour data of the object under test. The texture enhancement of the image under test includes the following steps: designing second-order differential filters in four directions to enhance the contour texture, with the second-order differential filters in each direction evenly distributed at 90-degree intervals.

[0022] Scheme 7, based on Scheme 6, defines the filter design formula as follows: In an M*N image... Above, using an m*n filter, the following calculations are performed:

[0023] ,in, , , These are filter coefficients. It is an image value.

[0024] Option 8, based on Option 3, uses a linear Gaussian filter and the gradient consistency principle to extract the contour data of the object under test.

[0025] Scheme Nine, based on Scheme Eight, utilizes the following extraction principle of the Gaussian algorithm: The partial derivatives of a Gaussian smoothing filter are used to determine the parameters of a quadratic polynomial for the x and y coordinates of each pixel in the image. These parameters are used to calculate the line direction of each pixel. Pixels with local maxima in the second partial derivative perpendicular to this line direction are marked as points on the line. These points are then connected to form a contour. First, the gradient of each pixel and the difference in gradient magnitude and direction between adjacent pixels are calculated. The first derivative of the Gaussian function is used as the suboptimal detection operator for straight-line edges. The two-dimensional Gaussian function is shown in the following equation: In one direction superior The first directional derivative is In the formula , , It is a direction vector. It is the gradient vector. With G n Perform convolution, and simultaneously change direction, At the maximum value It refers to the direction orthogonal to the detection edge;

[0026] Then, based on the principle of consistency between the gradient direction and magnitude of the line, pixels are tracked and marked. This mainly involves two steps: First, the image is scanned line by line. When the gradient value of a pixel is lower than a certain threshold, it is marked as 0, indicating that the pixel is in a flat region. When the gradient value of a pixel is greater than a certain threshold, it is marked with numbers in turn, indicating that the pixel is in a non-flat region. Second, the gradient direction difference and magnitude difference of the pixels marked as non-zero are calculated. When the gradient direction difference and magnitude difference are within a certain threshold range, it is determined that the two pixels are on the same line support set, and the two points are marked with the same number. Finally, the corresponding line is fitted from the line support set.

[0027] Option 10, based on Option 3, involves two line scanning cameras arranged side-by-side along a direction perpendicular to the first direction and stitched together to form the image to be tested.

[0028] As can be seen from the above description of the present invention, compared with the prior art, the present invention has the following beneficial effects:

[0029] 1. Option 1: Due to the flexible nature of the airbag fabric roll and the limitations of the production line layout, a rigid calibration plate cannot be used to calibrate the camera. Therefore, a high-precision flexible calibration plate is used, which is attached to the fabric roll for scanning and imaging. Because the rollers tightening the fabric roll cannot fully stretch the airbag fabric, the edges of the roll will periodically undulate during actual operation (due to uneven force on the fabric roll during operation or centrifugal force from the rollers being biased towards the ends). When calculating the feature point coordinates of the calibration plate, this results in varying degrees of undulation at both ends of the flexible calibration plate, causing the corrected image to deviate from the true image. To address this, a second corrected image is obtained by processing the region corresponding to the middle position of the flexible component in the first image using a quadratic polynomial; and by processing the regions corresponding to the two ends of the flexible component in the first image using a cubic spline curve. This results in an image that is closer to the true size after correction, facilitating subsequent inspection.

[0030] 2. Option 2: The second image is calibrated. In the horizontal direction, Zhang Zhengyou calibration method is used to calibrate the relationship between the camera's horizontal distortion and pixel scale scaling. In the vertical direction, a linear calibration method is used to perform distortion correction and obtain an image closer to the real size through pixel scale scaling.

[0031] 3. Option 3: A measurement method for large-size flexible components, comprising the following steps:

[0032] To obtain the theoretical dimensional information required for the object under test;

[0033] A line scan camera captures images of the large flexible component moving along a first direction, and divides the image at two adjacent marker lines to obtain the original image of the object between the two marker lines. Using a line scan camera makes it easy to acquire images of objects with a large length.

[0034] Frequency domain processing is used to eliminate noise in the original image to obtain the first image. Because the fabric roll exhibits periodic up-and-down undulations at both edges during actual operation, the edge portion of the original image acquired by the line scan camera will show periodic brightness changes, which is detrimental to extracting the line contours of the airbag pattern to be measured. Conventional image processing methods based on grayscale values ​​cannot obtain stable and effective results. Considering its periodic characteristics, frequency domain processing is used to eliminate noise in the original image.

[0035] The image to be tested is obtained by processing the first image using the above image processing method and then stitching the images together.

[0036] Extract the contour data of the object to be tested from the image to be tested; measure the actual size information of the object to be tested based on the contour data;

[0037] The measurement result is obtained by comparing the actual size information of the object to be measured with the theoretical size information;

[0038] This measurement method can detect all quantities, improving product quality, and can measure online in real time, resulting in high detection efficiency.

[0039] 4. Option 4: Use the CAD drawing of the object to be measured as the matching template for the measurement profile, and set the theoretical dimension information according to the CAD drawing. The theoretical dimension information includes dimension items, measurement type, measurement tolerance and corresponding geometric elements. The setting method is convenient and facilitates quick model changeover on the production line. It is suitable for the measurement and testing requirements of different product models.

[0040] 5. Option 5: After performing a Fourier transform on the original image, the grayscale features are converted to frequency domain features, and then a filter is designed to filter and eliminate noise; finally, an inverse Fourier transform is performed on the noise-eliminated image to obtain the first image, thereby achieving noise elimination.

[0041] 6. Option Six: Due to the limited resolution of the line scan camera, the image appears blurry with indistinct grayscale contrast. That is, the outline of the geometric element to be measured is not clearly visible in the background image, making it difficult to extract stably. Texture enhancement of the image to be measured can facilitate extraction.

[0042] 7. Scheme 7: The second-order differential filter can enhance the texture of the image under test by using this design formula.

[0043] 8. Option 8 uses a linear Gaussian filter and the gradient consistency principle to extract the contour data of the object under test, which can ensure the consistency between the extracted contour data and the actual geometric elements, so as to obtain the elements of the measured contour set.

[0044] 9. Option 10: Two line scan cameras are used. The two line scan cameras are set up side by side and stitched together to form the original image, in order to solve the problem of the excessive width of the object to be measured. Attached Figure Description

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments are briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 A step-by-step diagram illustrating a measurement method for a large-sized flexible component;

[0047] Figure 2 This refers to the calibration plate style used in this scheme;

[0048] Figure 3 This is a schematic diagram of the airbag pattern of a type of airbag;

[0049] Figure 4 for Figure 2 CAD drawing of the airbag pattern.

[0050] Figure 5 This is a flowchart illustrating the steps of a processing method.

[0051] Figure 6 This is a diagram showing the transmission between the roller and the fabric. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are preferred embodiments of the present invention and should not be considered as excluding other embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0053] Unless otherwise expressly defined, the use of terms such as "first," "second," or "third" in the claims, description, and accompanying drawings of this invention is for distinguishing different objects and not for describing a specific order.

[0054] Unless otherwise expressly defined, in the claims, description, and accompanying drawings of this invention, the use of directional terms such as "center," "lateral," "longitudinal," "horizontal," "vertical," "top," "bottom," "inner," "outer," "upper," "lower," "front," "rear," "left," "right," "clockwise," and "counterclockwise" to indicate orientation or positional relationships is based on the orientation and positional relationships shown in the accompanying drawings and is only for the convenience of describing the invention and simplifying the description, and is not intended to indicate or imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the specific scope of protection of this invention.

[0055] Unless otherwise expressly defined, the terms "fixed connection" or "fixed connection" used in the claims, description and drawings of this invention should be interpreted broadly to refer to any connection in which there is no displacement or relative rotation relationship between the two parties, including non-removable fixed connection, detachable fixed connection, integral connection and fixed connection by other means or components.

[0056] In the claims, description and accompanying drawings of this invention, the terms "comprising," "having," and variations thereof are used to mean "including but not limited to."

[0057] refer to Figure 1 A method for measuring large-sized flexible components, used to measure the dimensions of the large-sized flexible components, the large-sized flexible components comprising several test objects equidistantly arranged along a first direction, two test objects separated by marking lines, the marking lines being equally spaced, and rollers provided at both ends of the large-sized flexible components. Figure 6 The two rollers are parallel and tensioned by the rollers to make the large flexible component move along the first direction;

[0058] In this embodiment, the large-size flexible component is an automotive airbag, the object to be tested is the airbag pattern, and the marking line is black yarn. During the production of automotive airbags, the dimensions of the black yarn and the airbag pattern on the fabric roll need to be measured online. Due to the flexible nature of the airbag fabric roll, the rollers tightening the roll cannot fully stretch the airbag fabric, resulting in a state where the middle is tight and the sides are loose. This causes the fabric roll to exhibit a periodic up-and-down undulation phenomenon along the two ends perpendicular to the first direction during actual operation, such as... Figure 6 As shown.

[0059] Before measurement, the theoretical dimensional information of the object to be measured must be obtained. This involves using the CAD drawing of the object as a matching template for the measurement profile and setting the theoretical dimensional information based on the CAD drawing. The theoretical dimensional information includes dimensional items, measurement type, measurement tolerance, and corresponding geometric elements. Since the dimensional measurement of airbag patterns needs to be carried out during the continuous operation of the production line, there are many dimensional items that need to be measured, and different models of airbag products will have certain differences in the dimensional items that need to be measured. Since CAD drawings are used for product design, the theoretical dimensional information can be set directly by importing the CAD drawings, providing an extremely convenient way for product changeovers and personnel operations on the production line.

[0060] For details, please refer to Figure 3 , Figure 4 The CAD drawing contains all the dimension annotations, measurement types (points, lines, arcs, etc.), and the start and end positions of the measurement. By developing functional software, the CAD drawing file is searched and parsed to read the drawing and annotation information, forming a measurement selection list. Each row in the list corresponds to a measurement dimension, including the definition of the measurement item and the start and end positions of the measurement, and then the corresponding selection process is performed.

[0061] A line scan camera photographs the fabric roll moving along a first direction and segments it at two adjacent marked lines to obtain the original image of the object under test between the two marked lines. In the actual measurement process, the line scan camera scans and photographs the entire roll of fabric, which continuously contains multiple airbag pattern areas. Based on the black yarn on the fabric, the detection area image of each individual airbag pattern is segmented to obtain the measurement and detection image of each airbag, i.e., the original image.

[0062] Because the width of the entire fabric roll is large, a single camera cannot achieve the required measurement accuracy. Therefore, two line scan cameras are used. The two line scan cameras are arranged side by side above the airbag production line in a direction perpendicular to the first direction, and the scanning direction of the line scan cameras is parallel to the running direction of the fabric roll.

[0063] To obtain the first image, a frequency domain processing method is used to eliminate noise in the original image. Because the fabric roll exhibits periodic up-and-down fluctuations at both edges during actual operation, the edge portions of the original image captured by the line scan camera show periodic brightness variations, which is detrimental to extracting the outline of the airbag pattern to be measured. Conventional image processing methods based on grayscale values ​​cannot obtain stable and effective results. Considering its periodic characteristics, a frequency domain processing method is adopted. After performing a Fourier transform on the original image, the grayscale features are converted to frequency domain features. The frequencies corresponding to the noise characteristics are eliminated in the image spectrum. Finally, an inverse Fourier transform is performed on the noise-reduced image to obtain the first image.

[0064] The discrete Fourier transform F(u,v) of a function f(x,y) with image size M*N.

[0065]

[0066] u,v=0,1,2…N-1

[0067] Inverse transform of F(u,v):

[0068]

[0069] x,y=0,1,2…N-1

[0070] The calibration is performed based on the projection matrix obtained from the standardization. Specifically, before measurement, the two line scan cameras for splicing detection can be pre-calibrated. Due to the flexible nature of the airbag fabric roll and the limitations of the production line layout, a rigid calibration plate cannot be used to calibrate the camera. Therefore, a high-precision flexible calibration plate is used, which is attached to the fabric roll and scanned by the camera. Because the rollers that tighten the fabric roll cannot fully tighten the airbag fabric, the edges of the fabric roll will periodically undulate during actual operation. When calculating the feature point coordinates of the calibration plate, the two ends of the flexible calibration plate will have different degrees of undulation, and the corrected image will deviate from the real image. Specifically, the following method is used for processing: the area corresponding to the middle position of the flexible part in the first image is processed using a quadratic polynomial; the area corresponding to the two ends of the flexible part in the first image is processed using a cubic spline curve to obtain the corrected second image, that is, an image that is closer to the real size after correction.

[0071] The cubic spline curve is processed as follows:

[0072] The general equation of the cubic spline curve is:

[0073] in, These are the characteristic points of the control curve;

[0074] The basis functions in the equation of a cubic spline curve are:

[0075] in Indicates social class, that is

[0076]

[0077]

[0078]

[0079]

[0080] The equation of the cubic spline curve is:

[0081] .

[0082] Then, Zhang Zhengyou's calibration method was used to calibrate the lateral distortion and pixel scale relationship of the camera. Longitudinal calibration is related to trigger sampling and the uniformity of fabric movement, and mainly uses a linear calibration method. Finally, the two line scan cameras are stitched together to form the image under test.

[0083] After enhancing the texture of the image to be tested, the contour data of the object to be tested is extracted. Since the contours of the geometric elements to be measured are not obvious in the background image, through analysis and testing, a second-order differential filter in four directions is designed to enhance the contour texture. The second-order differential filters in each direction are evenly distributed at 90-degree intervals. The design formula of the second-order differential filter can be defined as follows: In an M*N image... Above, using an m*n filter, the following calculations are performed:

[0084] ,in, , , These are filter coefficients. These are image values ​​that enhance the texture of the geometric elements to be measured, making them easier to extract; parameter a represents the size of the image noise in the X direction, and parameter b represents the size of the image noise in the Y direction.

[0085] Then, the contour data of the object under test is extracted. The contour data of the object under test is mainly extracted using a linear Gaussian filter and the gradient consistency principle. This ensures that the extracted contour data is consistent with the actual geometric elements, so as to obtain the measurement contour set elements.

[0086] The specific extraction principle is as follows: The extraction principle of the Gaussian algorithm is as follows: The partial derivatives of a Gaussian smoothing filter are used to determine the parameters of a quadratic polynomial for the x and y coordinates of each pixel in the image. The parameters of the quadratic polynomial are used to calculate the line direction of each pixel. Pixels with local maxima in the second partial derivative perpendicular to this line direction are marked as points on the line. The points on the line found in this way are connected to form a contour. First, the gradient of the pixel and the gradient magnitude and direction difference between adjacent pixels are calculated. The first derivative of the Gaussian function is used as the suboptimal detection operator for straight-line edges. Let the two-dimensional Gaussian function be as shown in the equation: In one direction superior The first directional derivative is In the formula , , It is a direction vector. It is the gradient vector. With G n Perform convolution, and simultaneously change direction, At the maximum value It refers to the direction orthogonal to the detection edge; θ represents a certain direction. The direction and angle of differentiation;

[0087] Then, based on the principle of consistency between the gradient direction and magnitude of the line, pixels are tracked and labeled. This mainly involves two steps: First, the image is scanned line by line. When the gradient value of a pixel is below a certain threshold, it is marked as 0, indicating that the pixel is in a flat region. When the gradient value of a pixel is above a certain threshold, it is labeled with numbers, indicating that the pixel is in a non-flat region, i.e., the pixel may be an edge point. Second, the gradient direction difference and magnitude difference of the pixels marked as non-zero are calculated. When the gradient direction difference and magnitude difference are within a certain threshold range, the two pixels are determined to be on the same line support set, and these two points are labeled with the same number. That is, adjacent pixels in the image whose gradient direction difference and magnitude difference are within a certain threshold range are labeled with the same number, while pixels in different sets are labeled differently. Finally, the corresponding line is fitted from the line support set.

[0088] Then, based on the contour data, the actual dimensions of the object to be measured are measured. The actual dimensions are compared with the theoretical dimensions to obtain the measurement result. If the result meets the standard, it is an OK product; otherwise, it is an NG product. In the actual measurement process, the CAD drawing is matched with the image to be measured. An affine transformation is used to establish a mapping relationship between the drawing coordinate system and the image coordinate system. Then, based on the positional distribution of the measurement items, the specified geometric elements are searched in the image coordinate system for measurement.

[0089] This measurement method can perform full inspection of airbag patterns, improving product quality. It can also perform real-time online measurement with high inspection efficiency. The setup method is convenient, facilitating quick model changeover on the production line, and is suitable for the measurement and inspection requirements of different product models.

[0090] The foregoing description of the specifications and embodiments is intended to explain the scope of protection of this invention, but does not constitute a limitation on the scope of protection of this invention. Modifications, equivalent substitutions, or other improvements to the embodiments of this invention or a portion thereof that can be obtained by those skilled in the art through logical analysis, reasoning, or limited experimentation, based on the teachings of this invention or the foregoing embodiments, in conjunction with common knowledge, general technical knowledge, and / or existing technology, should all be included within the scope of protection of this invention.

Claims

1. A method for measuring a large-size flexible component, used to measure the size of the object to be measured in the large-size flexible component, wherein two parallel rollers are provided at both ends of the large-size flexible component to drive the large-size flexible component along a first direction; the large-size flexible component includes a plurality of objects to be measured arranged at equal intervals along the first direction, two objects to be measured are separated by marking lines, and the marking lines are arranged at equal intervals, characterized in that: Includes the following steps: To obtain the theoretical dimensional information required for the object under test; The line scan camera scans and captures images of the large flexible component that is moving along the first direction, and divides the image at two adjacent marker lines to obtain the original image of the object under test between the two marker lines. The noise in the original image is eliminated using frequency domain processing to obtain the first image; An image processing method is used to process a first image to obtain a test image; an image processing method is used to correct a first image of a flexible component acquired by a camera, wherein the flexible component is tensioned by two parallel rollers at both ends along a first direction and is in a state of being tight in the middle and loose at both ends, so that the flexible component undulates periodically up and down along the two ends perpendicular to the first direction. Includes the following steps: The region in the first image corresponding to the middle position of the flexible component is processed using a quadratic polynomial; The regions on the first image corresponding to the positions of the two ends of the flexible component are processed using cubic spline curves; To obtain the corrected second image; to calibrate the second image, the Zhang Zhengyou calibration method is used to calibrate the horizontal distortion and pixel scale scaling relationship of the camera in the horizontal direction, and the linear calibration method is used in the vertical direction to unify the pixel scale scaling relationship with the horizontal direction. Extract the contour data of the object to be tested from the image to be tested; The actual dimensions of the object under test are measured based on the contour data. The measurement result is obtained by comparing the actual size information of the object under test with the theoretical size information.

2. The measurement method for a large-size flexible component as described in claim 1, characterized in that: Obtaining the theoretical dimensional information required for the object under test includes the following steps: The CAD drawing of the object to be measured is used as the matching template for the measurement profile, and theoretical dimension information is set according to the CAD drawing. The theoretical dimension information includes dimension items, measurement type, measurement tolerance and corresponding geometric elements.

3. The measurement method for a large-size flexible component as described in claim 1, characterized in that: The process of eliminating noise from the original image using frequency domain processing to obtain the first image includes the following steps: After performing a Fourier transform on the original image, the grayscale features are converted to frequency domain features, and the frequencies corresponding to the noise features are eliminated in the image spectrum. Finally, an inverse Fourier transform is performed on the noise-eliminated image to obtain the first image.

4. The measurement method for a large-size flexible component as described in claim 1, characterized in that: After enhancing the geometric texture features of the image to be tested, the contour data of the object to be tested is extracted. The enhancement of the image to be tested includes the following steps: designing second-order differential filters in four directions to enhance the contour texture, with the second-order differential filters in each direction evenly distributed at 90-degree intervals.

5. The measurement method for a large-size flexible component as described in claim 4, characterized in that: The design formula for a second-order differential filter is defined as: in an M*N image... Above, using an m*n filter, the following calculations are performed: ,in, , , These are filter coefficients. These are image values; parameter a represents the size of the image noise in the X direction, and parameter b represents the size of the image noise in the Y direction.

6. The measurement method for a large-size flexible component as described in claim 1, characterized in that... The contour data of the object under test is extracted using a linear Gaussian filter and the gradient consistency principle.

7. The measurement method for a large-size flexible component as described in claim 6, characterized in that... The principle of linear feature extraction using the Gaussian algorithm is as follows: The partial derivatives of a Gaussian smoothing filter are used to determine the parameters of a quadratic polynomial for the x and y coordinates of each pixel in the image. These parameters are used to calculate the line direction for each pixel. Pixels with local maxima in the second partial derivative perpendicular to this line direction are marked as points on the line. These points are then connected to form a contour. First, the gradient of the pixel and the difference in gradient magnitude and direction between adjacent pixels are calculated. The first derivative of the Gaussian function is used as a suboptimal operator for detecting straight-line edges. The two-dimensional Gaussian function is shown in the following equation: In one direction superior The first directional derivative is In the formula , , It is a direction vector. It is the gradient vector. With G n Perform convolution, and simultaneously change direction, At the maximum value It refers to the direction orthogonal to the detection edge; θ represents a certain direction. The direction and angle of differentiation; Then, based on the principle of consistency between the gradient direction and magnitude of the line, pixels are tracked and labeled. This mainly involves two steps: First, the image is scanned line by line. When the gradient value of a pixel is lower than a certain threshold, it is labeled as 0, indicating that the pixel is in a flat region. When the gradient value of a pixel is greater than a certain threshold, it is labeled with numbers in turn, indicating that the pixel is in a non-flat region. Second, the gradient direction difference and magnitude difference of the pixels labeled as non-zero are calculated. When the gradient direction difference and magnitude difference are within a certain threshold range, it is determined that the two pixels are on the same line support set, and the two points are labeled with the same number. Finally, the corresponding line is fitted from the line support set.

8. The measurement method for a large-size flexible component as described in claim 1, characterized in that... The line scan camera consists of two cameras, which are arranged side by side along a direction perpendicular to the first direction and stitched together to form the image to be tested.

Images