Building steel stamping forming quality detection method based on binocular vision

Abstract

This invention relates to the field of steel plate defect detection technology, specifically to a method for inspecting the quality of stamped steel forming in construction based on binocular vision. The invention divides distorted regions from all segmented areas of the visual image based on the differences in grayscale values ​​of pixels in segmented regions; it then filters out mismatched points in the image to be analyzed based on the distribution differences between the image to be analyzed and the reference image; it determines the latest disparity value of the mismatched points based on their local disparity values; it updates the disparity image based on the latest disparity value of the mismatched points in the image to be analyzed, obtaining an updated disparity image; it constructs a three-dimensional model of the steel plate based on the updated disparity image; and it performs quality inspection on the steel plate based on the differences between the three-dimensional model and the standard steel plate model. This invention improves the rationality of steel plate quality inspection by reasonably updating the disparity values ​​and accurately constructing the three-dimensional model of the steel plate.

Description

Technical Field

[0001] This invention relates to the field of steel plate defect detection technology, specifically to a method for detecting the stamping quality of architectural steel based on binocular vision. Background Technology

[0002] Applying pressure to high-strength metal sheets using a stamping press to induce plastic deformation allows for the rapid fabrication of various shapes and sizes of architectural steel structure components, such as stamped steel scaffolding boards, decorative components, ceiling materials, or other building templates. Traditional inspection methods for most stamped parts rely on auxiliary techniques such as oiling stones, applying oil for illumination, and scribing and counting to assess quality. However, these methods are dependent on external conditions such as the inspection environment, equipment operating status, and judgment standards. Small, hidden, and difficult-to-detect defects are easily overlooked, leading to inaccurate results. Binocular vision technology can calculate the parallax between two visual images to obtain the three-dimensional shape information of an object. During the stamping process, the steel sheet deforms, changing its size and shape. Binocular vision technology monitors the size and shape of the steel sheet in real time to determine whether it meets design requirements.

[0003] However, when determining the parallax value, the existing technology does not fully consider the presence of distorted areas on the steel plate surface. These distorted areas will have inconsistent effects on the light reflection of the two cameras, resulting in unreasonable parallax values ​​for the distorted pixels. This makes it difficult to accurately construct a visual model of the steel structure, thereby affecting the accuracy of steel plate quality inspection. Summary of the Invention

[0004] To address the inaccuracy of existing technologies in steel plate quality inspection, the present invention aims to provide a binocular vision-based method for inspecting the stamping quality of architectural steel. The specific technical solution adopted is as follows:

[0005] A method for quality inspection of stamping forming of architectural steel based on binocular vision, the method comprising:

[0006] Using binocular vision, images of the steel plate from two different perspectives and a parallax image are obtained;

[0007] The viewpoint image is divided into blocks to obtain various segmented regions. Based on the differences in grayscale values ​​of pixels within these blocks, distortion regions are identified from all segmented regions of the viewpoint image. The areas of the distortion regions in the two viewpoint images are used to determine the image to be analyzed and a reference image. Based on the distribution differences between the image to be analyzed and the reference image, mismatched points in the image to be analyzed are identified. Based on the local disparity values ​​of the mismatched points in the image to be analyzed, the latest disparity value of the mismatched points is determined. Based on the latest disparity value of the mismatched points in the image to be analyzed, the disparity image is updated to obtain the updated disparity image.

[0008] A 3D model of the steel plate is constructed based on the updated parallax image; the steel plate is then subjected to quality inspection based on the differences between the 3D model and the standard steel plate model.

[0009] Furthermore, the method for obtaining the distorted region includes:

[0010] Based on the differences in grayscale values ​​of pixels in the segmented region, a distortion anomaly measure of the segmented region is obtained;

[0011] The distortion region is divided from all block regions of the viewpoint image based on the distortion anomaly metric.

[0012] Furthermore, the method for obtaining the distortion anomaly measure includes:

[0013] The mean of the absolute values ​​of the differences between the gray values ​​of every two pixels in the segmented region is calculated to obtain the first anomaly measure of the segmented region.

[0014] The ratio of the area of ​​the segmented region to the area of ​​the corresponding viewpoint image is calculated and negative correlation mapping is performed to obtain the second anomaly measure of the segmented region;

[0015] The first anomaly measure and the second anomaly measure are positively fused to obtain the distortion anomaly measure of the block region.

[0016] Furthermore, the method for defining the distorted region includes:

[0017] In the viewpoint image, each block region whose distortion anomaly measurement is greater than a preset distortion threshold is marked as a distortion region.

[0018] Furthermore, the method for obtaining the image to be analyzed and the reference image includes:

[0019] The total area of ​​all distorted regions in the viewpoint image is taken as the degree of distortion of the viewpoint image; the viewpoint image corresponding to the largest degree of distortion is taken as the image to be analyzed; and the viewpoint image corresponding to the smallest degree of distortion is taken as the reference image.

[0020] Furthermore, the method for obtaining the mismatch points includes:

[0021] Each distorted region in the image to be analyzed is taken as a region to be analyzed; each block region in the reference image is taken as a region to be matched.

[0022] Based on the differences in feature point distribution between the region to be analyzed and the region to be matched, a matching metric between the region to be analyzed and the region to be matched is obtained.

[0023] Based on the matching metrics between the region to be analyzed and all regions to be matched, determine the best matching region for the region to be analyzed;

[0024] Based on the disparity between the region to be analyzed and the best matching region, obtain the mismatched points in the region to be analyzed.

[0025] Furthermore, the method for obtaining the matching metric includes:

[0026] Calculate the absolute value of the difference between the number of feature points in the region to be analyzed and the region to be matched to obtain the first mismatch index between the region to be analyzed and the region to be matched.

[0027] Calculate the mean of the gradient values ​​of all feature points in the region to be analyzed, and use it as the feature gradient value of the region to be analyzed; calculate the mean of the gradient values ​​of all feature points in the region to be matched, and use it as the feature gradient value of the region to be matched; calculate the absolute value of the difference between the feature gradient values ​​of the region to be analyzed and the region to be matched, and obtain the second mismatch index between the region to be analyzed and the region to be matched.

[0028] Calculate the product of the first mismatch index and the second mismatch index and perform a negative correlation mapping to obtain the matching metric between the region to be analyzed and the region to be matched.

[0029] Furthermore, the method for obtaining the mismatch points includes:

[0030] Taking any pixel as the target pixel, calculate the mean of the absolute values ​​of the differences between the target pixel and the disparity values ​​of all its neighboring pixels to obtain the local change value of the target pixel; in the region to be analyzed, calculate the mean of the local change values ​​of all the pixels to obtain the overall change value; calculate the absolute value of the difference between the local change value of the target pixel and the overall change value and perform normalization processing to obtain the mismatch evaluation index of the target pixel; mark the pixels whose mismatch evaluation index is greater than a preset evaluation threshold as mismatch points.

[0031] Furthermore, the method for obtaining the latest disparity value includes:

[0032] Within the preset neighborhood range of the mismatched point, the mean disparity value of all neighboring data points is calculated and used as the latest disparity value of the mismatched point.

[0033] Furthermore, the method for quality inspection of the steel plate includes:

[0034] Map all data points in the 3D model of the steel plate to the standard steel plate model, and obtain the mapped data points corresponding to each data point;

[0035] The mean of the absolute values ​​of the differences between the curvature values ​​of all data points and their mapped data points is calculated and normalized to obtain the steel plate defect measure.

[0036] When the defect measurement of the steel plate is less than the preset defect threshold, the steel plate quality inspection level is determined to be qualified; when the defect measurement of the steel plate is not less than the preset defect threshold, the steel plate quality inspection level is determined to be unqualified.

[0037] The present invention has the following beneficial effects:

[0038] First, the viewpoint image is divided into several blocks to obtain individual regions. Considering that distortion defects can cause significant changes in light distribution, which are typically manifested as significant changes in grayscale values, distorted regions are identified from all blocks of the viewpoint image. After determining the image to be analyzed and the reference image, the distribution differences between the two images need to be compared to filter out mismatched points in the image to be analyzed. Mismatched points reflect pixels corresponding to inconsistent light reflection caused by distorted regions. To correct the disparity error caused by the influence of distorted regions, the latest disparity value needs to be determined for the selected mismatched points based on their local disparity values. Finally, the disparity image is updated based on the latest disparity values ​​of the mismatched points in the image to be analyzed, obtaining the updated disparity image. By reasonably updating the disparity image, the 3D model of the steel plate can be accurately constructed, improving the rationality of steel plate quality inspection. Attached Figure Description

[0039] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 A flowchart illustrating a method for quality inspection of stamped building steel based on binocular vision, provided in one embodiment of the present invention;

[0041] Figure 2 This is a flowchart illustrating a method for obtaining a distorted region according to an embodiment of the present invention.

[0042] Figure 3 This is a flowchart of a method for obtaining mismatched points according to an embodiment of the present invention. Detailed Implementation

[0043] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a binocular vision-based method for inspecting the quality of stamped steel forming in construction, as proposed by the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0044] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0045] The following description, in conjunction with the accompanying drawings, details a specific scheme for a binocular vision-based method for inspecting the quality of stamped steel building materials provided by this invention.

[0046] This invention provides a method for quality inspection of stamping and forming of architectural steel based on binocular vision. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of a binocular vision-based method for inspecting the quality of stamped steel forming in construction, according to an embodiment of the present invention. The method includes the following steps:

[0047] Step S1: Use binocular vision to obtain two viewpoint images and a parallax image of the steel plate.

[0048] The inspection system acquires two perspective images and a parallax image of the steel plate. It should be noted that using binocular vision to acquire these images is a well-known technique. Binocular vision, by simulating the stereoscopic vision principle of the human eye, can acquire images of the same object from different perspectives using left and right cameras, thereby calculating the object's three-dimensional information. A brief description of using binocular vision to acquire these images and parallax image of the stamped steel plate is provided below:

[0049] Two high-resolution industrial cameras were selected and fixed in appropriate relative positions to ensure that different perspectives of the steel plate could be captured. The two cameras were placed parallel to each other, ensuring a suitable baseline length between them to guarantee depth measurement accuracy. The industrial cameras were calibrated using a checkerboard calibration board or other calibration objects to obtain their intrinsic and extrinsic parameters. After calibration, the system could correctly correct image distortion and achieve accurate depth calculation. The steel plate was simultaneously captured from two different perspectives using binocular industrial cameras, resulting in images of the steel plate captured by the left and right cameras in the binocular vision system. Since the two acquired steel plate images contained noise, which affected image accuracy, denoising processing was performed to obtain denoised images. These denoised images were then converted to grayscale to obtain two perspective images of the steel plate. These two perspective images correspond to the images captured by the left and right cameras in the binocular vision system, and are referred to as the left view and right view, respectively. Stereo matching was performed on the two captured perspective images to obtain the parallax image of the steel plate. Each disparity value in a disparity image actually represents the difference in horizontal position between a pixel in the left view and its matching pixel in the right view. This difference is closely related to the depth information of objects in the scene and can be used for subsequent tasks such as depth estimation and 3D reconstruction.

[0050] In this embodiment of the invention, a bilateral filter is used for image noise reduction, and a weighted average method is used for grayscale conversion. The implementer can set these parameters according to the implementation scenario. It should be noted that, for ease of calculation, all index data involved in the calculation in this embodiment of the invention have undergone data preprocessing to eliminate the influence of dimensions. The specific methods for removing the influence of dimensions are well-known to those skilled in the art and are not limited here.

[0051] During the cold stamping process of building steel components, insufficient plastic deformation can occur, leading to distorted areas such as scratches, dents, or protrusions on the steel plate surface. These distorted areas cause significant changes in the angle, intensity, and distribution of light reflection, especially when the steel plate surface is uneven. When determining a parallax image using images from two viewpoints, the presence of these distorted areas can cause inconsistent light reflection from the two cameras, resulting in unreasonable parallax values ​​for some pixels within the distorted areas. Therefore, it is necessary to update the parallax image with these unreasonable parallax values.

[0052] Step S2: Divide the viewpoint image to obtain each block region of the viewpoint image; based on the difference in grayscale values ​​of pixels in the block regions, divide the distortion region from all the block regions of the viewpoint image; determine the image to be analyzed and the reference image based on the area of ​​the distortion region in the two viewpoint images; based on the distribution difference between the image to be analyzed and the reference image, filter out the mismatched points in the image to be analyzed; based on the local disparity value of the mismatched points in the image to be analyzed, determine the latest disparity value of the mismatched points; based on the latest disparity value of the mismatched points in the image to be analyzed, update the disparity image to obtain the updated disparity image.

[0053] The viewpoint image is divided into several regions. Considering that distortion defects can significantly alter light distribution, typically manifesting as significant changes in grayscale values, distorted regions are identified from all regions of the viewpoint image. After determining the image to be analyzed and the reference image, the distribution differences between the two images are compared to identify mismatched points in the image to be analyzed. Mismatched points reflect pixels corresponding to inconsistent light reflection caused by distorted regions. To correct disparity errors caused by distorted regions, the latest disparity value is determined for each identified mismatched point based on its local disparity value. Finally, the disparity image is updated based on the latest disparity values ​​of the mismatched points in the image to be analyzed, resulting in an updated disparity image for subsequent accurate 3D model construction.

[0054] To perform region-based analysis, the viewpoint image is first divided into segments to obtain various block regions. Preferably, in one embodiment of the present invention, the method for obtaining the block regions includes:

[0055] Based on the region growing algorithm, image segmentation is performed on the viewpoint image to obtain various block regions of the viewpoint image. It should be noted that the region growing algorithm is a well-known existing technology in the art and will not be described in detail here.

[0056] For the steps described above, the region growing algorithm is a segmentation method based on pixel similarity. It starts with multiple seed points and gradually adds neighboring pixels similar to the seed points into the region until a stopping condition is met. In steel plate quality inspection, surface distortions such as scratches, dents, and protrusions cause significant changes in grayscale values. The region growing algorithm can effectively capture these changes and separate defective areas from normal areas.

[0057] To identify the distorted regions from all partitioned regions, please refer to [link / reference]. Figure 2 The diagram illustrates a flowchart of a method for obtaining a distorted region according to an embodiment of the present invention. Preferably, in one embodiment of the present invention, the method for obtaining the distorted region includes:

[0058] Step S201: Based on the difference in grayscale values ​​of pixels in the segmented region, obtain the distortion anomaly measurement of the segmented region.

[0059] The likelihood of a block region being a distortion defect region is quantified by constructing a distortion anomaly metric.

[0060] Preferably, in one embodiment of the present invention, the method for obtaining the distortion anomaly measure includes:

[0061] The first anomaly measure of the block region is obtained by calculating the mean of the absolute values ​​of the differences between the grayscale values ​​of every two pixels in the segmented region. The second anomaly measure is obtained by calculating the ratio of the area of ​​the segmented region to the area of ​​the corresponding viewpoint image and performing a negative correlation mapping. The first and second anomaly measures are then positively fused to obtain the distortion anomaly measure of the segmented region. The negative correlation mapping in this invention can be inversely proportional or a negative exponential form, and is not limited here. It should be noted that positive fusion is a prior art well known to those skilled in the art. Positive fusion can employ simple multiplication, arithmetic mean, or other suitable fusion methods. In one embodiment of this invention, the product of the first and second anomaly measures is calculated to obtain the distortion anomaly measure of the segmented region.

[0062] Regarding the above steps, considering that distortion defects on the surface of stamped steel plates, such as scratches, dents, and protrusions, can cause significant changes in grayscale values, the greater the variation in grayscale values ​​within a segmented area, the more likely the segmented area is to be a distortion defect area. At the same time, considering that distortion areas are usually caused by local damage or stress release on the steel plate surface, they often manifest as small-area grayscale value changes. Based on the region growing algorithm, the smaller the area of ​​the segmented area, the more likely the segmented area is to be a distortion defect area.

[0063] Within a segmented region, all pixels are traversed, and the absolute value of the difference between the grayscale values ​​of every two pixels is calculated. The mean of these absolute differences is then taken. This mean is defined as the first anomaly measure of the segmented region, reflecting the magnitude of grayscale value variation within the region. The larger the value of the first anomaly measure, the more drastic the grayscale value variation within the segmented region, and the greater the likelihood of distortion defects.

[0064] Secondly, the ratio of the area of ​​the segmented region to the area of ​​the entire viewpoint image is calculated, and a negative correlation mapping is performed to obtain the second anomaly metric of the segmented region. The negative correlation mapping means that the smaller the segmented region, the larger its second anomaly metric value. This is because distorted regions typically exhibit small-area grayscale value changes, so smaller segmented regions are more likely to contain distortion defects.

[0065] Finally, by combining the first and second anomaly measures, a distortion anomaly measure for the segmented region is obtained. This distortion anomaly measure can more comprehensively reflect the possibility that the segmented region is a distortion defect region.

[0066] Step S202: Based on the distortion anomaly metric, divide the distortion region from all block regions of the viewpoint image.

[0067] In order to identify the distorted regions from the segmented regions, considering that the larger the distortion anomaly measure, the greater the probability that the segmented region is a distortion defect region, the distortion anomaly measure is used to identify the distorted regions from all segmented regions of the viewpoint image.

[0068] Preferably, in one embodiment of the present invention, the method for dividing the distorted region includes:

[0069] In the viewpoint image, each block region whose distortion anomaly measure is greater than a preset distortion threshold is marked as a distortion region. In one embodiment of the present invention, the preset distortion threshold is 0.56. The preset distortion threshold is set according to the actual application scenario and image quality requirements, and is used to distinguish distortion regions.

[0070] Preferably, in one embodiment of the present invention, the method for obtaining the image to be analyzed and the reference image includes:

[0071] The total area of ​​all distorted regions in the viewpoint image is taken as the degree of distortion of the viewpoint image; the viewpoint image with the largest degree of distortion is taken as the image to be analyzed; and the viewpoint image with the smallest degree of distortion is taken as the reference image.

[0072] For the above steps, firstly, the total area of ​​all distorted regions in each viewpoint image is calculated; this area represents the degree of distortion of that viewpoint image. Then, the degree of distortion of the two viewpoint images is compared, and the viewpoint image with the greatest distortion is identified. Since this image contains the most distorted regions, it is used as the image to be analyzed. Similarly, the viewpoint image with the least distortion is identified. This image is relatively normal and is therefore used as the reference image.

[0073] Please see Figure 3 The diagram illustrates a flowchart of a method for obtaining mismatched points according to an embodiment of the present invention. Preferably, in one embodiment of the present invention, the method for obtaining mismatched points includes:

[0074] Step S211: Take each distorted region in the image to be analyzed as a region to be analyzed; take each block region in the reference image as a region to be matched.

[0075] Following the steps described above, distorted regions in the image to be analyzed are extracted and used as the objects for subsequent matching analysis. Since distorted regions may contain defects or anomalies on the steel plate surface, using these regions as the areas to be analyzed helps to more accurately identify mismatch points. Each block region in the reference image can potentially become an object to be matched with the area to be analyzed; therefore, each block region in the reference image is used as a different area to be matched.

[0076] Step S212: Based on the differences in feature point distribution between the region to be analyzed and the region to be matched, obtain the matching metric between the region to be analyzed and the region to be matched.

[0077] A matching metric is constructed to evaluate the degree of matching between the region to be analyzed and the region to be matched.

[0078] Preferably, in one embodiment of the present invention, the method for obtaining the matching metric includes:

[0079] The absolute value of the difference in the number of feature points between the region to be analyzed and the region to be matched is calculated to obtain the first mismatch index. The mean of the gradient values ​​of all feature points in the region to be analyzed is calculated as the feature gradient value of the region to be analyzed. The mean of the gradient values ​​of all feature points in the region to be matched is calculated as the feature gradient value of the region to be matched. The absolute value of the difference in the corresponding feature gradient values ​​between the region to be analyzed and the region to be matched is calculated to obtain the second mismatch index. The product of the first and second mismatch indices is calculated and negative correlation mapping is performed to obtain the matching metric between the region to be analyzed and the region to be matched. It should be noted that the method for obtaining feature points is a well-known existing technique, which can be achieved using SIFT (Scale-Invariant Feature Transform) detection. The specific acquisition process is not detailed here. It should also be noted that negative correlation mapping is a well-known technique, and negative correlation mapping can take the form of inverse proportion or negative exponential power, without limitation.

[0080] Regarding the steps described above, feature points in the distorted region are less affected by changes in lighting and maintain relative stability across images from different viewpoints. This unique characteristic makes feature points play a crucial role in image matching. First, the absolute value of the difference in the number of feature points between the region to be analyzed and the region to be matched is calculated to obtain the first mismatch index. The first mismatch index reflects the difference in the number of feature points between the two regions. A larger first mismatch index indicates a potentially significant difference in structure or content between the two regions. The greater the mismatch, the less likely they are to represent the same part of the steel plate. Next, the mean of the gradient values ​​of all feature points in the region to be analyzed is calculated as the feature gradient value of the region to be analyzed. Similarly, the mean of the gradient values ​​of all feature points in the region to be matched is calculated as the feature gradient value of the region to be matched. The feature gradient value reflects the overall texture or structural features of the region. Finally, the absolute value of the difference in the feature gradient values ​​between the two regions is calculated to obtain the second mismatch index. The second mismatch index measures the difference in texture or structural features between the two regions to be analyzed. The larger the second mismatch index, the more significant the differences in texture or structure between the region to be analyzed and the region to be matched. The greater the mismatch, the less likely they are to represent the same part of the steel plate. A negative correlation mapping is applied to the product of the first and second mismatch indices to obtain a matching metric. The matching metric more comprehensively assesses the degree of matching between the region to be analyzed and the region to be matched; a larger matching metric indicates a better match between the two regions.

[0081] Step S213: Determine the best matching region for the region to be analyzed based on the matching metrics of the region to be analyzed and all regions to be matched.

[0082] To determine the best matching region for the region to be analyzed, the optimal matching region is determined based on the matching metrics between the region to be analyzed and all regions to be matched.

[0083] Preferably, in one embodiment of the present invention, the method for obtaining the optimal matching region includes:

[0084] Among the matching metrics of the region to be analyzed and all regions to be matched, the region with the largest matching metric is taken as the best matching region for the region to be analyzed.

[0085] By identifying the maximum matching metric through the above steps, the corresponding region to be matched can be determined as the optimal matching region for the region to be analyzed. This optimal matching region is visually, structurally, or in terms of features most similar to the region to be analyzed, and therefore is most likely the corresponding steel plate region of the region to be analyzed from another perspective or under other conditions.

[0086] Step S214: Based on the disparity between the region to be analyzed and the best matching region, obtain the mismatch points in the region to be analyzed.

[0087] By analyzing the parallax between the region to be analyzed and the optimal matching region, mismatched points in the region to be analyzed are identified. These mismatched points reflect differences in light reflection caused by distortion defects on the steel plate surface, resulting in inaccurate matching at certain points.

[0088] Preferably, in one embodiment of the present invention, the method for obtaining mismatch points includes:

[0089] Based on the region to be analyzed and the best matching region, obtain the disparity value corresponding to each pixel in the region to be analyzed;

[0090] Taking any pixel as the target pixel, the mean of the absolute values ​​of the differences between the target pixel and the disparity values ​​of all its neighboring pixels is calculated to obtain the local change value of the target pixel. In the region to be analyzed, the mean of the local change values ​​of all pixels is calculated to obtain the overall change value. The absolute value of the difference between the local change value and the overall change value of the target pixel is calculated and normalized to obtain the mismatch evaluation index of the target pixel. Pixels with mismatch evaluation indices greater than a preset evaluation threshold are marked as mismatch points. It should be noted that the specific method for obtaining the disparity value is a well-known prior art and can be obtained through the SAD binocular vision algorithm. In one embodiment of this invention, pixels within the eight-neighborhood of the target pixel are considered as neighboring pixels of the target pixel. The eight-neighborhood is a well-known prior art and will not be elaborated here. Normalization processing in this invention can employ linear normalization, etc., and is not limited here. The preset evaluation threshold is set to 0.63 in this invention, and the implementer can set it according to the implementation scenario.

[0091] Following the steps described above, the SAD binocular vision algorithm is used to process the region to be analyzed and the best-matching region to obtain a disparity map between them. This disparity map contains the disparity value corresponding to each pixel in the region to be analyzed, i.e., the displacement of the pixel from the region to be analyzed to the best-matching region. For each pixel in the region to be analyzed, the mean of the absolute values ​​of the differences between its disparity values ​​and those of all its neighboring pixels is calculated. This mean is called the local change value of the target pixel, reflecting the degree of difference between the target pixel and its surrounding pixels in the depth direction. In the region to be analyzed, the local change values ​​of all pixels are statistically analyzed, and their mean is calculated to obtain the overall change value. The overall change value reflects the overall trend of change in the depth direction of the region to be analyzed. For each target pixel, the absolute value of the difference between its local change value and the overall change value is calculated and normalized. Normalization ensures that the value of the mismatch evaluation index is within a reasonable range, facilitating subsequent comparison and judgment. The value obtained after normalizing this absolute value of the difference is called the mismatch evaluation index of the target pixel. A preset evaluation threshold is set to determine whether a pixel is a mismatch point. If the mismatch evaluation index of a target pixel exceeds this preset evaluation threshold, it is marked as a mismatch point. These mismatch points are likely due to differences in light reflection caused by distortion defects on the steel plate surface, resulting in inaccurate matching at these points.

[0092] The above method can effectively identify mismatch points caused by uneven depth on the steel plate surface. These mismatch points may negatively impact subsequent image processing and matching accuracy. Therefore, in subsequent processing steps, these mismatch points can be corrected to improve the overall processing accuracy and effect.

[0093] To correct the disparity values ​​corresponding to mismatched points, preferably, in one embodiment of the present invention, the method for obtaining the latest disparity value includes:

[0094] Within a preset neighborhood of the mismatched point, the mean disparity value of all neighboring data points is calculated and used as the latest disparity value of the mismatched point. In one embodiment of the present invention, a system of pixels with a size of 5 is constructed with the mismatched point as the center pixel. A preset neighborhood range of 5 is defined, with the center pixel being the center pixel. In this invention, all pixels within the preset neighborhood range except the center pixel are considered as neighborhood data points. It should be noted that the disparity value of the neighborhood data points is obtained by comparing the region to be analyzed with the optimal matching region, thus determining the disparity value corresponding to each pixel in the region to be analyzed.

[0095] To correct the disparity value of a mismatched point by utilizing the effective disparity values ​​around it, the mean of the disparity values ​​of all neighboring data points within a preset neighborhood of the mismatched point is calculated and used as the latest disparity value for that point. This value is then used to update the disparity value of the mismatched point in the disparity image, thereby improving the accuracy and reliability of the disparity image.

[0096] The disparity image is updated based on the latest disparity values ​​calculated for the mismatched points. Specifically, the disparity values ​​for the corresponding mismatched points in the disparity image are replaced with the latest calculated disparity values. This results in an updated disparity image where the disparity values ​​at the mismatched points are more accurate and reliable.

[0097] Step S3: Construct a 3D model of the steel plate based on the updated parallax image; perform quality inspection on the steel plate based on the differences between the 3D model and the standard steel plate model.

[0098] The updated parallax image more accurately reflects the depth information of the steel plate surface. By updating the parallax image appropriately, a three-dimensional model of the steel plate can be accurately constructed, thereby improving the rationality of steel plate quality inspection.

[0099] To construct an accurate three-dimensional model of the steel plate, preferably, in one embodiment of the present invention, the method for obtaining the three-dimensional model of the steel plate includes:

[0100] A three-dimensional model of the steel plate is constructed using binocular vision based on an updated disparity image. This model includes the steel plate's geometry and surface features, serving as the basis for subsequent quality inspection. It should be noted that constructing a three-dimensional model of the steel plate using binocular vision based on an updated disparity image is a well-known technique in the art and will not be elaborated upon here.

[0101] To accurately inspect the quality of steel plates, preferably, in one embodiment of the present invention, the method for inspecting the quality of steel plates includes:

[0102] All data points in the 3D model of the steel plate are mapped to a standard steel plate model to obtain the mapped data point for each data point. The mean of the absolute values ​​of the differences between the curvature values ​​of all data points and their mapped data points is calculated and normalized to obtain the steel plate defect metric. When the steel plate defect metric is less than a preset defect threshold, the steel plate quality inspection level is determined to be qualified; when the steel plate defect metric is not less than the preset defect threshold, the steel plate quality inspection level is determined to be unqualified. It should be noted that the method for obtaining the curvature value is a prior art well known to those skilled in the art, and can be obtained through the least squares method, which will not be elaborated here. In one embodiment of the present invention, the preset defect threshold is 0.58, and the implementer can set it according to the implementation scenario. The normalization process in the present invention can adopt linear normalization, etc., and is not limited here.

[0103] Following the steps outlined above, all data points in the 3D model of the steel plate are mapped one by one to the standard steel plate model. This mapping process ensures that each data point can find a corresponding mapping point in the standard model. This step is fundamental for subsequent difference calculations. For each data point and its mapped data point, their curvature values ​​are calculated. Curvature is an important indicator describing the degree of surface curvature. The absolute value of the difference between the curvature values ​​of each data point and its mapped point is calculated, and the mean of all absolute differences is taken. This mean reflects the overall difference in curvature between the 3D model of the steel plate and the standard model. This mean is then normalized to obtain the steel plate defect metric. Normalization eliminates the influence of different sizes or shapes of steel plates on the defect metric results, making the results more comparable. A preset defect threshold is set to determine the quality inspection level of the steel plate. When the steel plate defect metric is less than the preset defect threshold, the steel plate quality inspection level is determined to be acceptable; when the steel plate defect metric is not less than the preset defect threshold, the steel plate quality inspection level is determined to be unacceptable.

[0104] In summary, this invention provides a method for quality inspection of stamped steel forming in construction based on binocular vision. This invention divides distorted regions from all segmented regions of the visual image based on the differences in grayscale values ​​of pixels in segmented regions; it filters out mismatched points in the image to be analyzed based on the distribution differences between the image to be analyzed and the reference image; it determines the latest disparity value of the mismatched points based on their local disparity values; it updates the disparity image based on the latest disparity value of the mismatched points in the image to be analyzed, obtaining the updated disparity image; it constructs a three-dimensional model of the steel plate based on the updated disparity image; and it performs quality inspection on the steel plate based on the differences between the three-dimensional steel plate model and the standard steel plate model. This invention improves the rationality of steel plate quality inspection by reasonably updating the disparity values ​​and accurately constructing the three-dimensional model of the steel plate.

[0105] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0106] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A method for detecting the quality of building steel stamping forming based on binocular vision, characterized in that, The method includes: Using binocular vision, images of the steel plate from two different perspectives and a parallax image are obtained; The viewpoint image is divided into blocks to obtain various segmented regions. Based on the differences in grayscale values ​​of pixels within these blocks, distortion regions are identified from all segmented regions of the viewpoint image. The areas of the distortion regions in the two viewpoint images are used to determine the image to be analyzed and a reference image. Based on the distribution differences between the image to be analyzed and the reference image, mismatched points in the image to be analyzed are identified. Based on the local disparity values ​​of the mismatched points in the image to be analyzed, the latest disparity value of the mismatched points is determined. Based on the latest disparity value of the mismatched points in the image to be analyzed, the disparity image is updated to obtain the updated disparity image. Based on the updated parallax image, a 3D model of the steel plate is constructed; based on the differences between the 3D model of the steel plate and the standard steel plate model, the quality of the steel plate is inspected. The methods for obtaining the image to be analyzed and the reference image include: The total area of ​​all distorted regions in the viewpoint image is taken as the degree of distortion of the viewpoint image; the viewpoint image corresponding to the largest degree of distortion is taken as the image to be analyzed; and the viewpoint image corresponding to the smallest degree of distortion is taken as the reference image. The method for obtaining the mismatch points includes: Each distorted region in the image to be analyzed is taken as a region to be analyzed; each block region in the reference image is taken as a region to be matched. Based on the differences in feature point distribution between the region to be analyzed and the region to be matched, a matching metric between the region to be analyzed and the region to be matched is obtained. Based on the matching metrics between the region to be analyzed and all regions to be matched, determine the best matching region for the region to be analyzed; Based on the disparity between the region to be analyzed and the best matching region, obtain the mismatch points in the region to be analyzed; The method for obtaining the mismatch points includes: Taking any pixel as the target pixel, calculate the mean of the absolute values ​​of the differences between the target pixel and the disparity values ​​of all its neighboring pixels to obtain the local change value of the target pixel; in the region to be analyzed, calculate the mean of the local change values ​​of all the pixels to obtain the overall change value; calculate the absolute value of the difference between the local change value of the target pixel and the overall change value and perform normalization processing to obtain the mismatch evaluation index of the target pixel; mark the pixels whose mismatch evaluation index is greater than a preset evaluation threshold as mismatch points.

2. The method for detecting the quality of the building steel stamping forming based on binocular vision according to claim 1, characterized in that, The method for obtaining the distorted region includes: Based on the differences in grayscale values ​​of pixels in the segmented region, a distortion anomaly measure of the segmented region is obtained; The distortion region is divided from all block regions of the viewpoint image based on the distortion anomaly metric.

3. The method for detecting the stamping forming quality of building steel based on binocular vision according to claim 2, characterized in that, The method for obtaining the distortion anomaly measurement includes: The mean of the absolute values ​​of the differences between the gray values ​​of every two pixels in the segmented region is calculated to obtain the first anomaly measure of the segmented region. The ratio of the area of ​​the segmented region to the area of ​​the corresponding viewpoint image is calculated and negative correlation mapping is performed to obtain the second anomaly measure of the segmented region; The first anomaly measure and the second anomaly measure are positively fused to obtain the distortion anomaly measure of the block region.

4. The method for quality inspection of stamping and forming of architectural steel based on binocular vision according to claim 2, characterized in that, The method for dividing the distorted region includes: In the viewpoint image, each block region whose distortion anomaly measurement is greater than a preset distortion threshold is marked as a distortion region.

5. The method for quality inspection of stamping and forming of architectural steel based on binocular vision according to claim 1, characterized in that, The method for obtaining the matching metric includes: Calculate the absolute value of the difference between the number of feature points in the region to be analyzed and the region to be matched to obtain the first mismatch index between the region to be analyzed and the region to be matched. Calculate the mean of the gradient values ​​of all feature points in the region to be analyzed, and use it as the feature gradient value of the region to be analyzed; calculate the mean of the gradient values ​​of all feature points in the region to be matched, and use it as the feature gradient value of the region to be matched; calculate the absolute value of the difference between the feature gradient values ​​of the region to be analyzed and the region to be matched, and obtain the second mismatch index between the region to be analyzed and the region to be matched. Calculate the product of the first mismatch index and the second mismatch index and perform a negative correlation mapping to obtain the matching metric between the region to be analyzed and the region to be matched.

6. The method for quality inspection of stamping and forming of architectural steel based on binocular vision according to claim 1, characterized in that, The method for obtaining the latest disparity value includes: Within the preset neighborhood range of the mismatched point, the mean disparity value of all neighboring data points is calculated and used as the latest disparity value of the mismatched point.

7. The method for detecting the quality of the building steel stamping forming based on binocular vision according to claim 1, characterized in that, The method for quality inspection of steel plates includes: Map all data points in the 3D model of the steel plate to the standard steel plate model, and obtain the mapped data points corresponding to each data point; The mean of the absolute values ​​of the differences between the curvature values ​​of all data points and their mapped data points is calculated and normalized to obtain the steel plate defect measure. When the defect measurement of the steel plate is less than the preset defect threshold, the steel plate quality inspection level is determined to be qualified; when the defect measurement of the steel plate is not less than the preset defect threshold, the steel plate quality inspection level is determined to be unqualified.

Images