Aluminum profile positioning and recognition method based on image recognition

Abstract

The present invention relates to the field of image processing technology, and more specifically, to an aluminum profile positioning and identification method based on image recognition, comprising: obtaining a grayscale image of an aluminum profile image; calculating a response function value of each pixel in the grayscale image using a Harris algorithm, and weighting the response function value according to the grayscale anomaly and geometric anomaly of the pixel to obtain a final response function value of each pixel, wherein the grayscale anomaly of the pixel is positively correlated with the variance of the grayscale values ​​of all pixels in the neighborhood of the pixel. The present invention obtains a response weight of each pixel based on the grayscale anomaly and geometric anomaly of each pixel, obtains a final response function value based on the response weight of each pixel and the response function value, and marks pixels with a high final response function value as corner points, thereby reducing the problem of false corner points generated on the boundaries of shadow and reflective areas, making subsequent corner point matching more accurate, and improving the positioning accuracy of the aluminum profile.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and more specifically, to an aluminum profile positioning and recognition method based on image recognition. Background Art

[0002] Aluminum profiles, made primarily of aluminum, have specific shapes and cross-sections. They are primarily used in a wide range of fields, including construction and industry, and are indispensable for modern construction. During the extrusion process, freshly extruded profiles must be removed promptly to ensure continuous production. A tractor pulls the aluminum profiles at a steady speed, preventing them from accumulating at the die exit and potentially disrupting production. Precise positioning of the profiles is crucial in controlling the tractor's pulling of the aluminum profiles. This ensures uniform force and deformation during the pulling process, ensuring product quality, enabling smooth transitions to subsequent processing steps, and improving the efficiency and accuracy of the entire production process. With the advancement of computer vision and image recognition technologies, automated inspection systems based on image recognition have become an effective solution to this problem. Image recognition technology captures images of the aluminum profile surface using a camera and then processes and analyzes them using computer vision algorithms, enabling precise positioning and identification.

[0003] In related technologies, for example, Chinese patent document CN108898148B discloses a digital image corner detection method, system, and computer-readable storage medium. This method can accurately find all real corner points in the image contour, has the characteristics of high positioning accuracy and strong robustness, and can effectively realize corner detection and positioning of two-dimensional images, overcoming the disadvantage of low corner positioning accuracy.

[0004] Currently, when positioning aluminum profiles using Harris corner detection, the image of the aluminum profile contains shadows and reflective areas due to the uneven intensity of reflected light on the surface of the aluminum profile. Points with high corner response function values ​​will also appear on the boundaries of these areas and will be identified as corners, resulting in inaccurate subsequent corner point matching and ultimately inaccurate positioning of the aluminum profile. Summary of the Invention

[0005] The present invention provides an aluminum profile positioning and identification method based on image recognition, which aims to solve the problem in related technologies that shadows and reflective areas exist in aluminum profile images, and points with high corner point response function values ​​are also generated on the boundaries of such areas, which will be identified as corner points, resulting in inaccurate subsequent corner point matching.

[0006] The present invention provides an aluminum profile positioning and recognition method based on image recognition, comprising: obtaining a grayscale image of an aluminum profile image; calculating a response function value of each pixel point in the grayscale image using a Harris algorithm, and weighting the response function value according to the grayscale anomaly and geometric anomaly of the pixel point to obtain a final response function value of each pixel point, wherein the grayscale anomaly of the pixel point is positively correlated with the variance of the grayscale values ​​of all pixels in the pixel point neighborhood, and determining the pixel point whose final response function value is greater than a preset threshold as a corner point, and realizing positioning of the aluminum profile based on the corner point; wherein calculating the geometric anomaly of the pixel point comprises: detecting the edge of the grayscale image to obtain an edge set composed of multiple edges, wherein the geometric anomaly of the pixel point is negatively correlated with the Euclidean distance between the pixel point and the edge n center point, and is also negatively correlated with the edge The reliability of aluminum profile is negatively correlated; the edge Aluminum profile reliability The calculation formula is: , where For the edge All pixels to the edge Shortest Euclidean distance to the centroid; is the total number of edges in the edge set, is the normalization function; 、 Edge ,edge The roller confidence reflects the number of non-repeated pixels in the horizontal coordinates of all pixels on the edge; 、 Edge ,edge The smoothness of the image is calculated using the edge chain code method. Based on the grayscale changes in each pixel's neighborhood, we preliminarily determine which pixels are likely to be in shadow or reflective areas. Furthermore, based on the edges near each pixel, we further determine which pixels are likely to be in shadow or reflective areas. Finally, we obtain the response weight of each pixel and use it to weight the response function value to obtain the final response function value. This allows us to accurately screen corner points and improve positioning accuracy.

[0007] Furthermore, positioning the aluminum profile based on the corner points includes: obtaining a template image and extracting the corner points in the template image; constructing corresponding feature matrices based on the neighborhood range of each corner point in the aluminum profile image and the template image, obtaining the normalized cross-correlation value between the feature matrices, and selecting the set of corner points with the highest normalized cross-correlation value as matching corner points; and using the matching corner points to obtain a transformation matrix to locate the aluminum profile. This method can achieve precise positioning of the aluminum profile.

[0008] Furthermore, the response function value is weighted, including: calculating the response weight of each pixel point, and the calculation formula is: Where, Pixel The response weight, Pixel The grayscale abnormality, Pixel The geometric anomaly degree of the pixel is calculated; the final response function value of the pixel is obtained by multiplying the pixel's response weight by the pixel's response function value. The response weight of the corresponding response function value of the pixel is assigned according to the pixel's grayscale anomaly degree and geometric anomaly degree, which improves the accuracy of the subsequent calculation of the final response function of each pixel and improves the accuracy of corner point identification.

[0009] Furthermore, the calculation formula for the geometric anomaly degree of the pixel point is: Where, is the geometric abnormality of the pixel point, For the edge Belong to the credibility of aluminum profile, Pixels and edges The Euclidean distance of the center point, is the total number of edges in the edge set of the aluminum profile image, The function is An exponential function with base .

[0010] Furthermore, the edge chain code method is used to calculate edge smoothness. This method involves extracting an 8-directional chain code for each edge in each grayscale image, starting from the upper-left pixel of the grayscale image; obtaining the sum of the absolute values ​​of the first-order difference sequence of the linearly normalized chain code for each edge as the edge's suddenness, and subtracting the suddenness of each edge from one to obtain the smoothness of each edge. The chain code method can effectively describe edge morphology in different directions, particularly in areas with complex shapes or edges in multiple directions. The 8-directional chain code not only avoids directional limitations but also adapts to the changing morphology of complex surfaces such as aluminum profiles, further improving the comprehensiveness and robustness of edge recognition.

[0011] Furthermore, obtaining the roller confidence of an edge involves: obtaining the set of horizontal coordinates of all pixels on each edge, determining a single, non-repeating horizontal coordinate in the set of horizontal coordinates for each edge as a measurement coordinate; obtaining the total number of pixels on each edge whose horizontal coordinates fall within a preset fluctuation range of the measurement coordinate as a measurement value, and using the highest linearly normalized measurement value among the edges as the roller confidence of the edge, where the preset fluctuation range is 2. During the pulling process, a normal roller edge has a large number of pixels with small horizontal coordinate variations. Therefore, this characteristic of the large number of pixels with small horizontal coordinate variations on the edge can be used to accurately calculate the roller confidence of each edge.

[0012] Furthermore, the calculation formula for the grayscale abnormality of a pixel is: Where, is the grayscale abnormality of the pixel; is the variance of the grayscale values ​​of all neighboring pixels of the pixel; The pixel The neighboring pixels and The distance between neighboring pixels; The pixel Gray value of neighboring pixels; The pixel Gray value of neighboring pixels; For the natural constant An exponential function with base ; is the normalization function, is the number of pixels in the neighborhood pixel set, where the neighborhood pixel set is composed of all pixels in the neighborhood range centered on the pixel.

[0013] Furthermore, the empirical value of the preset threshold is 0.8.

[0014] Beneficial effect: By obtaining the grayscale anomaly and geometric anomaly of each pixel point according to the aluminum profile image, obtaining the response weight of each pixel point according to the grayscale anomaly and geometric anomaly of each pixel point, and obtaining the final response function value according to the response weight of each pixel point and the corner point response function value, the pixel points with high final response function values ​​are marked as corner points, thereby reducing the problem of false corner points generated on the boundaries of shadow and reflective areas, making subsequent corner point matching more accurate, and improving the positioning accuracy of aluminum profiles. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The following detailed description is read with reference to the accompanying drawings, which illustrate several embodiments of the present invention by way of example and not limitation, and in which like or corresponding reference numerals designate like or corresponding parts, wherein:

[0016] Figure 1FIG. 4 is a flowchart schematically illustrating a method for identifying corner points according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0018] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0019] It should be noted that during the aluminum profile extrusion process, the aluminum profile just extruded from the die needs to be removed in a timely manner to make room for the subsequent continuous extrusion process, so that the tractor can pull out the aluminum profile at a stable speed and the extrusion production line can continue to operate, thereby avoiding production interruptions caused by the accumulation of aluminum profiles at the die outlet. In order to ensure that the tractor accurately pulls the aluminum profile, it is necessary to accurately locate the position of the aluminum profile based on image recognition.

[0020] In one embodiment, when locating aluminum profiles using Harris corner detection, the uneven intensity of reflected light on the aluminum profile surface causes shadows and reflective areas in the profile image. Points with high corner response function values ​​also appear on the boundaries of these areas and are identified as corners, leading to inaccurate subsequent corner matching and ultimately inaccurate positioning of the aluminum profile. Therefore, the present invention addresses this issue through the following steps. Harris corner detection is a state-of-the-art technique and will not be described in detail here.

[0021] In one embodiment, the grayscale and geometric anomaly degrees of each pixel are obtained from the aluminum profile image. A response weight is then determined for each pixel based on these values. The corner response function value is weighted according to the pixel response weight to obtain a final response function value. Pixels with high final response function values ​​are marked as corner points, completing corner point matching. This reduces the problem of false corner points generated at the boundaries of shadows and reflective areas, making subsequent corner point matching more accurate and improving the positioning accuracy of aluminum profiles.

[0022] like Figure 1 As shown, S101: obtaining a grayscale image of the aluminum profile image.

[0023] Specifically, an industrial camera is placed on top of the tractor to capture real-time images of the aluminum profile in a vertically downward direction. These images are then transmitted to a host computer and converted into grayscale images, resulting in a grayscale image of the aluminum profile. It should be noted that due to the workshop environment, mechanical vibrations from the tractor and nearby electromagnetic interference can cause noise in the image. Therefore, the grayscale image must be denoised using Gaussian filtering to reduce the noise impact.

[0024] In one embodiment, pixels with high corner response function values ​​at the boundaries of shadow and light reflection areas caused by uneven reflected light intensity on the aluminum profile surface are primarily distributed at these boundaries. The boundaries of shadow and light reflection areas on the aluminum profile surface appear as gradient grayscale values ​​in the image, so the grayscale changes within the neighborhood of each pixel can be used to preliminarily determine which pixels are likely to be in shadow or light reflection areas.

[0025] S102: Calculate the grayscale abnormality of the pixel.

[0026] In one embodiment, for any pixel in the grayscale image, an empirical neighborhood range of the pixel is set. In this embodiment, the size of the empirical neighborhood range is set to , select the pixel centered All pixels within the range of are taken as the set of neighboring pixels of the pixel point. The Euclidean distance of the coordinates of any two different pixels in the image among all the neighboring pixels of the pixel point is taken as the distance between the two pixels. The variance of the grayscale values ​​of all the neighboring pixels of the pixel point is obtained. Then the grayscale abnormality of the pixel point is calculated. The calculation formula is: Where, is the grayscale abnormality of the pixel; is the variance of the grayscale values ​​of all neighboring pixels of the pixel; The pixel The neighboring pixels and The distance between neighboring pixels; The pixel Gray value of neighboring pixels; The pixel Gray value of neighboring pixels; For the natural constant An exponential function with base ; is the normalization function, is the number of pixels in the neighborhood pixel set.

[0027] in, Represents the grayscale smoothness of the pixel's neighborhood pixel set. When the absolute grayscale difference between all the pixel's neighborhood pixels and the closer pixels is smaller, the grayscale smoothness of the pixel's neighborhood pixel set is higher. When the variance of the grayscale values ​​of all the pixel's neighborhood pixels is higher and the grayscale smoothness of the pixel's neighborhood pixel set is higher, the grayscale change in the local range of the pixel is more consistent with the boundary of the shadow and reflective area caused by the uneven intensity of reflected light on the aluminum profile surface, and the grayscale anomaly of the pixel is higher.

[0028] In one embodiment, due to the complex grayscale of the background area in the aluminum profile image, the grayscale of some pixels located near the aluminum profile outline may also be abnormally high, and the boundaries of the shadow and reflective areas on the aluminum profile surface are usually located within the aluminum profile outline. Therefore, the pixels that are likely to be located in the shadow and reflective areas can be further judged based on the edges near each pixel.

[0029] S103: Calculate the geometric abnormality of the pixel point.

[0030] In one embodiment, the geometric abnormality of a pixel point is calculated, including: detecting the edge of the grayscale image to obtain an edge set consisting of multiple edges, the geometric abnormality of the pixel point is negatively correlated with the Euclidean distance between the pixel point and the edge n center point, and is also negatively correlated with the edge The reliability of the aluminum profile is negatively correlated.

[0031] It should be noted that when aluminum profiles are too long, they may have a certain curvature. Since the roller in the background area is fixed in direction, the edge of the roller must be used to determine the edge of the aluminum profile. During image capture, the camera is fixed directly above the tractor, which is pulling the aluminum profile in the image. For example, the tractor is pulling from left to right. During this process, a normal roller edge will have a large number of pixels with small changes in the horizontal coordinate. Therefore, based on this characteristic of a large number of pixels with small changes in the horizontal coordinate, the roller confidence of each edge can be calculated. This means that the probability of the edge being a roller edge reflects the number of unique horizontal coordinates of all pixels on the edge.

[0032] Specifically, the horizontal coordinate set of all pixel points on each edge is obtained, and a non-repeated single horizontal coordinate in the horizontal coordinate set of each edge is determined as the measurement coordinate; the total number of pixel points of each edge whose horizontal coordinate is within the preset fluctuation range of the measurement coordinate is obtained as the measurement value, and the highest measurement value of the edge after linear normalization is used as the roller confidence of the edge, where the preset fluctuation range is 2.

[0033] Exemplary, edge The horizontal coordinate set of all pixel points on 、 、 、 、 ], then the horizontal coordinate set has two non-repeated horizontal coordinates respectively 、 , and then count the edges within the preset fluctuation range of the two horizontal axes The number of pixels on the horizontal axis is 、 The total number of pixels within these two ranges is used as the edge The measurement value of .

[0034] In one embodiment, the edge smoothness is calculated using an edge chain code method, including: extracting the 8-directional chain code of the edge in each grayscale image with the upper left corner pixel point of the grayscale image as the starting point; obtaining the sum of the absolute values ​​of the first-order difference sequence of the linearly normalized chain code of each edge as the suddenness of each edge, and subtracting the suddenness of each edge from one to obtain the smoothness of each edge. After obtaining the smoothness of each edge, the influence of the complex edges in the background area can be reduced according to the magnitude of the edge smoothness. It should be noted that since the edges of the shadow and reflective areas are equally smooth, but these edges are located within the aluminum profile contour and are farther from the roller edge than the aluminum profile contour, the edges with a high degree of possibility of the aluminum profile contour can be identified according to the distance from the edge with a high degree of possibility to the roller edge, that is, the edges belong to the aluminum profile contour credibility.

[0035] It should be noted that since the aluminum profile moves horizontally during the pulling process, the possibility of the roller corresponding to the edge of the aluminum profile's true contour is low. Also, because the edges of the shadow and reflective areas are located within the true contour of the aluminum profile, they are farther away from the edge of the roller than the true contour of the aluminum profile. Therefore, the edge of the aluminum profile with high credibility can be identified based on the roller confidence.

[0036] Therefore, the edge can be calculated according to the above method Aluminum profile reliability , calculation formula: , where For the edge All pixels to the edge Shortest Euclidean distance to the centroid; is the total number of edges in the edge set, is the normalization function; 、 Edge ,edge Roller confidence; 、 For the edge ,edge smoothness.

[0037] in, Representing edges The possible extent that it does not belong to the edge of the drum; Representing edges The degree to which there are edges with high roller confidence nearby; when the probability that any edge is a roller edge near edge n is high, the probability of the true aluminum profile profile of edge n is higher. When the smoothness of edge n is high, the probability that edge n is not a roller edge is high, and the degree to which there are edges with high roller confidence near edge n is high, the probability of the true aluminum profile profile of edge n is high.

[0038] At this point, the geometric anomaly of the pixel point can be calculated based on the credibility of each edge being an aluminum profile outline. The calculation formula is: Where, is the geometric abnormality of the pixel point, For the edge For the reliability of aluminum profile, Pixels and edges The Euclidean distance of the center point, is the total number of edges in the edge set of the aluminum profile image, The function is It should be noted that the center point of an edge is the pixel point corresponding to the mean of the horizontal and vertical coordinates of all pixels on the edge.

[0039] Step S104: Calculate the final response function value of the pixel point and determine the corner point.

[0040] In one embodiment, pixels with high grayscale and geometric anomalies are more likely to be located on the boundaries of shadowed or reflective areas. Therefore, a smaller weight should be assigned when calculating the corner response function value to prevent them from being mistakenly identified as corners, which could lead to inaccurate corner matching and positioning of the aluminum profile. Therefore, the response function value needs to be weighted according to the pixel's grayscale and geometric anomalies to obtain the final response function value for each pixel.

[0041] In one embodiment, weighting the response function value includes calculating the response weight of each pixel point using the following formula: Where, Pixel The response weight, Pixel The grayscale abnormality, Pixel The geometric anomaly degree is calculated; the final response function value of the pixel is obtained based on the product of the pixel's response weight and the pixel's response function value. At this point, the final response function value of each pixel is obtained. Pixels with final response function values ​​greater than a preset threshold are determined as corner points, and the aluminum profile is positioned based on these corner points. The empirical value of the preset threshold is 0.8. In other embodiments, the empirical value of the preset threshold may be 0.85, 0.87, or the like.

[0042] S105: Positioning the aluminum profile based on the corner points.

[0043] In one embodiment, a template image is obtained and corner points in the template image are extracted; corresponding feature matrices are respectively constructed based on the neighborhood range of each corner point in the aluminum profile image and the template image, and the normalized mutual correlation value of the feature matrix between the above two is obtained, and a group of corner points with the highest normalized mutual correlation value is selected as matching corner points; the transformation matrix is ​​obtained using the matching corner points to realize the positioning of the aluminum profile.

[0044] It should be noted that since the tractor's towing position is marked in the template image, the transformation matrix between the aluminum profile image and the template image can be obtained through the corner point matching result, thereby locating the aluminum profile and controlling the tractor to tow the aluminum profile.

[0045] For example, when the image captured by the tractor does not contain aluminum profiles (when there are no edges with aluminum profile contour confidence higher than a threshold value of 0.8 in the edge detection results of the image), the tractor is set to move at a uniform speed in the horizontal direction toward the aluminum profile extrusion equipment until the image captured by the tractor contains aluminum profiles. At this time, the transformation matrix is ​​obtained based on the corner point matching results of the aluminum profile image and the template image, and the position of the tractor is controlled to move in the corresponding direction. After the adjustment is completed, the tractor is controlled to tow the aluminum profile, thereby achieving precise positioning of the aluminum profile, and the tractor is controlled to automatically tow the aluminum profile based on the positioning results.

[0046] Through the above steps, the grayscale anomaly and geometric anomaly of each pixel point are obtained according to the aluminum profile image, the response weight of each pixel point is obtained according to the grayscale anomaly and geometric anomaly of each pixel point, and the final response function value is obtained according to the response weight of each pixel point and the corner point response function value. The pixel points with high final response function values ​​are marked as corner points, thereby reducing the problem of false corner points generated on the boundaries of shadow and reflective areas, making subsequent corner point matching more accurate, and improving the positioning accuracy of aluminum profiles.

[0047] In the description of this specification, "multiple" and "several" mean at least two, such as two, three or more, unless otherwise clearly defined.

[0048] The technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0049] The above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, and all such variations and improvements fall within the scope of protection of the present invention.

Claims

1. A method for positioning and identifying aluminum profiles based on image recognition, characterized in that: include: Get the grayscale image of the aluminum profile image; The Harris algorithm is used to calculate the response function value of each pixel in the grayscale image, and the response function value is weighted according to the grayscale abnormality and geometric abnormality of the pixel to obtain the final response function value of each pixel. The pixel point with the final response function value greater than the preset threshold is determined as the corner point, and the aluminum profile is positioned based on the corner point. The calculation formula of the grayscale abnormality of the pixel point is: ； Where, is the grayscale abnormality of the pixel; is the variance of the grayscale values ​​of all neighboring pixels of the pixel; The pixel The neighboring pixels and The distance between neighboring pixels; The pixel Gray value of neighboring pixels; The pixel Gray value of neighboring pixels; For the natural constant An exponential function with base ; is the normalization function, is the number of pixels in the neighborhood pixel set, where the neighborhood pixel set is composed of all pixels in the neighborhood with the pixel as the center; Obtaining the geometric abnormality of the pixel point includes: performing edge detection on the grayscale image to obtain an edge set consisting of multiple edges, and calculating the geometric abnormality of the pixel point: ； Where, is the geometric abnormality of the pixel point, For the edge Belong to the credibility of aluminum profile, Pixels and edges The Euclidean distance of the center point, is the total number of edges in the edge set of the aluminum profile image; edge Aluminum profile reliability The calculation formula is: , where For the edge All pixels to the edge Shortest Euclidean distance to the centroid; is the total number of edges in the edge set; 、 Edge ,edge The roller confidence reflects the number of non-repeated pixels in the horizontal coordinates of all pixels on the edge; 、 Edge ,edge The smoothness is calculated using the edge chain code method.

2. The aluminum profile positioning and identification method based on image recognition according to claim 1, characterized in that: Positioning the aluminum profile based on the corner points includes: Acquire a template image and extract corner points in the template image; Based on the neighborhood range of each corner point in the aluminum profile image and the template image, corresponding feature matrices are respectively constructed, normalized cross-correlation values ​​of the feature matrices between the two are obtained, and a group of corner points with the highest normalized cross-correlation values ​​are selected as matching corner points; The transformation matrix is ​​obtained by matching corner points to realize the positioning of the aluminum profile.

3. The aluminum profile positioning and identification method based on image recognition according to claim 1, characterized in that: Weighting the response function value includes: Calculate the response weight of each pixel, the calculation formula is: ； Where, Pixel The response weight, Pixel The grayscale abnormality, Pixel The geometric anomaly degree; The final response function value of the pixel is obtained based on the product of the response weight of the pixel and the response function value of the pixel.

4. The aluminum profile positioning and identification method based on image recognition according to claim 1, characterized in that: The edge chain code method is used to calculate the smoothness of the edge, including: Taking the upper left corner pixel of the grayscale image as the starting point, extract the 8-directional chain code of the edge in each grayscale image; The sum of the absolute values ​​of the first-order difference sequence of the linearly normalized chain code of each edge is obtained as the mutation degree of each edge, and the smoothness of each edge is obtained by subtracting the mutation degree of each edge from one.

5. The aluminum profile positioning and identification method based on image recognition according to claim 1, characterized in that: Get the roller confidence of the edge, including: Obtain the horizontal coordinate set of all pixel points on each edge, and determine a non-repeated single horizontal coordinate in the horizontal coordinate set of each edge as the measurement coordinate; The total number of pixel points whose horizontal coordinates of each edge are within the preset fluctuation range of the measurement coordinates is obtained as the measurement value, and the highest measurement value among the edges after linear normalization is taken as the roller confidence of the edge. The roller confidence reflects the number of non-repetitive pixels in the horizontal coordinates of all pixel points on the edge, where the preset fluctuation range is 2.

6. The aluminum profile positioning and identification method based on image recognition according to claim 1, characterized in that: The empirical value of the preset threshold is 0.8.

Images