A method for extracting the center of a line laser stripe based on geometric and grayscale fusion decision.

Abstract

To address the challenge of existing linear laser bar center extraction algorithms failing to achieve sub-pixel-level precision extraction from images of thin-walled, complex curved surfaces of aero-engines with locally broken or anomalous linear laser bars, this invention proposes a geometric and grayscale fusion decision-making method for linear laser bar center extraction. This invention considers the inherent geometrical continuity of the linear laser bars on thin-walled, complex curved surfaces. For each column in the linear laser bar region (excluding the first column), candidate sub-pixel points and predicted center points are obtained through linear laser bar grayscale response calculation and geometrical continuity prediction, respectively. Then, a comprehensive score of the geometrical continuity and grayscale response of the candidate sub-pixel points is calculated based on an adaptive weighting factor. Finally, the linear laser bar center point is determined based on the comprehensive score and a preset score threshold, thus solving the technical difficulty of existing linear laser bar center extraction methods in handling locally broken or anomalous linear laser bars.

Description

Technical Field

[0001] This invention belongs to the field of image data processing, and specifically relates to a method for extracting the center of a line laser stripe. Background Technology

[0002] The high-quality manufacturing of key components for aero-engines largely relies on in-flight measurement technology to accurately acquire surface machining error data, thereby achieving machining quality control. Line laser sensors are widely used due to their advantages such as flexibility, efficiency, and low cost. Key components of aero-engines are typically designed as thin-walled, complex curved surface structures and manufactured using difficult-to-machine materials such as titanium alloys or high-temperature alloys, including blades, integral bladed disks, centrifugal impellers, and high-temperature alloy sealing rings. During in-flight laser measurement, due to the complex geometry and high reflectivity of the surface, localized broken or abnormal light streaks are easily observed in the light streak image, severely affecting the final point cloud accuracy. Therefore, accurately extracting the center of the light streak in such images is of great significance for achieving precise measurement of thin-walled, complex curved surface components of aero-engines.

[0003] Currently, light stripe center extraction algorithms mainly fall into four categories: those based on grayscale features, those based on Hessian matrix, those based on morphology, and those based on deep learning.

[0004] Methods based on grayscale features typically rely on the local or global grayscale distribution patterns of light stripe images to locate the center, such as the extreme value method, threshold method, grayscale centroid method, directional template method, and Gaussian template method. These methods usually have advantages such as simple structure, low computational cost, and strong real-time performance, but they are more sensitive to non-uniform light stripes, are prone to center point shift, and have low extraction accuracy.

[0005] Compared to the methods mentioned above, the Hessian matrix-based method no longer relies on the symmetry of the gray-level distribution of the light stripe. Instead, it achieves sub-pixel localization of the light stripe ridge by analyzing differential features, making it the most accurate method currently available. To achieve stable calculation of the second-order differential of the Hessian matrix, such methods typically incorporate Gaussian filtering for preprocessing. Among these, the Steger method and its variants are the most widely used. However, in the extraction of the light stripe center of thin-walled, complex curved surface parts, this method has the following limitations: In the local fracture region of the light stripe, the gray-level distribution is difficult to satisfy the eigenvalue conditions of the ridge, leading to incorrect judgment of the normal direction and thus misjudgment of the center point, resulting in low extraction accuracy; In abnormal light stripe regions, interfering gray-level features are easily included in the calculation of the Hessian matrix eigenvector of the local neighborhood, causing the center point to shift, resulting in low extraction accuracy.

[0006] Morphological methods typically rely on binary processing and morphological operations to maintain the continuity of light stripes, such as the morphological skeleton method. This method can achieve connection in locally broken areas of the light stripe, but it is prone to generating false features, resulting in low accuracy in extracting the connection region.

[0007] In addition, deep learning-based methods are gradually being explored. These methods can learn light stripe features from a large number of images and achieve end-to-end center extraction, showing potential in low-quality light stripe images. However, their performance heavily relies on large-scale, high-quality labeled data, which is often scarce in the measurement scenarios of thin-walled, complex curved surface parts of aero-engines, limiting their practical application.

[0008] In summary, existing methods for extracting the center of light stripes are insufficient to achieve sub-pixel-level accurate extraction of the center of light stripes in laser light stripe images of thin-walled complex curved surface parts of aero-engines with localized broken light stripes and abnormal light stripes, necessitating further research. Summary of the Invention

[0009] To address the technical challenge of existing light stripe center extraction algorithms failing to achieve sub-pixel-level accurate extraction from line laser light stripe images of thin-walled, complex curved surfaces of aero-engines with locally broken or anomalous light stripes, this invention proposes a line laser light stripe center extraction method based on geometric and grayscale fusion decision-making.

[0010] The technical solution of this invention is:

[0011] A method for extracting the center of a line laser stripe based on geometric and grayscale fusion decisions, characterized by the following steps:

[0012] Extracting the light stripe region from a grayscale image of a line laser light stripe;

[0013] Coarse extraction of the center of the light stripe within the light stripe region:

[0014] The response value of each column of pixels within the light stripe region is obtained using the Gaussian template matching method, and the response curve is then fitted.

[0015] Find local maxima on the response curves of each column and use them as candidate pixels for the center of the light stripe;

[0016] For each candidate pixel, a quadratic term is fitted based on its own row position and response value, and the row position and response value of its adjacent pixels in the column direction. The extreme points and their corresponding response values ​​are calculated as candidate sub-pixels and their corresponding response values.

[0017] Coarse extraction of the center point of the first column of light stripes: Extract the candidate sub-pixel with the largest response value from the candidate sub-pixel points located on the first column of light stripes and use it as the center point of the first column of light stripes, and store it in the first light stripe center point set;

[0018] Coarse extraction of light bar center points for subsequent columns: Predicted center points for subsequent columns are extracted through geometric continuity prediction; the comprehensive score of geometric continuity and grayscale response of candidate sub-pixel points in each column is calculated through adaptive weighted summation; the light bar center point is decided based on the comprehensive score and a preset score threshold. If a column has a set of candidate sub-pixel points with a score greater than the preset score threshold, the point with the highest comprehensive score in the candidate sub-pixel point set is taken as the light bar center point of that column and stored in the first light bar center point set; otherwise, the predicted center point of that column is taken as its light bar center point and stored in the first light bar center point set.

[0019] The precise search area for the light stripe is determined based on the set of center points of the first light stripe.

[0020] Within the precise search area of ​​the light stripe, the center point of the light stripe is extracted to obtain the second set of light stripe center points.

[0021] Furthermore, the specific method for binarizing the grayscale image of the line laser stripe is as follows:

[0022] The segmentation thresholds I for the foreground and background of a grayscale image of a line laser stripe are obtained using the maximum inter-class variance method. Th ;

[0023] Based on the segmentation threshold I Th and preset regulatory factors Th The grayscale image of the line laser stripe is binarized to obtain a binary image; Th .

[0024] Furthermore, the method for extracting the light stripe region from the grayscale image of the line laser light stripe is as follows: after binarizing the grayscale image of the line laser light stripe, the start and end columns of the light stripe are located in the binarized image, and the light stripe region is extracted from the grayscale image of the line laser light stripe based on the located start and end columns of the light stripe.

[0025] Furthermore, the formula for calculating the comprehensive score is as follows:

[0026]

[0027] In the formula, As an adaptive weighting factor, ; For the first The maximum response of candidate subpixels in the column, R GM The global maximum response for all candidate subpixels;

[0028] For located OK, Candidate sub-pixels of the column Geometric continuity score, ; It is a geometric continuity constraint quantity, and its value ranges from 0.01 to 0.1 of the total number of rows of pixels in the grayscale image of the line laser light stripe;

[0029] For located OK, Candidate sub-pixels of the column The grayscale response score, ; For located OK, Candidate sub-pixels of the column The grayscale response value.

[0030] Furthermore, the method for extracting the prediction center points of subsequent columns through geometric continuity prediction is as follows: the prediction center point of the second column is obtained based on the idea of ​​the nearest point, the prediction center point of the third column is obtained through linear extrapolation, the prediction center point of the fourth column is obtained through quadratic polynomial extrapolation, and the prediction center points of the fifth column and subsequent columns are obtained through spline curve fitting.

[0031] Furthermore, the method for determining the precise search region of the light stripe based on the set of center points of the first light stripe is as follows:

[0032] Based on the row containing the center points of each light bar in the first light bar center point set, several rows are selected above and below to define the search range;

[0033] Binarize the search range;

[0034] The actual width of each column of light stripes within the binarized search range is calculated by one-dimensional connected component analysis, thus obtaining the precise search area for the light stripes.

[0035] Furthermore, the Steger method is used to precisely extract the center point of the light stripe within the accurate search area.

[0036] The beneficial effects of this invention are as follows:

[0037] 1. In the coarse extraction stage of the light stripe center, this invention establishes a light stripe center decision mechanism that integrates geometric continuity and grayscale response.

[0038] This invention takes into account the inherent geometric continuity of light stripes on thin-walled complex curved surface parts. For each column in the light stripe region except the first column, candidate sub-pixel points and predicted center points of the light stripe center are obtained by calculating the light stripe grayscale response and predicting geometric continuity, respectively. Then, the comprehensive score of the geometric continuity and grayscale response of the candidate sub-pixel points is calculated based on the adaptive weighting factor. Finally, the light stripe center point is determined based on the comprehensive score and the preset score threshold. This invention solves the technical problem that existing light stripe center extraction methods are difficult to handle locally broken light stripes or abnormal light stripes.

[0039] 2. This invention establishes a framework for layered extraction of light stripe centers.

[0040] The framework mainly consists of three core modules: coarse extraction of the light stripe center, fine localization of the light stripe region, and fine extraction of the light stripe center. After obtaining the coarse extraction result of the light stripe center through a decision mechanism that fuses geometric continuity and grayscale response, one-dimensional connected component analysis is used to determine the actual width of the light stripe and accurately locate the precise search area for the light stripe. Based on this, the fine extraction of the light stripe center of thin-walled complex curved surface parts of aero-engines is performed.

[0041] 3. This invention overcomes the technical challenge of existing light stripe center extraction methods, which struggle to handle light stripe images with locally broken or abnormal light stripes due to their failure to consider the geometric continuity characteristics of thin-walled, complex curved surface parts. The technical difficulty of this invention lies in balancing the influence weights of the geometric continuity characteristics of the measured part's contour and pixel grayscale features on light stripe center extraction, and in achieving accurate extraction of the light stripe center within a layered extraction framework. To address these challenges, this invention solves them through the aforementioned technological innovations. For the former, this invention innovatively introduces a comprehensive score based on adaptive weighting factors for geometric continuity and grayscale response, resolving the decision-making difficulty between candidate sub-pixel points (grayscale response) and predicted center points (geometric continuity) for the light stripe center. For the latter, this invention proposes a layered extraction processing framework, using the light stripe center extraction result based on the fusion of geometric continuity and grayscale response as a coarse extraction result, employing one-dimensional connected component analysis to accurately locate the precise search region for the light stripe, and then performing fine extraction of the light stripe center within this precise search region. Finally, practical technical results demonstrate the advantages of this invention compared to the seven existing methods.

[0042] 4. The present invention uses the Steger method for fine extraction at the center of the light stripe, which further improves the accuracy of fine extraction at the center of the light stripe. Attached Figure Description

[0043] Figure 1 This is a flowchart of the method of the present invention.

[0044] Figure 2 This is a schematic diagram of a one-dimensional connected region in this invention.

[0045] Figure 3 The output results of this invention are compared with the output results of existing light stripe center extraction methods. Detailed Implementation

[0046] The present invention will be further explained below with reference to the accompanying drawings, taking the extraction of the center of the line laser light stripe grayscale image of the air intake edge of the compressor blade in an aero-engine as an example.

[0047] Define the grayscale image of the line laser stripe acquired during compressor blade inlet edge measurement as: Where I(m, n) represents the gray value of the pixel in the m-th row and n-th column of the line laser bar grayscale image I, m represents the row index, n represents the column index, and M and N represent the number of pixels in the height and width directions of the image, respectively.

[0048] Reference Figure 1 The method for extracting the center of a line laser stripe based on geometric and grayscale fusion decision proposed in this invention specifically includes the following steps:

[0049] Step 1: Obtain the segmentation threshold between the foreground and background of the image.

[0050] The segmentation thresholds I for the foreground and background of the grayscale image I of a line laser stripe are obtained by the maximum inter-class variance method. Th .

[0051] Step 2: Image binarization.

[0052] Based on the segmentation threshold I of foreground and background Th and regulatory factors Th The grayscale image I of the line laser stripe is binarized to obtain a binary image. :

[0053] When the grayscale value I(m, n) of the pixel in the m-th row and n-th column of the line laser stripe grayscale image I is greater than or equal to When the pixel value is set to 1, the corresponding pixel value in the binary image is set to 1. BI (m, n) is set to 1;

[0054] When the grayscale value I(m, n) of the pixel in the m-th row and n-th column of the grayscale image I is less than When the pixel value is set to 0, the corresponding pixel value in the binary image is set to 0, i.e., I... BI (m, n) is set to 0.

[0055] Among them, regulatory factors Th The values ​​are selected within the range of [0, 2] to suppress errors in the calculation of foreground and background segmentation thresholds for the Otsu's method in the grayscale image I of the line laser stripe on the inlet edge of an aero-engine compressor blade. In this embodiment... Th Take 0.4.

[0056] Step 3: Locate the start and end columns of the light stripe and extract the light stripe area.

[0057] For binary image I BI Sum the pixel values ​​column by column:

[0058]

[0059] In the formula, I BI_S (n) represents the sum of pixels in the nth column of the binary image.

[0060] I BI_S (n) The non-zero starting and ending columns are the starting and ending columns of the line laser bar grayscale image I, and the starting and ending column indices are denoted as n. E1 , n E2 The light stripe region extracted from the grayscale image I based on the start column index and the end column index is:

[0061]

[0062] Step 4: Coarse extraction of the center of the light stripe.

[0063] Step 4.1: Use the Gaussian template matching method to obtain the light stripe region I. ROI1 The response curve for each column;

[0064] In the light stripe region I ROI1 Within the array, a Gaussian template is used to perform a convolution operation on each column of pixels to obtain the grayscale response value of each pixel:

[0065]

[0066]

[0067]

[0068] In the formula, L represents the grayscale response value. Ts For Gaussian template size (odd number), GF Te_Norm For the normalized Gaussian template, GF Te Use a Gaussian template. is the standard deviation of the Gaussian template, and i and j represent the position indices in the Gaussian template.

[0069] After convolution processing, the response values ​​of each column of pixels are fitted, and then the light stripe region I... ROI1 Each column corresponds to a response curve. In this embodiment, L Ts The value is 17. The value is 3.

[0070] Step 4.2: Obtain the set of candidate pixels at the center of the light stripe;

[0071] In the light stripe region I ROI1On the response curve corresponding to each column of pixels, find the local maximum point. This local maximum point is the candidate pixel point for the center of the light stripe. All candidate pixels for the center of the light stripe constitute the set of candidate pixels for the center of the light stripe.

[0072] The set of candidate pixels at the center of the acquired light stripe is denoted as... , This indicates the position of the grayscale image I containing the line laser stripe. OK, The grayscale value of the candidate pixel at the center of the column of light stripes, and the corresponding response value are represented as: , , .

[0073] In the light stripe region I ROI1 Each column of pixels has at least one candidate pixel at the center of the light stripe.

[0074] Step 4.3: Extract candidate sub-pixel points at the center of the light stripe;

[0075] For the set of candidate pixels at the center of the light stripe Each candidate pixel in the array is determined based on its row position and response value. The row position and response value of adjacent pixels in the column direction Perform a quadratic fitting, and the quadratic fitting equation is:

[0076]

[0077] in, , , These are the fitting coefficients for the quadratic term;

[0078] After obtaining the quadratic fitting coefficients, the extreme points and corresponding response values ​​are calculated based on the quadratic fitting formula. The calculated extreme points are the extracted candidate sub-pixel points, and the row index of the extreme points is the row index of the candidate sub-pixel points.

[0079]

[0080] The set of candidate sub-pixel points is represented as , , , This indicates the position of the grayscale image I containing the line laser stripe. OK, The grayscale value of the candidate sub-pixel at the center of the column of light stripes, and the corresponding response value are represented as follows: , , .

[0081] Step 4.4: Coarse extraction of light stripe center points based on geometric and grayscale fusion decisions;

[0082] Step 4.4.1: Coarse extraction of the center point of the first column of light stripes;

[0083] For the light stripe region I ROI1 The set of candidate sub-pixels within When the column index of the candidate subpixel satisfies When the value is displayed, it indicates that it is the first candidate sub-pixel. Among the first candidate sub-pixels, the candidate sub-pixel with the largest response value is the extracted center point of the first column of light stripes.

[0084] The position of the center point of the extracted light stripe is uniformly recorded as... The response value is denoted as This information is then stored in the historical light bar center point queue and the first light bar center point set. For the position of the center point of the first light bar, .

[0085] Step 4.4.2: Coarse extraction of the center point of the subsequent light stripes;

[0086] Step 4.4.2.1: Extract the prediction center points of subsequent columns through geometric continuity prediction;

[0087] For the light stripe region I ROI1 The set of candidate sub-pixels within ,when hour:

[0088] When the number of points in the historical light bar center point queue is 1, based on the nearest point idea, the candidate sub-pixel point in the current column (second column) that is closest to the position of the previous historical light bar center point is taken as the predicted center point of the current column;

[0089] When the number of points in the historical light bar center point queue is 2, the predicted center point of the current column (the third column) is obtained by linear extrapolation;

[0090] When the number of points in the historical light bar center point queue is 3, the predicted center point of the current column (fourth column) is obtained by extrapolation using a quadratic polynomial.

[0091] When the number of points in the historical light bar center point queue is greater than or equal to 4, the predicted center point of the current column (the fifth column to the last column) is obtained by spline curve fitting;

[0092] The positions of the predicted center points of each column extracted above are denoted as... The corresponding response value is denoted as ,in, , .

[0093] Step 4.4.2.2: Calculate the light stripe region IROI1 Adaptive weighting factors for each column (excluding the first column) :

[0094]

[0095] In the formula, For the first The maximum response of candidate subpixels in the column, R GM The global maximum response for all candidate sub-pixels. Maximum response of the column When it is strong, that is Approaching R GM hour, Approximately 0.3; the Maximum response of the column When it is weaker, that is Much smaller than R GM hour, It is close to 0.8.

[0096] Step 4.4.2.3: Calculate the light stripe region I ROI1 Candidate sub-pixels in each column (excluding the first column) Geometric continuity score:

[0097]

[0098] In the formula, This is a geometric continuity constraint used to avoid erroneously extracting the center point of non-contour light stripes (abnormal light stripes). Its value ranges from 0.01 to 0.1 of the total number of rows of pixels in the grayscale image of the line laser light stripe. The closer a candidate sub-pixel is to the center point of the light stripe extracted in the previous step, the higher its geometric continuity score. In this embodiment, Take 0.05 of the total number of rows in the grayscale image I of the line laser light stripe.

[0099] Step 4.4.2.4: Calculate the light stripe region I ROI1 Candidate sub-pixels in each column (excluding the first column) Grayscale response score:

[0100]

[0101] The higher the grayscale response value of a candidate subpixel, the higher its score.

[0102] Step 4.4.2.5: Calculate the light stripe region I ROI1 (Except for the first column) The combined score of geometric continuity and grayscale response of candidate sub-pixels in each column:

[0103]

[0104] In the formula, the adaptive weighting factor Used to balance geometric continuity score and grayscale response score.

[0105] Step 4.4.2.6: Make a coarse extraction of the center points of the subsequent light stripes by making a decision between the candidate sub-pixel points and the predicted center points, and store the coarsely extracted light stripe center points into the first light stripe center point set;

[0106] From light stripe region I ROI1 (Except for the first column) Select the candidate sub-pixels in each column that have the highest overall score and are greater than the set score threshold CS. Th The points are used as the center points of the corresponding columns of light stripes and stored in the first set of light stripe center points; if the light stripe region I ROI1 There is no column with the highest overall score that is greater than the set score threshold CS. Th If a point is not specified, the corresponding predicted center point in that column is used as its light stripe center point and stored in the first light stripe center point set. In this embodiment, CS Th The value ranges from 0 to 1, and in this embodiment, the value is 0.2.

[0107] Finally, the set of center points of the first light stripe obtained through coarse extraction is represented as:

[0108]

[0109] Step 5: Precise positioning of the light stripe area.

[0110] Step 5.1: Define the search scope;

[0111] The set of center points of the first light stripe obtained by coarse extraction The center point of each light stripe in Define the search area around the center, specifically based on each pixel. The baseline of the behavior is defined by taking k above and below it. FL The search range is defined by the row, denoted as:

[0112]

[0113] In this embodiment, k FL Take 4.

[0114] Step 5.2: Calculate the actual width of the light stripe.

[0115] First, using the same method as steps 1-2, the search range I is... ROI2 Binarization is performed; after binarization, the search range I is... ROI2 Each pixel within the area has a value of 0 or 1.

[0116] Then, the binarized search range I is calculated through one-dimensional connected component analysis.ROI2 The actual width value of each column of light bars within the search range I. ROI2 In the middle, the set of center points of the first light stripe The pixel points contained within the center point of each light stripe and its corresponding actual width constitute the precise search area I for solving the light stripe problem. ROI3 .

[0117] See appendix Figure 2 A one-dimensional connected component is defined as a set of pixels in the current column whose pixel value is 1 (representing white) and which are consecutively adjacent in the column direction. Furthermore, at least one pixel in this set must have a row distance of less than or equal to 2 from the center point of the coarsely extracted light stripe in the current column. If multiple sets of pixels in the current column satisfy the above conditions, the set with the largest number of pixels is selected as the one-dimensional connected component. The width of the one-dimensional connected component is the actual width of the light stripe in its column.

[0118] Step 6: Fine extraction of the center point of the light stripe.

[0119] In the precise search area I of the light stripe ROI3 Within, the Steger method is used to refine the center points of the light stripes, resulting in the second set of light stripe center points, represented as:

[0120]

[0121] in, , . This is the final output of the present invention.

[0122] Technical effectiveness verification:

[0123] Taking the laser light stripe image of the inlet edge of a certain type of compressor blade as an example, the center of the light stripe is extracted using the following methods: extreme value method 1 (pixel level), extreme value method 2 (fitting, subpixel level), thresholding method (binary centroid, subpixel level), grayscale centroid method (subpixel level), Gaussian template method (subpixel level), Steger method (subpixel level), morphological skeleton method (fitting, subpixel level), and the method of this invention (subpixel level). The extraction results are then compared with the true values.

[0124] The extraction results of the above methods can be found in [link to relevant documentation]. Figure 3 Comparing the extraction results of the above methods with the true values ​​of the laser stripe centers in the compressor blade inlet edge image, the average deviations of the above methods are: 1.939 pixels, 1.825 pixels, 1.166 pixels, 1.114 pixels, 0.927 pixels, 0.839 pixels, 1.178 pixels, and 0.723 pixels, respectively. It can be concluded that the method of the present invention has the lowest extraction deviation.

Claims

1. A method for extracting a center of a line laser stripe by fusing geometry and gray scale decision, characterized in that, Includes the following steps: Extracting the light stripe region from a grayscale image of a line laser light stripe; Coarse extraction of the center of the light stripe within the light stripe region: The response value of each column of pixels within the light stripe region is obtained using the Gaussian template matching method, and the response curve is then fitted. Find local maxima on the response curves of each column and use them as candidate pixels for the center of the light stripe; For each candidate pixel, a quadratic term is fitted based on its own row position and response value, and the row position and response value of its adjacent pixels in the column direction. The extreme points and their corresponding response values ​​are calculated as candidate sub-pixels and their corresponding response values. Coarse extraction of the center point of the first column of light stripes: Extract the candidate sub-pixel with the largest response value from the candidate sub-pixel points located on the first column of light stripes and use it as the center point of the first column of light stripes, and store it in the first light stripe center point set; Coarse extraction of the center point of the light stripe in the subsequent column: The predicted center point of the subsequent column is extracted by geometric continuity prediction; The geometric continuity and grayscale response of candidate sub-pixels in each column are calculated by adaptive weighted summation. The center point of the light stripe is determined based on the comprehensive score and a preset score threshold. If there is a set of candidate sub-pixels in a column that is greater than the preset score threshold, the point with the highest comprehensive score in the set of candidate sub-pixels is taken as the center point of the light stripe in that column and stored in the first set of light stripe center points. Otherwise, the predicted center point of the column is taken as its center point of the light stripe and stored in the first set of light stripe center points. The precise search area for the light stripe is determined based on the set of center points of the first light stripe. Within the precise search area of ​​the light stripe, the center point of the light stripe is extracted to obtain the second set of light stripe center points.

2. The method for extracting the center of a line laser stripe based on geometric and grayscale fusion decision-making according to claim 1, characterized in that, The specific method for binarizing a grayscale image of a line laser stripe is as follows: A threshold I of the foreground and the background of the gray-scale image of the line laser light strip is obtained by the maximum inter-class variance method Th ; Based on the segmentation threshold I Th and preset regulatory factors Th The grayscale image of the line laser stripe is binarized to obtain a binary image; Th .

3. The method for extracting the center of a line laser stripe based on geometric and grayscale fusion decision-making according to claim 2, characterized in that, The method for extracting the light bar region from a grayscale image of a line laser light bar is as follows: After binarizing the grayscale image of the line laser light bar, locate the start and end columns of the light bar in the binarized image, and extract the light bar region from the grayscale image of the line laser light bar based on the located start and end columns.

4. The method for extracting the center of a line laser stripe based on geometric and grayscale fusion decision-making according to claim 1, characterized in that, The formula for calculating the overall score is as follows: In the formula, As an adaptive weighting factor, ; For the first The maximum response of candidate subpixels in the column, R GM The global maximum response for all candidate subpixels; For located OK, Candidate sub-pixels of the column Geometric continuity score, ; It is a geometric continuity constraint quantity, and its value ranges from 0.01 to 0.1 of the total number of rows of pixels in the grayscale image of the line laser light stripe; For located OK, Candidate sub-pixels of the column The grayscale response score, ; For located OK, Candidate sub-pixels of the column The grayscale response value.

5. The method for extracting the center of a line laser stripe based on geometric and grayscale fusion decision-making according to claim 1, characterized in that, The method for extracting the prediction center points of subsequent columns through geometric continuity prediction is as follows: the prediction center point of the second column is obtained based on the nearest point idea, the prediction center point of the third column is obtained through linear extrapolation, the prediction center point of the fourth column is obtained through quadratic polynomial extrapolation, and the prediction center points of the fifth column and subsequent columns are obtained through spline curve fitting.

6. The method for extracting the center of a line laser stripe based on geometric and grayscale fusion decision-making according to any one of claims 1-5, characterized in that, The method for determining the precise search region of a light stripe based on the set of center points of the first light stripe is as follows: Based on the row containing the center points of each light bar in the first light bar center point set, several rows are selected above and below to define the search range; Binarize the search range; The actual width of each column of light stripes within the binarized search range is calculated by one-dimensional connected component analysis, thus obtaining the precise search area for the light stripes.

7. The method for extracting the center of a line laser stripe based on geometric and grayscale fusion decision-making according to claim 6, characterized in that, The Steger method was used to precisely extract the center point of the light stripe within the accurate search area.

Images