Method for extracting a middle line of a surface of a shoe outsole by laser

Abstract

The application relates to the technical field of line laser center line extraction, and discloses a method for extracting a line laser center line of a shoe outsole surface, which comprises the following steps: collecting an original image of the shoe outsole, extracting a green channel in the original image, and then performing a Gaussian convolution operation to obtain a pretreated image; then, quantile of each column is calculated, and whether the position is a laser stripe is judged according to whether the quantile of each column tends to be normally distributed, so that the line laser center line of the shoe outsole is obtained. The application can accurately and quickly judge different regions such as a bright area and a dark area of the laser stripe, and solves the problem that the light intensity threshold method cannot accurately determine the position of the laser stripe of an irregular and rough object.

Description

Technical Field

[0001] This invention relates to the field of line laser midline extraction technology, specifically a method for line laser midline extraction on the surface of a shoe outsole. Background Technology

[0002] Applying adhesive to the shoe outsole is a crucial step in the shoemaking process. Traditional manual adhesive application methods suffer from drawbacks such as slow speed and inconsistent quality. Therefore, an increasing number of shoe factories are adopting vision-guided adhesive application equipment for automation. Among the vision equipment options, line laser-based 3D reconstruction devices are widely used by manufacturers due to their high cost-effectiveness.

[0003] Line laser measurement technology is a non-contact active measurement technique with advantages such as high speed, high accuracy, and low cost, and it has wide applications in dimensional measurement, 3D reconstruction, medical fields, and robotics. The principle of line laser dimensional measurement is to project a laser stripe onto the surface of the object being measured, and then calculate the world coordinates of the surface contour of the object being measured based on the stripe image captured by a camera and the system parameters during calibration. Therefore, extracting the center of the laser stripe is a crucial step and a prerequisite for achieving a stable, real-time, and accurate measurement system. Currently, the determination of whether a location is a laser stripe is based on whether the pixel intensity distribution of a local image tends to a normal distribution; however, ordinary normal distribution verification algorithms are computationally slow.

[0004] Traditional line laser midline extraction algorithms typically use thresholding methods such as the Otsu threshold to determine the position of the light stripe, and then design a processing algorithm to calculate the center of the light stripe. This approach often requires high-quality light stripe images. However, the surface of shoe outsoles is irregular and rough, resulting in diffuse reflection during laser irradiation. This causes the pixel intensity in most dark areas of the laser light stripe to be even lower than the pixel intensity of noise. Algorithms using thresholding methods to determine the light stripe inevitably produce missing edges and even many noise particles. Therefore, existing technologies need to be improved. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for extracting the laser centerline of the outsole surface of a shoe, so as to quickly and accurately extract the laser centerline of the outsole.

[0006] To address the aforementioned technical problems, this invention provides a method for laser-based extraction of the center line of a shoe outsole surface, comprising the following specific steps:

[0007] S1. Acquire the original image of the shoe sole, and then extract the green channel from the original image in the computer to obtain image I. gray Then, for image I gray Perform Gaussian convolution to obtain the preprocessed image Mat r*cr is the number of rows in the image, and c is the number of columns in the image;

[0008] S2, Calculate the preprocessed image Mat r*c The quantiles in the j-th column, j = {1, 2, 3, ..., c};

[0009] S3. If the repeatability of the quantile in column j is 0, return to step S2 to calculate the quantile in column (j+1). If the repeatability is 1, proceed to step S4.

[0010] S4. Determine whether the repeated position of the quantile in column j is in the middle of the set of quantiles in column j. If the repeated position is not in the middle, return to step S2 to calculate the quantile in column (j+1). If the repeated position is in the middle, proceed to step S5.

[0011] S5. Calculate the midpoint of the laser in column j using the GGM algorithm;

[0012] S6. Determine whether to iterate through the preprocessed image Mat. r*c If all columns have not been traversed, return to step S2 to calculate the quantile of column (j+1). If they have been traversed, connect the laser midpoints of each column obtained in step S5 to form the laser midline of the shoe outsole.

[0013] As an improvement to the method for laser centerline extraction of shoe outsole surface lines according to the present invention:

[0014] The formula for extracting the green channel in step S1 is:

[0015]

[0016] Among them, I gray The image after extracting the green channel is I(G), where I(R) is the pixel value of the G channel of the RGB image, I(R) is the pixel value of the R channel of the RGB image, I(B) is the pixel value of the B channel of the RGB image, and T0 is the grayscale threshold.

[0017] The preprocessed image Mat r*c :

[0018]

[0019] Where g(x,y) is the Gaussian convolution kernel:

[0020]

[0021] Wherein, the magnitude of σ should satisfy D is the width of the laser stripe.

[0022] As a further improvement to the method for laser centerline extraction of shoe outsole surface lines according to the present invention:

[0023] The quantiles mentioned in step S2 are:

[0024]

[0025] in, Q is the m-quantile m-1 The i-th quantile in , int(q i ) represents the equally divided position q i The integer part, float(q) i ) represents the equally divided position q i The decimal part, D n in int(q i The value of the position;

[0026]

[0027] Where m is the number of equal divisions, i∈[1,m-1], and n is D. n The number of data points in the middle;

[0028] D n For the preprocessed image Mat r*c The set of pixel sorting distributions in the j-th column:

[0029]

[0030] Among them, I (i,j) Let D be the gray value at position (i,j) in the image. n Given a set of data arranged in ascending order, containing n data items.

[0031] As a further improvement to the method for laser centerline extraction of shoe outsole surface lines according to the present invention:

[0032] The repeatability mentioned in step S3 is:

[0033]

[0034] Among them, T R0 The repeat threshold is used; repeat time is the number of repetitions, when two adjacent quantiles are equal. When the number of repetitions is 1, the number of repetitions is incremented by 1.

[0035] As a further improvement to the method for laser centerline extraction of shoe outsole surface lines according to the present invention:

[0036] The GGM method calculates the coordinates of the laser center point in the j-th column.

[0037]

[0038] In the formula:

[0039]

[0040] R max R is the maximum value among the repeated quantiles. min is the minimum value among the repeating quantiles; (u,v) is the pixel position on the image, the position of the u-th row and v-th column, and w is the laser linewidth.

[0041] The beneficial effects of this invention are mainly reflected in:

[0042] 1. This invention uses whether the pixel intensity distribution of a local image tends to a normal distribution to determine whether the location is a laser stripe. This method can accurately, stably, and quickly identify different areas such as bright and dark areas of the laser stripe, solving the problem that the light intensity threshold method cannot accurately determine the location of laser stripes on irregular and rough objects.

[0043] 2. This invention proposes that the normality of data can be determined based on the repetition of quantiles, which is fast, simple and convenient. Attached Figure Description

[0044] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0045] Figure 1 This is a flowchart of a method for laser midline extraction of a shoe outsole surface according to the present invention;

[0046] Figure 2 A schematic diagram of the experimental platform for acquiring shoe outsole images according to this invention;

[0047] Figure 3 This is a partial schematic diagram of the line laser midline processing results extracted by the present invention and the gray-scale centroid method based on Otsu's threshold, the Steger method based on Otsu's threshold, and the Steger method using a small threshold.

[0048] Figure 4 Comparison images of the original images of three different shoe soles with the point clouds generated after processing by the algorithm of this invention and the traditional thresholding algorithm;

[0049] Figure 5 This is a comparison chart of the number of valid points extracted by the present invention, the gray-scale centroid method based on Otsu's threshold, the Steger method based on Otsu's threshold, and the Steger method using a small threshold.

[0050] Figure 6 This is a comparison chart showing the number of invalid points extracted by the present invention, the gray-scale centroid method based on Otsu's threshold, the Steger method based on Otsu's threshold, and the Steger method using a small threshold.

[0051] Figure 7 This is a comparison chart of the overall error of the present invention, the error of taking the same column points in GGM-Otsu, and the error of the GGM-Otsu algorithm;

[0052] Figure 8 The image shows a partial image of the original shoe sole after preprocessing, where A is the bright area, B is the dark area, C is the astigmatic area, D is the light hole area, and E is the light hole area with light stripes. Detailed Implementation

[0053] The present invention will be further described below with reference to specific embodiments, but the scope of protection of the present invention is not limited thereto:

[0054] Example 1: A method for laser extraction of the center line of a shoe outsole surface, such as... Figure 1 As shown, the specific process is as follows:

[0055] 1. Acquire and preprocess the image

[0056] The shoe sole was placed on a backlight (model FJI-ASC350350-W), and the object was photographed from above at a 45° angle using a Hikvision MV-CA050-10GM industrial camera and a Hikvision 16mm lens. Figure 2 As shown, an original image of the shoe outsole is acquired for preprocessing. The acquired image is in BMP format and has a size of 2048*480.

[0057] Then, the original image of the shoe outsole is input into the computer, and the green channel is extracted from the original image using the following formula:

[0058]

[0059] In the formula, I represents the original image of the shoe outsole. gray The image after extracting the green channel (the green channel accounts for the largest proportion), I(G) is the pixel value of the G channel of the RGB image, I(R) is the pixel value of the R channel of the RGB image, I(B) is the pixel value of the B channel of the RGB image, and T0 is the preset grayscale threshold, which is an empirical value related to the camera's exposure, the power of the line laser, etc.

[0060] Then process the extracted image I gray Perform Gaussian convolution to obtain the preprocessed image Mat r*c The calculation is shown in equation (2).

[0061]

[0062] In the formula, r is the number of rows in the image, c is the number of columns in the image, and g(x,y) is the Gaussian convolution kernel:

[0063]

[0064] In the formula, the magnitude of σ should satisfy... D is the width of the laser stripe.

[0065] 2. Calculate the preprocessed image Mat r*c The set D of pixel sorting distribution in column j (j = {1, 2, 3, ..., c}) n The calculation method is as follows:

[0066]

[0067] In the formula, I (i,j) Let D be the gray value at position (i,j) in the image. n Given a set of data arranged in ascending order, containing n data items.

[0068] 3. Calculate D n The equal division positions in the middle, the equal division positions q i The calculation method is as follows:

[0069]

[0070] In the formula, m represents the number of equal parts, and q represents the number of equal parts. i Let n be the i-th equidistant position, where i ∈ [1, m-1]; n is the position of D. n The number of data points in the data.

[0071] Once the equally divided positions are determined, set D can be calculated. n m-quantile in:

[0072]

[0073] In the formula, Q is the m-quantile m-1 The i-th quantile in the array has a total of m-1 quantiles, int(q i ) is q i The integer part, float(q) i ) is q i The decimal part, D n in int(q i )Location;

[0074] 4. Calculate the quantile Q m-1 Repeating cases: when two adjacent quantiles are equal. To repeat once, increment the repeat time by 1; if the repeat time is less than the preset repeatness threshold T, the repeat count is incremented by 1. R0 (T R0When the value is empirical, the repeatability (Repeat) is recorded as 0, and then the process returns to step 2 to calculate the quantiles of the next column (j+1); otherwise, the repeatability (Repeat) is recorded as 1 and the process proceeds to the next step to calculate the repeating position of the quantiles.

[0075]

[0076] By calculating the repetition of quantiles, it is possible to quickly determine D. n Is the region normally distributed?

[0077] 5. Determine if the repeated position of a quantile is located in the middle of the set of quantiles in that column:

[0078]

[0079] If the repeated positions of the quantiles are not concentrated in the middle position, D n The distribution will not be a normal distribution, which will cause the algorithm to fail. When the repeated position is not in the middle position (Mid=0), return to step 2 to calculate the quantile of the next column (j+1). If the repeated position is in the middle position (Mid=1), it means that the quantile of the j-th column satisfies the normal distribution. By judging whether it tends to the normal distribution, we can determine whether the position is a laser light stripe.

[0080] Normality testing can usually be performed using the QQ plot method, but the QQ plot is a relatively intuitive method and not suitable for calculation. However, the present invention can determine the normality of data based on the m-quantile and quantile repetition, which is simple and fast.

[0081] 6. Calculate the coordinates of the laser center point in column j using the GGM method:

[0082]

[0083] In the formula:

[0084]

[0085] R max R is the maximum value among the repeated quantiles. min is the minimum value among the repeating quantiles; (u,v) is the pixel position on the image, the position of the u-th row and v-th column, and w is the laser linewidth.

[0086] 7. Repeat steps 2-6 until the preprocessed image Mat is traversed. r*c Extract the midpoint of the laser in each column of the image and connect them to form a line to obtain the laser midline.

[0087] Experiment 1

[0088] The experimental environment is shown in Table 1. The light source used in this experiment has a wavelength of 550nm, a power of 50mW, and a line width of 3mm; the camera resolution is 2448*2048 pixels. The image of the shoe outsole was acquired and the line laser centerline of the shoe outsole was obtained according to the method for extracting the line laser centerline of the shoe outsole surface according to Embodiment 1 of the present invention. At the same time, the line laser centerline of the shoe outsole was extracted and obtained from the same acquired shoe outsole image by the gray-scale centroid method based on Otsu threshold (GGM-Otsu), the Steger method based on Otsu threshold (Steger-Otsu), and the Steger method using small threshold (Steger-Tsmall) respectively for comparative experiments.

[0089] Table 1 Experimental Environment

[0090]

[0091] 1. Comparison of extraction results

[0092] A valid column is defined as a column from which only one point is extracted from the laser column.

[0093] Invalid columns are defined as columns in which multiple points are extracted from a laser column or columns in a non-laser column that lack extracted points.

[0094] Actual laser column count: If a pixel in a column of the image has a pixel value greater than or equal to a preset grayscale threshold T0, then that column is considered an actual laser column. The actual laser column count can be obtained by traversing the entire image.

[0095] The total number of laser beams can be calculated based on the start and end positions of each laser segment in the image. For example... Figure 8 Region (E) represents a laser segment, and the continuous laser regions containing (A), (B), and (C) are each laser segments; region (D) can be considered a non-laser region. Then, columns from each laser segment and each non-laser region are extracted from the image. The presence or absence of duplicate columns in each laser segment is calculated; columns without duplicates are considered valid; duplicate columns and columns from non-laser regions are considered invalid.

[0096] The results of this invention in extracting the midline of a linear laser, compared with those of the Gray-Scale Centroid Method (GGM-Otsu) based on the Otsu threshold, the Steger Method (Steger-Otsu) based on the Otsu threshold, and the Steger Method (Steger-Tsmall) using a small threshold, are as follows: Figure 3 As shown, the number of valid points extracted by various algorithms is as follows: Figure 5 As shown in Figure 6, the number of invalid points extracted by each algorithm is compared with the actual number of laser points P. all The ratio between them, the number of invalid points and the actual number of laser points P allThe ratios between them are shown in Table 2.

[0097] Table 2 shows the number of valid points and invalid points compared to the actual number of laser points P. all The ratio between

[0098]

[0099] As can be seen from Table 2, compared with other threshold algorithms, the algorithm in this paper can extract more valid columns while extracting fewer invalid columns.

[0100] 2. Extraction Error Analysis

[0101] Considering the unknown true value of the laser center, the midpoint value obtained by the Steger method is used as the standard for quantitative analysis of the extraction error, and the error is calculated based on this standard. Given that the GGM-Otsu algorithm is almost ineffective in extracting points in dark areas, the error analysis includes the overall error of this invention, the error of using the same column points in the GGM-Otsu algorithm, and the error of the GGM-Otsu algorithm itself. Error calculations were performed on 10 images, and the error calculation method is shown below:

[0102]

[0103] In the formula, bias represents the deviation between the center point extracted by different algorithms on the valid column and the center point extracted by the Steger algorithm on the same valid column. i is the number of valid columns extracted by different algorithms on this image.

[0104] The overall error of this invention (ours-all points), the error of this invention taking the same column points as GGM-Otsu (ours-same as GGM), and the error results of the GGM-Otsu algorithm are as follows: Figure 7 As shown in Table 3, the average error is as follows.

[0105] Table 3 Overall Average Error Analysis (pixels)

[0106]

[0107] The average error in Table 3 is the average error of 10 images. It can be seen that the overall error of the algorithm of the present invention is 0.0305 pixels smaller than that of GGM-Otsu. When the error of dark areas is not considered, that is, when the points used to calculate the error are the same as those of GGM-Otsu, the error is 0.0794 pixels smaller.

[0108] In addition, the errors in different regions of the 10 laser images were calculated using the same method as the overall average error calculation method, except that the value of i varied according to different regions.

[0109] The error results for different regions are shown in Table 4.

[0110] Table 4 Error Analysis in Different Regions (pixels)

[0111]

[0112] The algorithm of this invention is essentially based on GGM, employing QR detection to locate the ray and adjusting the laser width at each position according to the repeatability. Therefore, in bright areas and astigmatic areas where noise and brightness are minimal, the accuracy is similar. In light-hole areas with light streaks, due to excessive noise and the stable intensity of the laser itself, the extraction accuracy of this invention's algorithm is much higher than that of GGM-Otsu. In dark laser areas, the algorithm of this invention has a relatively large error, mainly because the laser streaks in this area are tilted in the image. Using the GGM method to calculate tilted lasers will result in relatively large errors, especially at the boundary between bright and dark areas.

[0113] 3. Algorithm stability

[0114] For example Figure 4 The three shoe outsoles shown were captured with 80, 120, and 150 images respectively. The extracted laser center points were converted into 3D point clouds for viewing the overall effect. One of the results is as follows: Figure 4 As shown, considering that threshold-based algorithms generally perform similarly, only the Steger-Otsu and Steger-Tsmall algorithms were selected for comparison. Figure 4 The results show that the algorithm of this invention extracts 3D point clouds with fewer noise points and more details. Especially at the high edge of the shoe outsole, the algorithm of this invention can stably extract these boundaries, while the Steger-Otsu method cannot. If the threshold of the Steger method is lowered, too many noise points will be introduced.

[0115] To test the operating speed of the present invention, 10 different shoe outsole images were used as input. The processing time of the present invention compared with GGM-Otsu, teger-Otsu and Steger-Tsmall is shown in Table 5 (the average time for processing 10 images is taken).

[0116] Table 5 Running Time

[0117]

[0118] The algorithm presented in this paper is essentially based on the traditional GGM algorithm, performing a statistical analysis of all pixel values ​​in the image and calculating the quantiles and repetition rates of each column. GGM-Ostu, on the other hand, calculates the inter-class variance of all pixels in the image, building upon the GGM algorithm. Both the runtime representation and the logical derivation demonstrate that this invention has lower runtime complexity, only about 0.007 seconds slower than the GGM-Ostu algorithm. The Steger method, however, requires five convolutions on the image, resulting in a relatively slower runtime.

[0119] Finally, it should be noted that the above examples are merely some specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments and many variations are possible. All variations that can be directly derived or conceived by those skilled in the art from the disclosure of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for laser extraction of the center line of a shoe outsole surface, characterized in that... The process includes the following: S1. Acquire the original image of the shoe sole, and then extract the green channel from the original image in the computer to obtain image I. gray Then, for image I gray Gaussian convolution operation is performed to obtain the preprocessed image. r is the number of rows in the image, and c is the number of columns in the image; S2, Calculate the preprocessed image The quantiles in the j-th column, j = {1, 2, 3, ..., c}: Computational preprocessing of images The set D of pixel sorting distribution in column j (j = {1, 2, 3, ..., c}) n The calculation method is as follows: In the formula, I (i,j) Let D be the gray value at position (i,j) in the image. n Given a set of data arranged in ascending order, containing a total of n data items; Calculate D n The equal division positions in the middle, the equal division positions q i The calculation method is as follows: In the formula, m represents the number of equal parts, and q represents the number of equal parts. i Let n be the i-th equidistant position, where i ∈ [1, m-1]; n is the position of D. n The number of data points in the middle; Calculate set D using the equally divided positions n m-quantile in: In the formula, Q is the m-quantile m-1 The i-th quantile in the array has a total of m-1 quantiles, int(q i ) is q i The integer part, float(q) i ) is q i The decimal part, D n in int(q i )Location; S3. If the repeatability of the quantile in column j is 0, return to step S2 to calculate the quantile in column (j+1). If the repeatability is 1, proceed to step S4. S4. Determine whether the repeated position of the quantile in column j is in the middle of the set of quantiles in column j. If the repeated position is not in the middle, return to step S2 to calculate the quantile in column (j+1). If the repeated position is in the middle, proceed to step S5. S5. Calculate the midpoint of the laser in column j using the GGM algorithm; S6. Determine whether to iterate through the preprocessed image. If all columns have not been traversed, return to step S2 to calculate the quantile of column (j+1). If they have been traversed, connect the laser midpoints of each column obtained in step S5 to form the laser midline of the shoe outsole.

2. The method for laser-based extraction of the center line of a shoe outsole surface according to claim 1, characterized in that: The formula for extracting the green channel in step S1 is: Among them, I gray The image after extracting the green channel is I(G), where I(R) is the pixel value of the G channel of the RGB image, I(R) is the pixel value of the R channel of the RGB image, I(B) is the pixel value of the B channel of the RGB image, and T0 is the grayscale threshold. The preprocessed image Where g(x,y) is the Gaussian convolution kernel: Wherein, the size of σ should satisfy D is the width of the laser stripe.

3. The method for laser-based extraction of the center line of a shoe outsole surface according to claim 2, characterized in that: The repeatability mentioned in step S3 is: Among them, T R0 The repeat threshold is used; repeat time is the number of repetitions, when two adjacent quantiles are equal. When the number of repetitions is 1, the number of repetitions is incremented by 1.

4. The method for laser centerline extraction of shoe outsole surface lines according to claim 3, characterized in that: The GGM algorithm calculates the coordinates of the laser center point in the j-th column. In the formula: R max R is the maximum value among the repeated quantiles. min It is the minimum value among the repeated quantiles; (u,v) represents the pixel position in the image, specifically the position in the u-th row and v-th column, where w is the laser linewidth.

Images