A method for detecting an automobile seat skeleton stamping part based on machine vision positioning

Abstract

The present application relates to the technical field of automobile manufacturing, in particular to a kind of automobile seat framework stamping part detection method based on machine vision positioning. Its method includes: first, the surface gray image of stamping part is collected, the edge profile is extracted and matched with standard template to obtain the region to be measured;Then based on the displacement vector of edge profile, construct deformation vector field, based on local texture feature distribution, construct texture entropy field;Then according to the length of deformation vector field, the window size is adaptively adjusted and the microtexture direction is extracted;Finally, the crack significant index is constructed and the defect is positioned by integrating streamline vector, microtexture direction and texture entropy field. The present application improves the texture extraction accuracy by adaptive window adjustment guided by deformation vector field, and significantly improves the accuracy and robustness of crack detection by comprehensive evaluation of direction difference and texture quality.

Description

Technical Field

[0001] This invention relates to the field of automotive manufacturing technology. More specifically, this invention relates to a method for inspecting stamped automotive seat frame parts based on machine vision positioning. Background Technology

[0002] The car seat frame is the core load-bearing component of the car seat system, and its quality directly affects the safety and comfort of the occupants. Seat frames are typically formed using a stamping process. During stamping, factors such as material properties, mold wear, and fluctuations in process parameters can easily lead to surface defects such as cracks, wrinkles, and scratches. These defects not only affect the product's appearance but can also reduce structural strength, posing safety hazards. Therefore, high-precision and high-efficiency quality inspection of stamped parts is crucial.

[0003] Currently, quality inspection of stamped automotive seat frames mainly relies on manual visual inspection and simple machine vision methods. Manual inspection methods are affected by subjective factors such as the inspector's experience and fatigue, resulting in low efficiency and poor consistency, making it difficult to meet the demands of modern production. Traditional machine vision inspection methods often employ techniques such as fixed threshold segmentation and template matching. These methods are prone to false positives and false negatives when dealing with complex surface textures, lighting variations, and deformation interference.

[0004] While existing edge detection-based methods can identify obvious contour defects, they are insufficient for detecting subtle defects such as microcracks and shallow scratches. Texture analysis-based methods can capture surface texture features, but when deformation occurs in stamped parts, it is difficult to effectively distinguish between normal deformation textures and defect textures. Furthermore, stamped parts inevitably experience positional shifts and shape deformations during production, and traditional methods fail to fully consider the impact of deformation on texture distribution, leading to decreased detection accuracy.

[0005] Therefore, how to effectively integrate deformation information and texture features, and design an intelligent detection algorithm that can adaptively adjust detection parameters and accurately distinguish between deformed textures and defect textures, in order to solve the problem of insufficient detection accuracy and robustness in the existing technology, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to propose a machine vision-based method for detecting stamped automotive seat frames, in order to solve the problems of insufficient accuracy and robustness in defect detection in the prior art.

[0007] This invention provides a machine vision-based method for detecting stamped automotive seat frame parts, comprising: acquiring a grayscale image of the stamped automotive seat frame part surface; extracting the edge contour of the stamped part and registering it with a standard template to obtain an image of the area to be tested; constructing a deformation vector field reflecting the deformation trend of the stamped part based on the displacement vector of the edge contour relative to the standard template in the image of the area to be tested; constructing a texture entropy field reflecting the surface texture quality based on the local texture feature distribution of pixels in the image of the area to be tested; adjusting the window size of different areas in the image of the area to be tested according to the modulus of the deformation vector field, and extracting the micro-texture direction of each pixel based on the window size; extracting the direction of the deformation vector field as a streamline vector, and constructing a crack significance index by combining the streamline vector, the micro-texture direction, and the texture entropy field; wherein, the crack significance index constructs a positive correlation term based on the difference between the direction of the micro-texture direction and the streamline vector, and constructs a negative correlation term based on the texture entropy field; and locating stamping defects based on the crack significance index.

[0008] This invention solves the false detection problem caused by deformation interference in traditional methods by quantifying the deformation distribution characteristics of stamped parts through the construction of a deformation vector field. Surface texture quality is evaluated using a texture entropy field, effectively distinguishing between normal and abnormal textures. The window size is adaptively adjusted based on the deformation vector field, maintaining high resolution in areas of severe deformation and suppressing noise in flat areas. By combining the differences between streamline vectors and micro-texture directions and constructing a crack significance index using the texture entropy field, the orthogonality between crack and deformation directions is fully utilized, improving the accuracy and robustness of defect detection.

[0009] Optionally, the standard template includes a standard CAD template, and the registration method is rigid registration, affine registration, or non-rigid registration.

[0010] The registration strategy ensures precise alignment between the image under test and the standard template, providing a reliable benchmark for subsequent deformation analysis and significantly improving the accuracy of defect localization.

[0011] Optionally, the process of registering the edge contour of the stamped part with the standard template includes: extracting rigid feature points from the surface grayscale image, using a point set registration algorithm to perform a rigid transformation between the rigid feature points and the corresponding points of the standard template to complete coarse registration; based on the coarse registration, extracting the sub-pixel level edge contour of the skeleton, and aligning the image of the area to be tested using a contour matching algorithm based on distance transformation.

[0012] Optionally, the process of constructing the deformation vector field includes: calculating the displacement vector of each edge point on the registered edge contour relative to the corresponding edge point of the standard template, wherein the displacement vector includes direction and magnitude; and using a spatial interpolation algorithm to interpolate the direction and magnitude of the displacement vector to obtain the displacement vector of each pixel, thereby generating a dense displacement vector field, i.e., the deformation vector field.

[0013] Optionally, the process of calculating the modulus of the deformation vector field includes: calculating the Euclidean norm of the displacement vector for each pixel in the displacement vector field to obtain the deformation modulus of that point; statistically analyzing the deformation modulus of all pixels in the region to be measured to construct a modulus distribution map; and normalizing the modulus distribution map to generate a normalized deformation vector field; wherein, the larger the modulus value, the more severe the deformation of the region relative to the standard template.

[0014] Optionally, the process of constructing the texture entropy field includes: setting a local neighborhood with each pixel as the center in the image to be tested; calculating the statistical distribution characteristics of the gray-level gradient within the local neighborhood; using the Shannon entropy algorithm to calculate the disorder level value of the statistical distribution characteristics, and assigning the value to the center pixel, and traversing all pixels to form the texture entropy field.

[0015] The complexity and irregularity of surface texture were quantified using a texture entropy field. This field can effectively identify abnormal regions with chaotic textures while suppressing interference from normal texture regions, providing a reliable quality evaluation basis for the construction of the crack significance index.

[0016] Optionally, the adjustment rule for the window size includes: establishing a negative correlation mapping relationship between the window size and the magnitude of the deformation vector field.

[0017] The adaptive window adjustment strategy dynamically optimizes texture extraction parameters based on the degree of local deformation, maintaining high resolution in areas of severe deformation to capture subtle defects, and increasing the window size in flat areas to suppress noise, significantly improving the accuracy and stability of detection.

[0018] Optionally, the process of constructing the micro-texture direction includes: performing principal component analysis on the gray-level gradient of the pixel within a determined window size; extracting the feature direction with the largest gradient change, and defining the orthogonal direction of the feature direction as the micro-texture direction.

[0019] Optionally, the process of constructing the crack significance index includes: defining an orthogonal direction factor, which is positively correlated with the sine value of the angle between the microtexture direction and the streamline vector; defining a texture interference factor, which is positively correlated with the texture entropy field value; the crack significance index is directly proportional to the orthogonal direction factor and inversely proportional to the texture interference factor.

[0020] A crack saliency index is constructed by combining an orthogonal direction factor and a texture interference factor, fully utilizing the physical property that the crack and deformation directions are orthogonal and the regular texture of the crack region. This comprehensive evaluation mechanism enables the saliency index to accurately highlight the crack region, effectively suppress deformation texture and noise interference, and significantly improve the accuracy of defect identification.

[0021] Optionally, the process of locating the stamping defect includes: generating a mapping map of the crack significance index; processing the mapping map using a connected region hysteresis threshold segmentation algorithm: setting a high confidence threshold and a low confidence threshold, retaining regions higher than the high confidence threshold as seed points, and growing outward to connect neighborhoods higher than the low confidence threshold, and finally determining the connected region as the stamping defect.

[0022] The beneficial effects of this invention are as follows: The machine vision-based method for detecting stamped automotive seat frames provided by this invention constructs a complete intelligent defect detection system through innovative technologies such as deformation vector field construction, texture entropy field evaluation, adaptive window adjustment, and comprehensive evaluation of directional differences. This method fully integrates deformation information and texture features, achieving differentiated processing for regions with different degrees of deformation. The improved crack saliency index has clear physical meaning and adaptive evaluation capabilities, significantly improving the accuracy, efficiency, and robustness of stamping defect detection. Attached Figure Description

[0023] Figure 1 This is a flowchart of a machine vision-based method for detecting stamped automotive seat frames according to an embodiment of the present invention.

[0024] Figure 2 This is an original image of a stamped automotive seat frame according to an embodiment of the present invention.

[0025] Figure 3 This is a deformation vector field distribution diagram of a stamped automotive seat frame according to an embodiment of the present invention.

[0026] Figure 4 It is the texture entropy field of the stamped part of the automobile seat frame according to an embodiment of the present invention.

[0027] Figure 5 This is a crack significance index diagram of a stamped automotive seat frame according to an embodiment of the present invention.

[0028] Figure 6 This is a defect result diagram of a stamped automotive seat frame according to an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Figure 1 The diagram shown is a flowchart of a machine vision-based method for detecting stamped automotive seat frames according to an embodiment of the present invention.

[0030] S1: Acquire and register the surface grayscale image of the stamped part of the car seat frame.

[0031] First, a grayscale image of the stamped automotive seat frame part is acquired. The image acquisition system includes an industrial camera, an LED surface light source, and a mounting bracket. The acquired raw image is an 8-bit grayscale image with a grayscale value range of 0 to 255. After acquiring the grayscale image, image preprocessing is performed. First, median filtering is performed to remove noise, with the filter window size set to 3×3 pixels. Then, histogram equalization is performed to enhance image contrast and make the grayscale distribution more uniform. The preprocessed image is used as input for subsequent processing. Figure 2 The image shown is an original image of a stamped automotive seat frame according to an embodiment of the present invention.

[0032] Next, the edge contour of the stamped part is extracted. The Canny edge detection algorithm is used to extract edges, with the high threshold set to the 70th percentile of the grayscale value and the low threshold set to 50% of the high threshold. After edge detection, morphological closing operations are performed to connect broken edges, with a 3×3 square kernel as the structuring element. The largest connected component is extracted as the outer contour of the stamped part.

[0033] To achieve accurate registration, coarse registration is first performed. Rigid feature points, such as hole centers and corners, are extracted from the surface grayscale image. The Harris corner detection algorithm is used to extract feature points, with the corner response threshold set to 10% of the maximum image response value. The extracted feature points are matched with corresponding points in the standard CAD template. The RANSAC algorithm is used to estimate the rigid transformation matrix, including rotation and translation parameters. The transformation parameters are optimized using the least squares method to minimize the sum of squared Euclidean distances between the feature points and the template points. After coarse registration, the positional deviation between the stamped part image and the standard template is reduced to within 10 pixels.

[0034] Based on coarse registration, fine registration is performed. Subpixel-level edge contours of the skeleton are extracted, and a subpixel edge detection algorithm is used to achieve a positioning accuracy of 0.1 pixels. The image of the test region is aligned using a contour matching algorithm based on distance transform. The shortest distance from each point on the test contour to the standard template contour is calculated to construct a distance field. An iterative nearest-point algorithm is used to optimize the registration parameters to minimize the sum of the distance fields. The iteration terminates when the distance change between two consecutive iterations is less than 0.01 pixels or the number of iterations reaches 100. After fine registration is completed, the image of the test region that is precisely aligned with the standard template is obtained.

[0035] S2: Construct the deformation vector field and texture entropy field.

[0036] After registration, calculate the displacement vector of the edge contour relative to the standard template. Extract the nodes on the registered edge contour, with a node spacing of 5 pixels. For each node, calculate its corresponding point position in the standard template. The displacement vector is defined as the difference between the node's current position and its corresponding point position, including components in the x and y directions.

[0037] Spatial interpolation algorithms are used to diffuse the displacement vectors of edge nodes across the entire image area to be measured, generating a dense displacement vector field, i.e., a deformation vector field. The radial basis function (RBF) interpolation method is employed, with a Gaussian kernel function chosen as the basis function. Interpolation parameters are determined through cross-validation to ensure smooth interpolation results while preserving deformation details. Each pixel in the deformation vector field corresponds to a displacement vector; the magnitude of the vector reflects the degree of deformation at that point, and the direction of the vector reflects the trend of deformation.

[0038] Calculate the modulus distribution map of the deformation vector field. Modulus is defined as the square root of the sum of the squares of the x and y components of the displacement vector. Regions with larger modulus indicate severe deformation, while regions with smaller modulus indicate gentler deformation. Normalize the modulus distribution map, mapping the modulus values ​​to the range of 0 to 1. The specific formula is: ; ; in, and Represents the horizontal and vertical displacement components of a pixel, measured in pixels. E represents the original deformation modulus, which physically represents the physical offset distance of that point relative to the standard template. min and E max These represent the minimum and maximum values ​​of the modulus within the region to be measured, respectively. This is the normalized deformation vector field value, dimensionless, ranging from 0 to 1. A larger value indicates a more severe deformation of the region relative to the standard template, and it will be used as a weighting coefficient for subsequent adaptive window adjustments. Figure 3 The figure shown is a deformation vector field distribution diagram of a stamped part of an automobile seat frame according to an embodiment of the present invention.

[0039] Next, a texture entropy field is constructed. In the image to be tested, a local neighborhood is defined centered on each pixel, with a neighborhood size of 15×15 pixels. The statistical distribution characteristics of the gray-level gradient within the local neighborhood are calculated. First, the gray-level gradient magnitude and direction of each pixel within the neighborhood are calculated, and the Sobel operator is used to calculate the gradient. The gradient direction is quantized into 8 directional intervals, and the cumulative sum of the gradient magnitudes within each directional interval is calculated to form a gradient direction histogram.

[0040] The degree of disorder in the gradient direction histogram is calculated using a statistical dispersion evaluation function. Shannon entropy is used as the evaluation function, calculated as follows: the entropy value equals the sum of the products of the negative probability intervals and their logarithms. The probability is defined as the ratio of the cumulative sum of gradient magnitudes in that interval to the total sum of gradient magnitudes. A larger entropy value indicates a more disordered gradient direction distribution and a more irregular texture; a smaller entropy value indicates a more concentrated gradient direction distribution and a more regular texture.

[0041] The calculated entropy value is assigned to the center pixel, and this process is repeated for all pixels to form a texture entropy field. The texture entropy field reflects the spatial distribution of surface texture quality; regions with high entropy values ​​correspond to abnormal regions with disordered textures, while regions with low entropy values ​​correspond to normal regions with regular textures or potential cracked areas. The specific formula is: ; in, In the formula, K represents the number of intervals whose gradient direction is quantized, and n k P represents the number of pixels whose gradient direction falls within the k-th interval of the local neighborhood window, or the cumulative sum of gradient magnitudes, where N is the total number of pixels within the window. k This represents the probability of the k-th directional interval occurring, and its dimensionless value. This represents the texture entropy value at the center pixel, which is dimensionless. A larger value indicates a more chaotic distribution of local texture directions, such as noise or a rough surface; a smaller value indicates a higher consistency of texture directions, such as cracks or edge features. Figure 4 The figure shows the texture entropy field of a stamped automotive seat frame according to an embodiment of the present invention.

[0042] S3: Adaptively adjusts window size and extracts micro-texture orientation.

[0043] The window size in different regions of the image under test is adjusted based on the magnitude of the deformation vector field. A negative correlation mapping relationship is established between the window size and the magnitude of the deformation vector field. The specific mapping function is set as follows: the window size equals the maximum window size minus the product of the normalized magnitude and the window size variation range. The maximum window size is set to 21 pixels, the minimum window size to 7 pixels, and the window size variation range is 14 pixels. The specific formula is: ; in, Represents the side length of the window used to extract texture features at a given pixel, in pixels. "Round" is the rounding symbol. W max and W min These are the preset maximum and minimum window sizes, respectively. This represents the normalized deformation vector field value mentioned above. This formula establishes a negative correlation mapping relationship, meaning that a larger deformation leads to... When the value approaches 1, the window size converges to a minimum to preserve high-frequency details; conversely, when the deformation is small, it leads to... When the value approaches 0, the window size expands to its maximum value to utilize the large window to suppress noise in flat areas.

[0044] In regions with a large deformation vector field magnitude, the normalized magnitude value is close to 1, and the window size is close to the minimum value of 7 pixels to preserve high-frequency texture details and capture subtle defects. In regions with a small deformation vector field magnitude, the normalized magnitude value is close to 0, and the window size is close to the maximum value of 21 pixels to smooth low-frequency noise and improve detection stability.

[0045] The micro-texture orientation of each pixel is extracted based on a defined window size. Within the window of each pixel, principal component analysis is performed on the grayscale gradient. First, the grayscale gradient vectors of all pixels within the window are calculated to construct the gradient matrix. The covariance matrix of the gradient matrix is ​​then calculated, followed by eigenvalue decomposition to obtain the two principal component directions and their corresponding eigenvalues.

[0046] Extract the feature direction with the largest gradient change, which corresponds to the feature vector direction of the largest eigenvalue. This direction reflects the dominant direction of the texture within the window. The orthogonal direction of this feature direction is defined as the microtexture direction. The microtexture direction is expressed as an angle value, ranging from 0 to 180 degrees. For cracked regions, the microtexture direction is usually along the crack propagation direction; for normally deformed regions, the microtexture direction is usually along the deformation direction.

[0047] S4: Construct a crack significance index and locate stamping defects.

[0048] The direction of the deformation vector field is extracted as the streamline vector. The direction of the streamline vector is defined as the direction of the displacement vector, representing the direction of the deformation tendency. The streamline vector direction is normalized to an angle value from 0 to 180 degrees.

[0049] A crack saliency index is constructed by combining streamline vectors, microtexture directions, and texture entropy fields. First, an orthogonality factor is defined, which monotonically increases as the angle between the microtexture direction and the streamline vector approaches 90 degrees. The absolute value of the angle is calculated, ranging from 0 to 90 degrees. The orthogonality factor is defined as the ratio of the angle to 90 degrees; it reaches its maximum value of 1 when the angle is 90 degrees and is 0 when the angle is 0 degrees.

[0050] A texture interference factor is defined, which monotonically increases with the texture entropy field value. The texture interference factor is defined as the normalized value of the texture entropy plus a small constant (0.1 in this example) to avoid a denominator of zero. A larger texture entropy value results in a larger texture interference factor, indicating that the texture in that area is disordered, possibly due to noise or a complex background.

[0051] The crack saliency index is defined as the ratio of the orientation orthogonality factor to the texture interference factor. This index reaches its maximum in regions where the texture direction is perpendicular to the streamline vector and the texture entropy value is low. These regions correspond to typical crack characteristics: the crack direction is orthogonal to the deformation direction, and the texture in the crack region is relatively regular. The specific formula is: ; CSI is a dimensionless crack significance index; a higher value indicates a higher probability of crack formation. (Numerator) is the orthogonality factor, where For the direction and angle of the micro-texture, Let be the streamline vector direction angle, and the absolute value of the difference between the two represents the included angle. γ is the sensitivity coefficient used to enhance contrast. Denominator term The texture interference factor is determined by the normalized texture entropy value. This formula is formed by adding the minimum constant ε. It utilizes the physical property that cracks are usually perpendicular to the tensile deformation direction and have relatively regular textures. This allows the crack region to obtain a very large significance exponent due to the large numerator and small denominator, while noise regions are suppressed due to their high entropy value, and normal deformation textures are filtered out because of their consistent direction. For example... Figure 5 The figure shown is a crack significance index diagram of a stamped part of an automobile seat frame according to an embodiment of the present invention.

[0052] Calculate the crack significance index for all pixels in the image of the region to be tested, and generate a crack significance index mapping map. Regions with higher significance indices in the mapping map are potential defect regions.

[0053] A connected component lag thresholding algorithm is used to process the mapping graph. A high-confidence threshold is set to 70% of the maximum saliency index, and a low-confidence threshold is set to 40%. First, pixels with a saliency index higher than the high-confidence threshold are extracted as seed points. Then, the algorithm grows outward from the seed points, connecting neighboring pixels with a saliency index higher than the low-confidence threshold. Eight-neighbor connectivity is used for judgment, and the algorithm iteratively grows until no new pixels are added.

[0054] The finally identified connected regions are the stamping defects. Morphological processing is performed on each connected region to remove noise areas smaller than 50 pixels. The retained connected regions are marked as defects, and their location, area, aspect ratio, and other characteristic parameters are recorded. The detection results are overlaid on the original image and displayed to the operator through a human-computer interaction interface, generating a defect detection report. Figure 6 The figure shown is a defect result diagram of a stamped part of an automobile seat frame according to an embodiment of the present invention.

[0055] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be within the scope of protection of the present invention.

Claims

1. A method for inspecting stamped automotive seat frames based on machine vision positioning, characterized in that, The detection method includes: Acquire grayscale images of the stamped parts of the car seat frame, extract the edge contours of the stamped parts and register them with a standard template to obtain images of the area to be tested; Based on the displacement vector of the edge contour in the image of the area to be tested relative to the standard template, a deformation vector field reflecting the deformation trend of the stamped part is constructed; based on the local texture feature distribution of the pixels in the image of the area to be tested, a texture entropy field reflecting the surface texture quality is constructed. The window size of different regions in the image to be tested is adjusted according to the magnitude of the deformation vector field, and the micro-texture direction of each pixel is extracted based on the window size; The direction of the deformation vector field is extracted as the streamline vector, and a crack significance index is constructed by combining the streamline vector, the micro-texture direction, and the texture entropy field; wherein, the crack significance index constructs a positive correlation term based on the difference between the direction of the micro-texture direction and the streamline vector, and constructs a negative correlation term based on the texture entropy field; The stamping defects are located based on the crack significance index.

2. The method for inspecting stamped automotive seat frames based on machine vision positioning according to claim 1, characterized in that, The standard template includes: The standard CAD template specifies that the registration method is rigid registration, affine registration, or non-rigid registration.

3. The method for inspecting stamped automotive seat frames based on machine vision positioning according to claim 1, characterized in that, The process of registering the edge contour of the stamped part with the standard template includes: Extract rigid feature points from the surface grayscale image, and use a point set registration algorithm to perform rigid transformation between the rigid feature points and the corresponding points of the standard template to complete coarse registration; Based on coarse registration, subpixel-level edge contours of the skeleton are extracted, and the image of the region to be tested is aligned using a contour matching algorithm based on distance transform.

4. The method for inspecting stamped automotive seat frames based on machine vision positioning according to claim 1, characterized in that, The process of constructing the deformation vector field includes: Calculate the displacement vector of each edge point on the registered edge contour relative to the corresponding edge point of the standard template, wherein the displacement vector includes direction and magnitude; Using a spatial interpolation algorithm, the direction and magnitude of the displacement vector are interpolated to obtain the displacement vector of each pixel, generating a dense displacement vector field, i.e., the deformation vector field.

5. The method for inspecting stamped automotive seat frames based on machine vision positioning according to claim 1, characterized in that, The process of calculating the modulus of the deformation vector field includes: For each pixel in the displacement vector field, calculate the Euclidean norm of its displacement vector to obtain the deformation modulus of that point. Statistically analyze the deformation modulus of all pixels within the test area and construct a modulus distribution map; The modulus distribution map is normalized to generate a normalized deformation vector field; wherein, the larger the modulus value, the more severe the deformation of the region relative to the standard template.

6. The method for inspecting stamped automotive seat frames based on machine vision positioning according to claim 1, characterized in that, The process of constructing the texture entropy field includes: In the image of the region to be tested, a local neighborhood is set with each pixel as the center; Calculate the statistical distribution characteristics of the gray-level gradient within the local neighborhood; The degree of disorder of the statistical distribution feature is calculated using the Shannon entropy algorithm, and this value is assigned to the center pixel. The texture entropy field is formed by traversing all pixels.

7. The method for inspecting stamped automotive seat frames based on machine vision positioning according to claim 1, characterized in that, The rules for adjusting the window size include: Establish a negative correlation mapping relationship between window size and deformation vector field magnitude.

8. The method for inspecting stamped automotive seat frames based on machine vision positioning according to claim 1, characterized in that, The process of constructing the micro-texture direction includes: Within a defined window size, principal component analysis is performed on the grayscale gradient of each pixel. Extract the feature direction with the largest gradient change, and define the orthogonal direction of the feature direction as the microtexture direction.

9. The method for inspecting stamped automotive seat frames based on machine vision positioning according to claim 1, characterized in that, The process of constructing the crack significance index includes: Define an orthogonality factor, which is positively correlated with the sine value of the angle between the microtexture direction and the streamline vector; Define a texture interference factor, which is positively correlated with the texture entropy field value; The crack significance index is directly proportional to the orthogonality factor and inversely proportional to the texture interference factor.

10. The method for inspecting stamped automotive seat frames based on machine vision positioning according to claim 1, characterized in that, The process of locating stamping defects includes: Generate a mapping map of the crack significance index; The mapping graph is processed using a connected component lag threshold segmentation algorithm: a high confidence threshold and a low confidence threshold are set, regions above the high confidence threshold are retained as seed points, and neighboring regions above the low confidence threshold are grown outwards. The finally determined connected region is the stamping defect.

Images