An Online Skin Boundary Feature Point Extraction Method Based on Line Structured Light

Abstract

The present invention provides an online method for extracting feature points of the skin boundary based on line structured light. This method includes: 1) extraction and reconstruction of the center line of the laser stripe; 2) preprocessing and dimensionality reduction of the reconstructed point cloud; 3) obtaining the control vertices of polygon approximation; 4) extracting precise boundary feature points by directed rotation angles, and realizing the extraction of the boundary points of the box skin through the processing of the scanned line point cloud. The method of the present invention can accurately extract the boundary points in the reconstructed point cloud under various cross-sectional shapes and multiple scanning postures in an online manner.

Description

Technical Field

[0001] The present invention relates to the field of line structured light vision measurement, and specifically to an online skin boundary feature point extraction method based on line structured light. Background Art

[0002] There are strict manufacturing and fitting requirements for the skin of the emission box and the box body. In traditional skin processing, the boundary to be processed is obtained by manual scribing. This method has low efficiency, poor accuracy, and is greatly affected by the subjectivity of workers, and the production quality is difficult to meet the airtightness requirements of the box body. In addition, the production department has also tried digital detection means such as laser scanning and structured light scanning. These methods all use scanning equipment to obtain the overall point cloud data of the box body, and then cooperate with the boundary line extraction algorithm for large-scale point cloud data to obtain the boundary line. Such a processing method has a large amount of point cloud data and high algorithm complexity, and cannot achieve real-time boundary line extraction.

[0003] Line structured light measurement is a digital measurement method with fast reconstruction speed, high accuracy, and strong anti-noise ability. It projects laser stripes onto the surface of the object to be measured, refines the laser stripes to the sub-pixel level, and reconstructs the three-dimensional shape of the object to be measured according to the triangulation method. The reconstructed point cloud is arranged in an orderly manner on the light plane. Summary of the Invention

[0004] The purpose of the present invention is to propose an online skin boundary feature point extraction method based on line structured light in view of the shortcomings of the existing box skin boundary extraction. This method can effectively achieve high-precision real-time extraction for boxes with various cross-sectional shapes.

[0005] To achieve the above purpose, the technical solution of the present invention is as follows:

[0006] An online skin boundary feature point extraction method based on line structured light, comprising the following steps:

[0007] S1. Combine single binocular and handheld line structured light measurement to reconstruct the laser line on the surface of the highly reflective box body, and obtain the three-dimensional point cloud data of the three-dimensional box body surface;

[0008] S2. Reduce the dimension of the scanned line point cloud and perform data preprocessing;

[0009] S3. Determine the boundary point feature points in the scanned line point cloud by a two-step method;

[0010] S4. Determine the skin cross-section type according to the distribution of the boundary points and optimize the boundary feature points;

[0011] Further, in step S1, the specific process of combining single binocular and handheld line structured light measurement to reconstruct the laser line on the surface of the highly reflective box body and obtain the three-dimensional point cloud data of the three-dimensional box body surface is as follows:

[0012] Two sets of laser line structured light reconstruction systems are used to collect images respectively. The problem of high reflectivity is solved by the homography relationship between the light strips in the left and right views. The center of the light strip is extracted by the gray square weighted centroid method. The central point set of the stripes is smoothed by five-point moving average, and finally the three-dimensional reconstruction of the stripes is completed according to the laser triangulation method.

[0013] Further, the specific process of the scan line point cloud dimensionality reduction and data preprocessing in step S2 is as follows:

[0014] S21. Use Rodriguez rotation to rotate the reconstructed three-dimensional data to a certain plane parallel to the plane. At this time, the axis coordinates are the same, that is, the dimensionality reduction of the three-dimensional data is completed;

[0015] S22. Use the chord height difference method to remove jump points and error points;

[0016] S23. Set the radius , determine all points within the radius of the point to be filtered and calculate their distances from the point to be filtered. Normalize the distances, and then substitute them into the Gaussian kernel function to calculate the Gaussian weights, and obtain the filtered points by weighted average.

[0017] Further, the specific process of the two-step method for determining the boundary point feature points in the scan line point cloud in step S3 is as follows:

[0018] S31. Use Douglas–Peucker polygon approximation to determine the control vertices on the scan line point cloud, and set the radius as the search area for boundary points;

[0019] S32. Construct the turning support domain of all points within the search area;

[0020] S33. Calculate the k-directed turning angles of each point, determine the positive maximum value and negative minimum value of all points in the search area, and determine the correct boundary points according to the turning angle and local same-sign uniqueness;

[0021] S34. Determine the box type according to the number of boundary points and the distance between adjacent boundary points, and then determine whether it is necessary to fit the boundary to obtain more accurate boundary points according to the number difference.

[0022] Further, the construction of the turning support domain of all points within the search area is to linearly interpolate the coordinates of the new points according to the given cumulative chord length sum and the new point step size . The specific process is as follows:

[0023] Establish a sequence number for the scan line data points in the original storage order. For the current point , at the given chord length sum Inside, first determine the farthest points in the left and right neighborhoods and such that it satisfies

[0024] ,

[0025] wherein is the cumulative chord length sum from point to . Then, according to the set step size , starting from point , calculate the new point in accordance with the step size. Suppose it falls into the interval . Then, determine the new point by linear interpolation between the two endpoints of the interval. Taking the right side of as an example, after the rd step size, the point is within . The cumulative chord length sums corresponding to the points to and are respectively and . , are respectively the proportionality coefficients. Then, the new point satisfies ;

[0026] wherein , .

[0027] Furthermore, the k-directed rotation angle is a directed feature quantity capable of expressing the change trend, and its definition is:

[0028] For a point , the left and right equal chord length distance points are respectively , . The vectors between them and are , . The directed vector angular value between the two vectors is ;

[0029]

[0030]

[0031]

[0032] When satisfies , , , then the point has a directed rotation angle of ;

[0033] When and are located on both sides of and collinear, at this time , has no change in the neighborhood and can be determined as a non-feature point. When and theoretically coincide, , which is a cusp point with obvious feature information. The larger the

[0034] value, the greater the possibility of it becoming a feature point. Therefore, it can be used as an attribute to characterize the discrete point as a boundary feature point.

[0035] 1) If the sequence difference between adjacent extreme points with different signs is less than 5 points and the Euclidean distance between the two points is greater than 1 mm and less than 5 mm, then it is a "step type" with missing data, types 1 and 2;

[0036] 2) If the sequence difference is greater than 5 points, it can be considered that the point cloud of the vertical edge is reconstructed between the upper and lower boundary points, which is type 3. Due to factors such as reconstruction resolution and filtering, the actual boundary points may not be exactly reconstructed and there is a contraction phenomenon at the cusp part. It needs to be further determined by the known points on both sides. The points on both sides of the upper boundary point can be respectively fitted with straight lines to find the intersection point as the final boundary point;

[0037] 3) If there is only one boundary point between adjacent extreme points with different signs or the distance between the two is greater than 5 mm, it can be judged as a "ramp type", and the intersection point of the point sets on both sides of the point is also fitted with a straight line as the final boundary point;

[0038] 4) There will be 2 upper boundary points in the frustum multi-step. The processing method is similar to the step type, and the second one is taken as the final boundary point.

[0039] The present invention has the following advantages and effects compared with the prior art:

[0040] (1) The present invention extracts the box boundary line in two steps, approximately determines the rough positioning area by polygon, and determines the precise positioning point by the directed k rotation angle, which can realize the precise extraction of boundary points;

[0041] (2) The present invention uses line structured light to extract while scanning, avoiding the reconstruction of a large number of data point clouds and meeting the real-time extraction requirements.

[0042] The gray square weighted method is used to extract the center line of the light strip on the surface of the box body, which has higher robustness and accuracy and is suitable for the three-dimensional reconstruction of the metal surface. Brief Description of the Drawings

[0043] Figure 1 It is a flow chart of an online skin boundary feature point extraction method based on line structured light according to the present invention;

[0044] Figure 2 It is a schematic diagram of the boundary of the line structured light measurement according to the present invention;

[0045] Figure 3 It is a schematic diagram of the precise positioning of the boundary points according to the present invention;

[0046] Figure 4 It is a result diagram of the extraction of the stepped boundary points according to the present invention;

[0047] Figure 5 It is a result diagram of the extraction of the multi-level stepped boundary points according to the present invention;

[0048] Figure 6 It is a result diagram of the extraction of the ramp-shaped boundary points according to the present invention;

[0049] Figure 7 It is a flow chart of the boundary extraction according to the present invention;

[0050] Figure 8 It is a schematic diagram of the directed corner according to the present invention. Detailed Embodiment

[0051] In order to enable those skilled in the art to better understand the technical solutions in this application, the following will clearly and completely describe the technical solutions in the embodiments of this application with reference to the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of this application.

[0052] Although the steps in the present invention are arranged with reference numerals, they are not used to limit the order of the steps. Unless the order of the steps is clearly stated or the execution of a certain step requires other steps as a basis, the relative order of the steps can be adjusted. It can be understood that the term "and / or" used herein relates to and encompasses any and all possible combinations of one or more of the associated listed items.

[0053] The following gives an exemplary description of an online skin boundary feature point extraction method based on line structured light according to the present invention.

[0054] S1. The line structured light measurement system projects laser light onto the surface of the box-shaped object. The light stripe image is preprocessed through median filtering and closing operation. By combining the OTSU threshold based on edge grayscale with the global threshold, the light stripe is segmented. Search for the boundaries of the light stripe for each row, and calculate the sub-pixel center points of the light stripe through the weighted grayscale centroid method:

[0055]

[0056] In the formula: where is the coordinate of the sub-pixel stripe center point obtained by extraction, is the pixel coordinate and the grayscale value at this point.

[0057] Three-dimensional reconstruction is completed through the laser triangulation method.

[0058] S2. Use the Rodriguez rotation to rotate the reconstructed three-dimensional data to a certain plane parallel to the plane. At this time, the axis coordinates are the same, that is, the three-dimensional data dimensionality reduction is completed.

[0059] S3. Use the chord height difference method to eliminate jump points and error points

[0060] S4. Set the radius , calculate all points within the radius of the point to be filtered, perform distance normalization, and then substitute it into the Gaussian kernel function to calculate the weight value. The neighborhood points calculate the filtered points through the weight value.

[0061] S5. Determine the control vertices through polygon approximation and use them as seed points to determine the set of points to be searched.

[0062] S6. Determine the neighborhood support domain of each point in the set through the cumulative chord length sum and the new step size, and calculate the k-directed rotation angle of each point.

[0063] S7. Calculate all the positive maxima and negative minima within all sets.

[0064] S8. Delete the extreme points of non-boundary points through the rotation angle threshold and the uniqueness of the same-direction extreme values.

[0065] Rotation angle threshold: In the actual detection object, the boundary point angle value needs to be greater than 40°.

[0066] Same-sign local uniqueness: There will also be extreme points with an angle greater than the set value in the neighborhood points of the maximum and minimum values. There will only be one extreme point with the same sign within the given Euclidean distance of the boundary point. Therefore, the extreme points of the local positive maximum and negative minimum are used as boundary points by sign, and the extreme points with the same sign within their neighborhood range are discarded.

[0067] S9. Determine the box type based on the number of boundary points and the distance between adjacent boundary points, and then determine whether it is necessary to fit the boundary to obtain more accurate boundary points according to the number difference.

[0068] If the difference in the sequence where adjacent extreme points with different signs is less than 5 points and the Euclidean distance between the two points is greater than 1 mm and less than 5 mm, then it is a "step type" with missing data, types 1 and 2.

[0069] If the difference in the sequence is greater than 5 points, it can be considered that the point cloud of the vertical edge is reconstructed between the upper and lower boundary points, which is type 3. Due to factors such as reconstruction resolution and filtering, the actual boundary points may not be exactly reconstructed and there is a contraction phenomenon at the cusp part, which needs to be further determined by the known points on both sides. The points on both sides of the upper boundary point can be respectively fitted with straight lines to find the intersection point as the final boundary point.

[0070] If there is only one boundary point for adjacent extreme points with different signs or the distance between the two is greater than 5 mm, it can be judged as a "ramp type", and the point sets on both sides of the point are also fitted with straight lines to find the intersection point as the final boundary point.

[0071] S10. Rotate the two-dimensional boundary points back to the original coordinate system by Rodriguez rotation again.

[0072] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

[0073] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An online skin boundary feature point extraction method based on line structured light, characterized in that It includes the following steps: S1. A hand-held single binocular combined line structured light measurement system reconstructs the laser line on the surface of the highly reflective skin to obtain the three-dimensional point cloud data of the box surface; S2. Scan line point cloud dimensionality reduction and data preprocessing; S3. Determine the boundary feature points in the scan line point cloud by a two-step method; S4. Determine the skin cross-section type according to the distribution of boundary points and optimize the boundary feature points; The two-step method in step S3 to determine the boundary point feature points in the scan line point cloud includes the following steps: S31. Use Douglas–Peucker polygon approximation to determine the control vertices on the scan line point cloud, and set the radius as the search area for boundary points according to the control vertices;​​ S32. Construct the corner support domain of all points in the search area; S33. Calculate the k-directed corner of each point, determine the positive maximum value and negative minimum value of all points in the search area, and determine the correct boundary points according to the corner threshold and local same-sign uniqueness.

2. The on-line skin boundary feature point extraction method based on line structured light according to claim 1, wherein In step S1, the single binocular combined hand-held line structured light measurement reconstructs the laser line on the surface of the highly reflective box to obtain the three-dimensional point cloud data of the three-dimensional box surface. The specific process is as follows: Two sets of laser line structured light reconstruction systems are used to collect images respectively. The problem of high reflectivity is solved through the homography relationship between the light strips in the left and right views, and the center of the light strip is extracted by the gray square weighted centroid method. The point set smoothing is completed for the stripe center point set by the five-point moving average method. Finally, the stripe three-dimensional reconstruction is completed according to the laser triangulation method.

3. An online skin boundary feature point extraction method based on line structured light according to claim 1, characterized in that, The scan line point cloud dimensionality reduction and data preprocessing in step S2 include the following steps: S21. Use Rodriguez rotation to rotate the reconstructed three-dimensional data to a certain plane parallel to the plane. At this time, the z-axis coordinates are the same, which completes the dimensionality reduction of the three-dimensional data. S22. Use the chord height difference method to remove the jump points and error points; S23. Set the radius , determine the radius of the point to be filtered for all points within it and calculate their distances from the point to be filtered, normalize the distances, then substitute them into the Gaussian kernel function to calculate the Gaussian weights, and obtain the filtered point through weighted averaging.

4. An online skin boundary feature point extraction method based on line structured light according to claim 1, characterized in that The turning support regions of all points within the constructed search region are determined by linear interpolation according to the given cumulative chord length and and the new point step size to determine the coordinates of the new point. The specific process is as follows: Establish a sequence number for the scanned line data points in the originally stored order. For the current point , within the given chord length and , first determine the farthest points and in the left and right neighborhoods to satisfy ， ； where is the cumulative chord length sum from point to . Then, according to the set step size , starting from point , calculate the new point in steps. Suppose it falls into the interval . Then, determine the new point by linear interpolation between the two endpoints of the interval; taking the right side of as an example, after the -th step, the point is in . The cumulative chord length sums corresponding to the points to and are and respectively. , are the proportionality coefficients respectively. Then, the new point satisfies ; Among them , .

5. An online skin boundary feature point extraction method based on line structured light according to claim 4, characterized in that, The k-directed corner is a directed feature quantity that can express the change trend, and its definition is: One point The points with equal chord lengths on the left and right sides are respectively , , and the vectors between them and are , , and the directed vector angular value between the two vectors is ; ； When is satisfied then , , then the directed angle of point is ; where is a specific value in , that is, the directed vector angular value between the relevant vectors of point When and are collinear and located on both sides of , at this time , there is no change in the neighborhood, and it can be determined as a non-feature point. When and theoretically coincide , it is a cusp point with obvious feature information . The larger the value, the greater the possibility of it becoming a feature point. Therefore, it can be used as an attribute to represent that the discrete point is a boundary feature point; The absolute value of the included angle between the vectors formed by the points and with equal chord lengths on the left and right of the representative point and the point .

6. The on-line skin boundary feature point extraction method based on line structured light according to claim 1, characterized in that In step S4, determine the skin cross-section type according to the distribution of boundary points and optimize the boundary feature points. The specific judgment is as follows: 1) If the sequence difference between adjacent extreme points with different signs is less than 5 points and the Euclidean distance between the two points is greater than 1 mm and less than 5 mm, then the skin cross-section type is the "step type" with missing data; 2) If the sequence difference is greater than 5 points, it can be considered that the point cloud of the vertical edge is reconstructed between the upper and lower boundary points, which is of type 3. Due to the influence of reconstruction resolution and filtering factors, the actual boundary points may not be exactly reconstructed and there is a contraction phenomenon in the cusp part, which needs to be further determined by the known points on both sides. The points on both sides of the upper boundary point can be respectively fitted with straight lines to find the intersection point as the final boundary point; 3) If there is only one boundary point or two points with a distance greater than 5 mm between adjacent extreme points with different signs, it can be judged that the skin cross-section type is the "ramp type". Similarly, fit the straight lines of the point sets on both sides of the point to find the intersection point as the final boundary point; 4) There will be 2 upper boundary points in the multi-stage step of the frustum. The processing method is similar to the step type, and the second one is taken as the final boundary point.

Images