An AI vision-based high-end fashion stitching trajectory high-speed quality inspection system

Abstract

This invention discloses a high-speed quality inspection system for high-end fashion stitching trajectories based on AI vision, relating to the field of visual recognition. It includes: a data acquisition module; a region identification module; a combination determination module, which identifies at least one needle hole position in the area to be sewn and uses adjacent needle hole positions as a needle hole combination; a coefficient calculation module, which uniformly takes at least one sampling point in the area to be sewn and identifies the average curvature coefficient of the sampling point; a stitch distance judgment module, which, based on the average curvature coefficient, identifies the distance between the needle hole positions in the needle hole combination as the actual stitch distance of the needle hole combination; and a flatness judgment module. By setting up the region identification module, coefficient calculation module, stitch distance judgment module, and flatness judgment module, problems with needle hole arrangement and stitch flatness can be accurately identified.

Description

Technical Field

[0001] This invention relates to the field of visual recognition, specifically to a high-speed quality inspection system for high-end fashion sewing thread trajectories based on AI vision. Background Technology

[0002] Seam trajectory detection is a classic and crucial topic in industrial automation and quality inspection. It involves technologies such as computer vision, image processing, and machine learning. High-end fashion demands high precision in seam trajectory detection, requiring consistent spacing between adjacent needle holes and smooth, close-fitting seams to the fabric surface, primarily for aesthetics and stitch quality. However, since fabric can be flat or curved in different locations, surface distances cannot be simply calculated through point-to-point straight-line measurements. Furthermore, image recognition distances are two-dimensional, differing from actual spatial distances. This further increases errors in recognizing needle hole spacing and seam smoothness, making accurate identification difficult. Summary of the Invention

[0003] To address the aforementioned technical issues, a high-speed quality inspection system for high-end fashion seam trajectories based on AI vision is provided. This technical solution resolves the problems mentioned in the background section.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A high-speed quality inspection system for high-end fashion seam trajectories based on AI vision, comprising: The data acquisition module analyzes and obtains the sample interval of the stitch length of high-end fashion seams, and analyzes and obtains the allowable error of the seams. The region identification module acquires at least one fabric that constitutes high-end fashion, takes two fabrics that have an intersection as a fabric combination, and forms a region to be sewn at the intersection of the two fabrics in the fabric combination. If the stitch line goes beyond the region to be sewn, there is a problem with the sewing; otherwise, no processing is done. The combination determination module identifies at least one needle hole location in the area to be sutured, and combines adjacent needle hole locations as needle hole combinations. A coefficient calculation module, wherein the coefficient calculation module uniformly takes at least one sampling point in the area to be stitched and identifies the average curvature coefficient of the sampling point; The stitch distance judgment module identifies the distance between the needle positions in the needle eye combination based on the average curvature coefficient, which is taken as the actual stitch distance of the needle eye combination. If the difference between the actual stitch distance and the sample interval exceeds the allowable error of the suture, then there is a problem with the distance between the needle positions in the needle eye combination; otherwise, no processing is performed. The flatness judgment module takes the suture portion between the needle holes in the needle hole combination as the feature suture of the needle hole combination. It takes at least one feature point evenly on the feature suture and analyzes the actual distance between adjacent feature points. It accumulates the actual distances between adjacent feature points in the feature suture to obtain the length of the feature suture. If the difference between the length of the feature suture and the sample interval exceeds the allowable error of the suture, then there is a problem with the flatness of the suture at the needle hole position in the needle hole combination. Otherwise, no processing is performed.

[0005] Preferably, the analysis to obtain the sample interval of the stitch length of high-end fashion seams includes the following steps: Obtain at least one sample fashion item, and use the flat portion of the sample fashion item as the sample plane; The needle holes of the sutures in the sample plane are identified to obtain the sample needle holes. The distance between adjacent sample needle holes is measured to obtain the sample needle distance. The average of the needle distances for at least one sample is taken to obtain the sample interval.

[0006] Preferably, the analysis to obtain the permissible error of the suture includes the following steps: The suture allowable error is obtained by subtracting the maximum and minimum values ​​of the stitch distance of at least one sample.

[0007] Preferably, forming the area to be sewn at the junction of two fabrics in the fabric assembly includes the following steps: In the sample fashion, the overlapping part of two fabrics when they are sewn together is taken as the sample strip, the width of the sample strip is taken as the sample width, and the two edges of the sample strip that are parallel to the seam trajectory inside the sample strip are taken as the sample edges. The distance from the seam trace within the sample fabric strip to the sampling edge is used as the first sample distance and the second sample distance, respectively. The exposed edge at the overlapping position of two fabrics in the fabric combination is taken as the feature edge, and the fabric where the feature edge is located is taken as the feature fabric. A feature strip is generated on the feature fabric, satisfying that one edge of the feature strip is a feature edge, the feature strip is obtained by translating the feature edge, and the width of the feature strip is equal to the sample width. Generate a feature trajectory in the feature cloth, where the distance from the feature trajectory to the feature edge is equal to the distance of the first sample or the distance of the second sample; Two feature boundaries are formed on both sides of the feature trajectory. The feature boundaries are obtained by translating the feature trajectory. The distance from the feature boundary to the feature trajectory is equal to the allowable error of the suture. The area between the two feature boundaries is taken as the area to be sutured.

[0008] Preferably, identifying at least one needle prick location in the area to be sutured includes the following steps: From a top-down perspective, at a preset shooting distance, an image of the area to be stitched is acquired. The pixels in the image of the area to be stitched are classified to obtain at least one set of pixels, satisfying that the pixel values ​​of the pixels in the set of pixels are consistent. Use the set of pixels with the smallest number of elements as the target set of pixels; Generate at least one initial set, which contains one of the pixels in the target pixel set; The initial set is expanded by adding adjacent pixels from the target pixel set to the initial set until no more elements are added to the initial set. At least one initial set is deduplicated to obtain a target initial set, and the area covered by the pixels in the target initial set is taken as the pinhole position.

[0009] Preferably, the identification of the average curvature coefficient of the sampling points includes the following steps: The line connecting adjacent sampling points is taken as a sampling line segment, and one of the sampling line segments generated by the sampling point is taken as the reference line segment of the sampling point. The three-view image of the baseline segment is obtained by performing three-view image acquisition, resulting in the top view image, left view image, and front view image of the baseline segment; In the top-view image, the difference in longitudinal distance between the two ends of the baseline segment is identified and used as the top-view length; In the left view image, the difference in lateral distance between the two ends of the baseline segment is identified and used as the left view length; In the front view image, the difference in longitudinal distance between the two ends of the baseline segment is identified and used as the front view length; The first distance is calculated using the first length formula, and the second distance is calculated using the second length formula. Divide the second distance by the first distance to obtain the average curvature coefficient of the sampling points corresponding to the baseline segment; The formula for the first length is as follows: , Where L is the first distance, a is the top-view length, and b is the left-view length; The formula for the second length is as follows: , Where S is the second distance and c is the frontal length.

[0010] Preferably, the acquisition of the three-view image of the baseline segment includes the following steps: A top-down view of the baseline segment is obtained by shooting at a preset distance. From a left-hand perspective, at a preset shooting distance, obtain a left-hand view image of the baseline segment; Using a frontal view and a preset shooting distance, obtain a frontal image of the baseline line segment.

[0011] Preferably, identifying the distance to the needle eye position in the needle eye combination based on the average curvature coefficient includes the following steps: The ratio of the actual size to the corresponding image size in the top view image is obtained in advance and used as a scaling factor; Using a top-down perspective and a preset shooting distance, a top-down image of the needle hole combination is obtained. The lines connecting the needle hole positions in the needle hole combination are obtained as feature lines. Sampling points whose distance from the feature lines is less than the allowable error of the suture are used as target sampling points. In the top view image at the pinhole position, draw a horizontal line through the target sampling point. The horizontal line is parallel to the horizontal direction and divides the feature line into at least one local line segment. Match the target sampling points on the two horizontal lines that form the local line segment to the local line segment. In the top view image at the needle hole position, obtain the image size of the local line segment, and take the average value of the average curvature coefficient of the target sampling points corresponding to the local line segment to obtain the reference coefficient of the local line segment. The image size of a local line segment is multiplied by its reference coefficient to obtain a preliminary size. The preliminary size of the local line segment is then multiplied by a scaling factor to obtain its actual size. The actual sizes of at least one local line segment are then superimposed to obtain the distance between the needle eye positions in the needle eye combination.

[0012] Preferably, the analysis to obtain the actual distance between adjacent feature points includes the following steps: Using a top-down perspective and a preset shooting distance, a top-down image of the feature point is obtained, and the sampling point closest to the feature point is paired with the feature point; The line connecting adjacent feature points is taken as the target line. The average curvature coefficient of the sampling points corresponding to the feature points at both ends of the target line is averaged to obtain the target coefficient of the target line. The actual distance between adjacent feature points is obtained by multiplying the image size of the target line in the top view image of the feature point, the target coefficient of the target line, and the scale coefficient.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: By setting up a region recognition module, a coefficient calculation module, a stitch distance judgment module, and a flatness judgment module, the relationship between distances in a planar image and spatial distances can be constructed. Thus, by recognizing distances in a top-view image, the spatial distance at a given location can be inferred. Subsequently, by converting the image size ratio, the actual distance can be obtained. This allows for more accurate recognition of the spacing between needle holes and the flatness of the seam in the image, enabling accurate identification of problems with needle hole arrangement and seam flatness, and timely correction of seam parameters. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the high-speed quality inspection system for high-end fashion seam trajectories based on AI vision, as described in this invention. Figure 2 This is a schematic diagram illustrating the process of obtaining the sample interval of the stitch length of high-end fashion seams through analysis according to the present invention. Figure 3 This is a schematic diagram illustrating the process of forming a sewing area at the junction of two fabrics in a fabric assembly according to the present invention. Figure 4 This is a schematic diagram of the process of identifying at least one needle hole location in the area to be sutured according to the present invention; Figure 5 This is a schematic diagram of the process for identifying the average curvature coefficient of the sampling points according to the present invention; Figure 6 This is a schematic diagram of the process for obtaining three-view images of a baseline line segment according to the present invention; Figure 7 This is a schematic diagram of the process of identifying the distance of the needle hole position in the needle hole assembly based on the average bending coefficient according to the present invention. Figure 8 This is a schematic diagram illustrating the process of obtaining the actual distance between adjacent feature points in the analysis of this invention. Detailed Implementation

[0015] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0016] Reference Figure 1 As shown, a high-speed quality inspection system for high-end fashion seam trajectories based on AI vision includes: The data acquisition module analyzes and obtains the sample interval of the stitch length of high-end fashion seams, and analyzes and obtains the allowable error of the seams. The region identification module acquires at least one fabric that constitutes high-end fashion, takes two fabrics that have an intersection as a fabric combination, and forms a region to be sewn at the intersection of the two fabrics in the fabric combination. If the stitch line goes beyond the region to be sewn, there is a problem with the sewing; otherwise, no processing is done. The combination determination module identifies at least one needle hole location in the area to be sutured, and combines adjacent needle hole locations as needle hole combinations. A coefficient calculation module, wherein the coefficient calculation module uniformly takes at least one sampling point in the area to be stitched and identifies the average curvature coefficient of the sampling point; The stitch distance judgment module identifies the distance between the needle positions in the needle eye combination based on the average curvature coefficient, which is taken as the actual stitch distance of the needle eye combination. If the difference between the actual stitch distance and the sample interval exceeds the allowable error of the suture, then there is a problem with the distance between the needle positions in the needle eye combination; otherwise, no processing is performed. The flatness judgment module takes the suture portion between the needle holes in the needle hole combination as the feature suture of the needle hole combination. It takes at least one feature point evenly on the feature suture and analyzes the actual distance between adjacent feature points. It accumulates the actual distances between adjacent feature points in the feature suture to obtain the length of the feature suture. If the difference between the length of the feature suture and the sample interval exceeds the allowable error of the suture, then there is a problem with the flatness of the suture at the needle hole position in the needle hole combination. Otherwise, no processing is performed.

[0017] When performing stitch trajectory detection, it is necessary to determine whether the stitching follows the predetermined line. This is relatively easy to determine. However, since there is an allowable error in stitching, the error needs to be taken into account during the identification process. A series of steps will be set up to handle this in the subsequent process. In addition, the stitch length needs to be consistent. If it is not consistent, firstly, the appearance will be insufficient, and secondly, the force generated by the sewing will be inconsistent. Some places may have less force and some places may have more force. In the places with less force, the strength may be insufficient. The smoothness of the stitching also needs to be considered. Insufficient smoothness will result in poor aesthetics. Furthermore, insufficient smoothness indicates that the thread is not tightly attached to the fabric, meaning the thread is not taut and straight, which indicates insufficient sewing strength. This can easily lead to problems with insufficient durability in these areas. Therefore, a series of steps will be set up to address this issue later.

[0018] Reference Figure 2 As shown, the sample interval for analyzing the stitch spacing of high-end fashion seams includes the following steps: Obtain at least one sample fashion item, and use the flat portion of the sample fashion item as the sample plane; The needle holes of the sutures in the sample plane are identified to obtain the sample needle holes. The distance between adjacent sample needle holes is measured to obtain the sample needle distance. The average of the needle distances for at least one sample is taken to obtain the sample interval.

[0019] The analysis to determine the permissible error of the suture includes the following steps: The suture allowable error is obtained by subtracting the maximum and minimum values ​​of the stitch distance of at least one sample.

[0020] Since stitches cannot be perfectly precise, it is necessary to set an allowable error for the stitches, which provides a certain degree of redundancy for identification.

[0021] Reference Figure 3 As shown, forming the area to be sewn at the junction of two fabrics in a fabric assembly includes the following steps: In the sample fashion, the overlapping part of two fabrics when they are sewn together is taken as the sample strip, the width of the sample strip is taken as the sample width, and the two edges of the sample strip that are parallel to the seam trajectory inside the sample strip are taken as the sample edges. The distance from the seam trace within the sample fabric strip to the sampling edge is used as the first sample distance and the second sample distance, respectively. The exposed edge at the overlapping position of two fabrics in the fabric combination is taken as the feature edge, and the fabric where the feature edge is located is taken as the feature fabric. A feature strip is generated on the feature fabric, satisfying that one edge of the feature strip is a feature edge, the feature strip is obtained by translating the feature edge, and the width of the feature strip is equal to the sample width. Generate a feature trajectory in the feature cloth, where the distance from the feature trajectory to the feature edge is equal to the distance of the first sample or the distance of the second sample; Two feature boundaries are formed on both sides of the feature trajectory. The feature boundaries are obtained by translating the feature trajectory. The distance from the feature boundary to the feature trajectory is equal to the allowable error of the suture. The area between the two feature boundaries is taken as the area to be sutured.

[0022] The area to be sutured is actually a redundant space formed based on the suture trajectory and the allowable error of the suture line. It is mainly used to identify the correctness of the suture trajectory route. Since the suture route cannot be completely precise, the error needs to be taken into account. Therefore, the area to be sutured is formed. As long as the suture trajectory can extend along the area to be sutured, it means that it meets the requirements. Normally, when sewing, the width of the overlapping part of two pieces of fabric is fixed, and the position of the sewing line in the overlapping part is fixed, that is, the distance from the sewing line to both sides of the overlapping part is fixed. Therefore, the accurate sewing line can be determined based on this, and the area to be sewn can be generated according to the allowable error of the sewing line.

[0023] Reference Figure 4 As shown, identifying at least one needle prick location in the area to be sutured includes the following steps: From a top-down perspective, at a preset shooting distance, an image of the area to be stitched is acquired. The pixels in the image of the area to be stitched are classified to obtain at least one set of pixels, satisfying that the pixel values ​​of the pixels in the set of pixels are consistent. Use the set of pixels with the smallest number of elements as the target set of pixels; Generate at least one initial set, which contains one of the pixels in the target pixel set; The initial set is expanded by adding adjacent pixels from the target pixel set to the initial set until no more elements are added to the initial set. At least one initial set is deduplicated to obtain a target initial set, and the area covered by the pixels in the target initial set is taken as the pinhole position.

[0024] The area to be stitched includes images of needle holes, seams, and fabric, with needle holes occupying the smallest proportion. The pixel values ​​of the needle holes, seams, and fabric are identical, thus three pixel sets can be identified. The set with the smallest number of elements represents the needle holes. However, each needle hole contains more than one pixel, so pixel aggregation is necessary. Different needle holes have non-adjacent pixels, while two pixels from the same needle hole are either adjacent or connected by one or more adjacent pixels. Therefore, an initial set can be formed, where each initial set contains a single needle hole pixel. However, duplicates exist, so deduplication is required.

[0025] Reference Figure 5 As shown, identifying the average curvature coefficient of the sampling points includes the following steps: The line connecting adjacent sampling points is taken as a sampling line segment, and one of the sampling line segments generated by the sampling point is taken as the reference line segment of the sampling point. The three-view image of the baseline segment is obtained by performing three-view image acquisition, resulting in the top view image, left view image, and front view image of the baseline segment; In the top-view image, the difference in longitudinal distance between the two ends of the baseline segment is identified and used as the top-view length; In the left view image, the difference in lateral distance between the two ends of the baseline segment is identified and used as the left view length; In the front view image, the difference in longitudinal distance between the two ends of the baseline segment is identified and used as the front view length; The first distance is calculated using the first length formula, and the second distance is calculated using the second length formula. Divide the second distance by the first distance to obtain the average curvature coefficient of the sampling points corresponding to the baseline segment; The formula for the first length is as follows: , Where L is the first distance, a is the top-view length, and b is the left-view length; The formula for the second length is as follows: , Where S is the second distance and c is the frontal length.

[0026] During recognition, the main method used here is to identify the image from above. However, there is a difference between the spatial distance and the image distance, as well as differences in the curvature of the fabric. Therefore, in order to take these factors into account and form the average curvature coefficient of the sampling points, the actual spatial distance can be inferred based on the average curvature coefficient of the sampling points during recognition.

[0027] Reference Figure 6 As shown, obtaining the three-view image of the baseline segment includes the following steps: A top-down view of the baseline segment is obtained by shooting at a preset distance. From a left-hand perspective, at a preset shooting distance, obtain a left-hand view image of the baseline segment; Using a frontal view and a preset shooting distance, obtain a frontal image of the baseline line segment.

[0028] Here, the methods for acquiring the top view, left view, and front view are the same, so their image sizes are all the same, and the scaling factor used is also the same as the actual size. Therefore, when acquiring the scaling factor, only the scaling factor of the top view needs to be acquired.

[0029] Reference Figure 7 As shown, identifying the distance to the needle hole position in a needle hole assembly based on the average curvature coefficient includes the following steps: The ratio of the actual size to the corresponding image size in the top view image is obtained in advance and used as a scaling factor; Using a top-down perspective and a preset shooting distance, a top-down image of the needle hole combination is obtained. The lines connecting the needle hole positions in the needle hole combination are obtained as feature lines. Sampling points whose distance from the feature lines is less than the allowable error of the suture are used as target sampling points. In the top view image at the pinhole position, draw a horizontal line through the target sampling point. The horizontal line is parallel to the horizontal direction and divides the feature line into at least one local line segment. Match the target sampling points on the two horizontal lines that form the local line segment to the local line segment. In the top view image at the needle hole position, obtain the image size of the local line segment, and take the average value of the average curvature coefficient of the target sampling points corresponding to the local line segment to obtain the reference coefficient of the local line segment. The image size of a local line segment is multiplied by its reference coefficient to obtain a preliminary size. The preliminary size of the local line segment is then multiplied by a scaling factor to obtain its actual size. The actual sizes of at least one local line segment are then superimposed to obtain the distance between the needle eye positions in the needle eye combination.

[0030] Here, we first determine the target sampling points used for distance calculation. Secondly, we need to determine the part that uses the average curvature coefficient of the target sampling points for calculation. Therefore, we use horizontal lines to segment the feature lines and match the target sampling points to local line segments. Since the local line segments are close to the matched target sampling points, we use the average curvature coefficient of the target sampling points for calculation, which matches the curvature of the local line segments. Therefore, the result calculated in this way is more in line with the actual situation. However, it should be noted that the distance calculated in this way is the image size. We still need to infer the true distance based on the relationship between the image size and the actual size. Here, the top-view image has two directions, horizontal and vertical, with the horizontal line parallel to the horizontal direction.

[0031] Reference Figure 8 As shown, the analysis to obtain the actual distance between adjacent feature points includes the following steps: Using a top-down perspective and a preset shooting distance, a top-down image of the feature point is obtained, and the sampling point closest to the feature point is paired with the feature point; The line connecting adjacent feature points is taken as the target line. The average curvature coefficient of the sampling points corresponding to the feature points at both ends of the target line is averaged to obtain the target coefficient of the target line. The actual distance between adjacent feature points is obtained by multiplying the image size of the target line in the top view image of the feature point, the target coefficient of the target line, and the scale coefficient.

[0032] The calculation process for the actual distance between adjacent feature points is similar to that for the distance to the needle eye.

[0033] Furthermore, this solution also proposes a storage medium on which a computer-readable program is stored. When the computer-readable program is invoked, it executes the aforementioned high-speed quality inspection system for high-end fashion seam trajectories based on AI vision.

[0034] It is understandable that the storage medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid-state drive (SSD).

[0035] In summary, the advantages of this invention are as follows: by setting up a region recognition module, a coefficient calculation module, a stitch distance judgment module, and a flatness judgment module, the relationship between distance and spatial distance in a planar image can be constructed. Thus, by recognizing the distance in the top-view image, the spatial distance at that location can be inferred. Consequently, by converting the image size ratio, the actual distance can be obtained. This allows for more accurate recognition of the spacing of needle holes and the flatness of the seam in the image, thereby accurately identifying problems with needle hole arrangement and seam flatness, and enabling timely correction of seam parameters.

[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A high-speed quality inspection system for high-end fashion seam trajectories based on AI vision, characterized in that, include: The data acquisition module analyzes and obtains the sample interval of the stitch length of high-end fashion seams, and analyzes and obtains the allowable error of the seams. The region identification module acquires at least one fabric that constitutes high-end fashion, takes two fabrics that have an intersection as a fabric combination, and forms a region to be sewn at the intersection of the two fabrics in the fabric combination. If the stitch line goes beyond the region to be sewn, there is a problem with the sewing; otherwise, no processing is done. The combination determination module identifies at least one needle hole location in the area to be sutured, and combines adjacent needle hole locations as needle hole combinations. A coefficient calculation module, wherein the coefficient calculation module uniformly takes at least one sampling point in the area to be stitched and identifies the average curvature coefficient of the sampling point; The stitch distance judgment module identifies the distance between the needle positions in the needle eye combination based on the average curvature coefficient, which is taken as the actual stitch distance of the needle eye combination. If the difference between the actual stitch distance and the sample interval exceeds the allowable error of the suture, then there is a problem with the distance between the needle positions in the needle eye combination; otherwise, no processing is performed. The flatness judgment module takes the suture portion between the needle holes in the needle hole combination as the feature suture of the needle hole combination. It takes at least one feature point evenly on the feature suture and analyzes the actual distance between adjacent feature points. It accumulates the actual distances between adjacent feature points in the feature suture to obtain the length of the feature suture. If the difference between the length of the feature suture and the sample interval exceeds the allowable error of the suture, then there is a problem with the flatness of the suture at the needle hole position in the needle hole combination. Otherwise, no processing is performed.

2. The high-speed quality inspection system for high-end fashion seam trajectories based on AI vision according to claim 1, characterized in that, The analysis to obtain the sample interval of the stitch length for high-end fashion seams includes the following steps: Obtain at least one sample fashion item, and use the flat portion of the sample fashion item as the sample plane; The needle holes of the sutures in the sample plane are identified to obtain the sample needle holes. The distance between adjacent sample needle holes is measured to obtain the sample needle distance. The average of the needle distances for at least one sample is taken to obtain the sample interval.

3. The high-speed quality inspection system for high-end fashion seam trajectories based on AI vision according to claim 2, characterized in that, The analysis to determine the permissible error of the suture includes the following steps: The suture allowable error is obtained by subtracting the maximum and minimum values ​​of the stitch distance of at least one sample.

4. The high-speed quality inspection system for high-end fashion seam trajectories based on AI vision according to claim 3, characterized in that, The process of forming the area to be sewn at the junction of two fabrics in a fabric assembly includes the following steps: In the sample fashion, the overlapping part of two fabrics when they are sewn together is taken as the sample strip, the width of the sample strip is taken as the sample width, and the two edges of the sample strip that are parallel to the seam trajectory inside the sample strip are taken as the sample edges. The distance from the seam trace within the sample fabric strip to the sampling edge is used as the first sample distance and the second sample distance, respectively. The exposed edge at the overlapping position of two fabrics in the fabric combination is taken as the feature edge, and the fabric where the feature edge is located is taken as the feature fabric. A feature strip is generated on the feature fabric, satisfying that one edge of the feature strip is a feature edge, the feature strip is obtained by translating the feature edge, and the width of the feature strip is equal to the sample width. Generate a feature trajectory in the feature cloth, where the distance from the feature trajectory to the feature edge is equal to the distance of the first sample or the distance of the second sample; Two feature boundaries are formed on both sides of the feature trajectory. The feature boundaries are obtained by translating the feature trajectory. The distance from the feature boundary to the feature trajectory is equal to the allowable error of the suture. The area between the two feature boundaries is taken as the area to be sutured.

5. The high-speed quality inspection system for high-end fashion seam trajectories based on AI vision according to claim 4, characterized in that, Identifying at least one needle prick location in the area to be sutured includes the following steps: From a top-down perspective, at a preset shooting distance, an image of the area to be stitched is acquired. The pixels in the image of the area to be stitched are classified to obtain at least one set of pixels, satisfying that the pixel values ​​of the pixels in the set of pixels are consistent. Use the set of pixels with the smallest number of elements as the target set of pixels; Generate at least one initial set, which contains one of the pixels in the target pixel set; The initial set is expanded by adding adjacent pixels from the target pixel set to the initial set until no more elements are added to the initial set. At least one initial set is deduplicated to obtain a target initial set, and the area covered by the pixels in the target initial set is taken as the pinhole position.

6. The high-speed quality inspection system for high-end fashion seam trajectories based on AI vision according to claim 5, characterized in that, The process of identifying the average curvature coefficient of the sampling points includes the following steps: The line connecting adjacent sampling points is taken as a sampling line segment, and one of the sampling line segments generated by the sampling point is taken as the reference line segment of the sampling point. The three-view image of the baseline segment is obtained by performing three-view image acquisition, resulting in the top view image, left view image, and front view image of the baseline segment; In the top-view image, the difference in longitudinal distance between the two ends of the baseline segment is identified and used as the top-view length; In the left view image, the difference in lateral distance between the two ends of the baseline segment is identified and used as the left view length; In the front view image, the difference in longitudinal distance between the two ends of the baseline segment is identified and used as the front view length; The first distance is calculated using the first length formula, and the second distance is calculated using the second length formula. Divide the second distance by the first distance to obtain the average curvature coefficient of the sampling points corresponding to the baseline segment; The formula for the first length is as follows: ， Where L is the first distance, a is the top-view length, and b is the left-view length; The formula for the second length is as follows: ， Where S is the second distance and c is the frontal length.

7. The high-speed quality inspection system for high-end fashion seam trajectories based on AI vision according to claim 6, characterized in that, The process of obtaining the three-view image of the baseline segment includes the following steps: A top-down view of the baseline segment is obtained by shooting at a preset distance. From a left-hand perspective, at a preset shooting distance, obtain a left-hand view image of the baseline segment; Using a frontal view and a preset shooting distance, obtain a frontal image of the baseline line segment.

8. The high-speed quality inspection system for high-end fashion seam trajectories based on AI vision according to claim 7, characterized in that, The method of identifying the distance to the needle eye position in the needle eye assembly based on the average curvature coefficient includes the following steps: The ratio of the actual size to the corresponding image size in the top view image is obtained in advance and used as a scaling factor; Using a top-down perspective and a preset shooting distance, a top-down image of the needle hole combination is obtained. The lines connecting the needle hole positions in the needle hole combination are obtained as feature lines. Sampling points whose distance from the feature lines is less than the allowable error of the suture are used as target sampling points. In the top view image at the pinhole position, draw a horizontal line through the target sampling point. The horizontal line is parallel to the horizontal direction and divides the feature line into at least one local line segment. Match the target sampling points on the two horizontal lines that form the local line segment to the local line segment. In the top view image at the needle hole position, obtain the image size of the local line segment, and take the average value of the average curvature coefficient of the target sampling points corresponding to the local line segment to obtain the reference coefficient of the local line segment. The image size of a local line segment is multiplied by its reference coefficient to obtain a preliminary size. The preliminary size of the local line segment is then multiplied by a scaling factor to obtain its actual size. The actual sizes of at least one local line segment are then superimposed to obtain the distance between the needle eye positions in the needle eye combination.

9. A high-speed quality inspection system for high-end fashion seam trajectories based on AI vision, as described in claim 8, is characterized in that... The analysis to obtain the actual distance between adjacent feature points includes the following steps: Using a top-down perspective and a preset shooting distance, a top-down image of the feature point is obtained, and the sampling point closest to the feature point is paired with the feature point; The line connecting adjacent feature points is taken as the target line. The average curvature coefficient of the sampling points corresponding to the feature points at both ends of the target line is averaged to obtain the target coefficient of the target line. The actual distance between adjacent feature points is obtained by multiplying the image size of the target line in the top view image of the feature point, the target coefficient of the target line, and the scale coefficient.

Images