An Automatic Method for Measuring Metal Thickness Based on 3D Point Cloud Data

Abstract

This invention discloses an automatic metal thickness measurement method based on 3D point cloud data, comprising the following steps: inputting raw point cloud data; extracting point cloud edge features to obtain an edge point data set D; calculating the principal component direction of each point in D based on its neighborhood points as the principal direction of that point; performing Euclidean clustering on D to obtain point sets D_class_0 to D_class_n; for each D_class, first finding an initial corresponding point q based on a point p to form an initial point pair; using the initial point pair, finding the corresponding thickness point pair using a region growing method; calculating the average value of the thickness point pair as the final thickness value; and finally outputting multiple sets of thickness point pairs and their corresponding thickness values. This invention, based on point cloud data reflecting the 3D information of an object's surface, comprehensively utilizes two edge feature extraction methods—normal curvature abrupt change and depth abrupt change—to automatically detect the thickness of scrap steel metal based on extensive and detailed sampling, improving estimation accuracy and generalization.

Description

Technical Field

[0001] This invention relates to the technical field of classifying the measurement of metallic materials using point cloud data processing. Specifically, it relates to a method for automatically measuring the thickness of scrap steel based on three-dimensional point cloud data. Background Technology

[0002] This invention is primarily applied to the automated measurement of certain metal scraps, such as scrap steel. Scrap steel refers to steel waste that is not used as a product during the steelmaking process, as well as steel materials from discarded equipment and components. In recent years, my country's bargaining power over iron ore imports has been significantly insufficient, resulting in high costs, making scrap steel a crucial raw material for steelmaking. For scrap steel grading, thickness is a decisive indicator. In the scrap steel recycling process of major steel mills, most still rely on manual judgment for grading. However, due to differences in perception among employees and the influence of subjective factors such as employee fatigue, the judgment results in the scrap steel recycling process can exhibit deviations and fluctuations.

[0003] With technological advancements, the scrap steel grading process is trending towards automation and intelligence. Some steel mills employ deep learning-based intelligent scrap steel grading systems. These systems use deep learning or machine learning to extract scrap steel samples, train discrimination algorithms, and directly predict the results after deployment. However, these methods have poor generalization ability across different types of scrap steel, low accuracy in identifying scrap steel thickness, and difficulty in accurately determining the classification grade of scrap steel, resulting in low scrap steel recycling efficiency.

[0004] Regarding algorithms for estimating scrap steel thickness, a search revealed Chinese patent application number 202110866985.6, published on December 24, 2021, entitled "Metal Body Thickness Identification Method, System, Storage Medium, and Electronic Device," which discloses the following method: First, multiple imaging units are used to repeatedly scan the scrap steel to obtain a first image and a second image of the metal body cross-section; then, the depth and pixel thickness of the metal cross-section are determined using the obtained images; finally, the thickness of the scrap steel is estimated using the depth, thickness, and imaging unit intrinsic parameters. While this method achieves improved accuracy compared to the aforementioned deep learning methods, it still falls short of the ideal level for practical production, and its hardware deployment is cumbersome, making it unsuitable for large-scale scrap steel thickness estimation.

[0005] Currently, there is no algorithm that can be widely applied, more efficient, and with a recognition accuracy that better meets the actual needs of production for the automated measurement of scrap steel thickness. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, this invention provides an automatic metal thickness measurement method based on three-dimensional point cloud data. This aims to solve the technical problems of low accuracy in scrap steel thickness identification and difficulty in accurately determining the classification grade of scrap steel in existing scrap steel identification schemes, resulting in low scrap steel recycling efficiency.

[0007] The automatic metal thickness measurement method based on three-dimensional point cloud data proposed in this invention includes the following steps:

[0008] Input the raw, ordered point cloud data;

[0009] Extract the edge features of the point cloud to obtain the edge point data set D;

[0010] For each point in point cloud D, calculate its principal component direction based on the points in its neighborhood and use it as the principal direction of that point.

[0011] Perform Euclidean clustering on the point cloud D to obtain point sets D_class_0 to D_class_n;

[0012] For each D_class, we first need to find an initial corresponding point q based on a point p to form an initial point pair;

[0013] Then, based on the initial point pairs, the region growing method is used to find the corresponding thickness point pairs that satisfy a certain estimated number;

[0014] The average value of the thickness point pairs is calculated as the final estimated thickness value.

[0015] Furthermore, the aforementioned edge feature extraction methods include two types: those based on depth abrupt change values ​​and those based on curvature abrupt change values. Both methods can obtain a set of edge points. The extracted edge points can be used for thickness estimation independently or in combination. For ease of representation, the set of edge points extracted based on depth abrupt change values ​​will be referred to as set A, and the set of edge points extracted based on curvature abrupt change values ​​will be referred to as set B.

[0016] Furthermore, the method for finding initial point pairs p and q described above differs slightly when performing thickness estimation independently on datasets A and B and when combining them for thickness estimation. The specific methods are as follows:

[0017] When independently estimating the thickness of scrap steel using datasets A or B, the specific steps are as follows:

[0018] The point set is formed into a "graph structure";

[0019] Find the point q corresponding to point p. Points p and q are edge point data belonging to opposite sides. The Euclidean distance between them can represent the thickness of the scrap steel edge.

[0020] The Euclidean distance between point q and point p is the smallest, and the "graphic distance" between point q and point p is large. The two points form an initial thickness point pair.

[0021] If we combine A and B to estimate the thickness, then we don't need to build a "graph structure". We can directly find the point in B with the minimum Euclidean distance based on the point p in A.

[0022] Furthermore, after finding the initial point pair, the specific details of the above-mentioned region growth method are as follows:

[0023] After finding all initial point pairs, add the initial point pairs to the "seed" set;

[0024] With the initial point pair vector pq as the normal vector, we pass through the plane plane1 passing through point p and the plane plane2 passing through point q;

[0025] Take the point sets w1 and w2 that are closest to plane1 and plane2 respectively;

[0026] Take a point p2 in w1, and construct a point-normal plane plane3 along the direction of point p2;

[0027] Find the set of points in w2 that are within a certain distance from plane3, and find point q2 from them;

[0028] Vector p2q2 satisfies two conditions: the angle between it and vector pq must be less than a preset threshold; and the difference between the thickness it forms and the thickness formed by vector pq must be less than a preset threshold.

[0029] Find all thickness point pairs that use point pair pq as the "seed" point pair, and find new "seed" point pairs from them.

[0030] Repeat the above steps until the seed set is empty.

[0031] Compared with existing technologies, the beneficial effects of this invention are as follows: First, this invention detects the thickness of scrap steel based on point cloud data. Point cloud is a discrete set of points distributed in N-dimensional space, representing discrete sampling of surface information of an object. Compared to RGB data used in many methods, point cloud data can better reflect the spatial shape information of an object. Currently, point cloud data is widely used in the construction of 3D scenes and the digital simulation of various workpieces, but no one has yet applied point cloud data to the grading of scrap steel. Utilizing point cloud data for automatic detection and measurement of scrap steel thickness has significant advantages and prospects. Second, this invention integrates two edge feature extraction methods: abrupt changes in normal curvature and abrupt changes in depth. It not only experiments with their individual estimation of scrap steel thickness but also with their combined estimation. Thus, this invention not only improves the accuracy of thickness detection but also enhances the versatility of the method, enabling it to handle more situations. Finally, the estimation of scrap steel thickness at the point cloud level is essentially an analysis and statistical study of the edge features of the scrap steel after extensive and detailed sampling. Compared to results obtained using relatively coarse methods such as deep learning, the thickness estimation method provided by this invention is more accurate and convincing. Attached Figure Description

[0032] Figure 1 This is a flowchart of an automatic method for measuring the thickness of scrap steel based on three-dimensional point cloud data proposed in this invention;

[0033] Figure 2 This is a flowchart of the process for finding the starting thickness point pair proposed in this invention;

[0034] Figure 3 This is a flowchart of the region growth method proposed in this invention;

[0035] Figure 4 This is a schematic diagram of the graph structure constructed by finding the starting thickness point pair proposed in this invention.

[0036] Figure 5 This is a schematic diagram of the region growth method proposed in this invention. Detailed Implementation

[0037] The specific embodiments of the present invention are described below with reference to the accompanying drawings and examples:

[0038] like Figure 1 The diagram shows a flowchart of an automatic method for measuring the thickness of scrap steel based on three-dimensional point cloud data proposed in this invention. This method includes the following steps:

[0039] Step S101: Input the original ordered point cloud data;

[0040] Step S102: Extract the edge features of the point cloud to obtain the edge point data set D;

[0041] Step S103: Calculate the principal component direction of each point. Specifically, for each point in the point cloud D, calculate its principal component direction based on the points in its neighborhood as the principal direction of this point.

[0042] Step S104: Perform Euclidean clustering on the point cloud D to obtain point sets D_class_0 to D_class_n.

[0043] Step S105, estimate the initial thickness point pair, specifically: for each D_class, first find the initial corresponding point q based on a point p to form an initial point pair;

[0044] Step S106: Based on the initial point pairs, use the region growing method to find more corresponding thickness point pairs that meet a certain estimated number;

[0045] Step S107: Calculate the average value of the thickness point pairs as the final estimated thickness.

[0046] In step S102, the present invention extracts point cloud edge features from the input point cloud data. The present invention provides two methods for point cloud edge extraction: a method based on depth abrupt change values ​​and a method based on curvature abrupt change values.

[0047] The method based on depth mutation values ​​first converts point cloud image data into depth image data, which uses depth values ​​as pixel values. Then, a certain threshold is set, and points in the neighborhood of each point whose depth values ​​change beyond the threshold are searched as edge feature points.

[0048] The method based on curvature abrupt change values ​​first requires calculating the normals of each point cloud. This invention uses an integral image method to calculate the normals of each point, and the formula for constructing the depth image integral image is as follows:

[0049]

[0050] Where I(x,y) is the sum of pixel values ​​within the rectangle from point (x,y) to (0,0).

[0051] After obtaining the normals, the Canny operator can be used to calculate edge feature points. Canny is an edge detection method.

[0052] The operator can remove false edges introduced by noise or irrelevant details, and obtain accurate edge data.

[0053] Figure 2 The flowchart for finding thickness point pairs proposed in this invention includes the following steps:

[0054] Step S201: Construct a graph structure for each clustered point set D_class. Specifically, when searching for thickness point pairs among similar feature points (i.e., all feature points obtained using the depth mutation value method or the curvature mutation value method), to distinguish between points on the same side and points on opposite sides, a graph structure must first be constructed as follows: Figure 4 The diagram structure shown has point p as the starting point and point q as the corresponding point to be found. p and q form the initial thickness point pair.

[0055] Step S202: Based on a point p, find the initial corresponding point q within a certain neighborhood. Specifically, using the direction of point p and point p to form a plane, find the set a of points a whose distance from the neighborhood to the plane is less than a certain threshold;

[0056] Step S203: To find the corresponding point q from the above set a, it is necessary to further evaluate the points in set a. Point q must satisfy both a small Euclidean distance from point p and a maximum graph distance to point p. For example... Figure 4 As shown, the distance on the graph from point p to point q is the length of the black trajectory in the graph;

[0057] In step S204, points p and q form an initial seed thickness pair and are stored in the seed point pair set. The point pairs stored in the seed point pair set will be used in the region growing method described below.

[0058] When combining edge feature data extracted by the depth catastrophe method and data extracted by the curvature catastrophe method for thickness estimation, since the point cloud data sources on both sides are different, it is sufficient to consider only the condition of the closest Euclidean distance, without constructing a graph structure.

[0059] The following is based on Figure 3 The process of the region growing method proposed in this invention to obtain more thickness point pairs includes the following steps:

[0060] Step S301: Find points p1 and q2 within a certain neighborhood of point p to obtain the point set w1 of p1; find points q2 within a certain neighborhood of point q to obtain the point set w2 of p2. Specifically: After obtaining the initial seed point pair p and q, as follows... Figure 5 As shown, we want to take vector pq as the normal vector and pass through plane 1 (point p) and plane 2 (point q), respectively. We then select the set of points w1 that is smaller in distance from plane 1 and the set of points w2 that is smaller in distance from plane 2.

[0061] Step S302: Find the corresponding point of w1 in w2 to form the thickness pair vector p2q2. Specifically: take a point p2 from the point set w1, where the direction of point p2 is basically the same as that of point p, such as... Figure 5As shown, construct a point-normal plane plane3 along the direction of point p2, find a set of points within a certain distance of plane3 from point set w2, select point q2 from this set, and then determine the angle and Euclidean distance between vectors p2q2 and pq. The angle must be less than a certain threshold (i.e., the two should be as parallel as possible), and the difference between the Euclidean distance of vector p2q2 and the Euclidean distance of vector pq must be less than a preset threshold.

[0062] Step S303: Add the point pair formed by vectors p2q2 to the thickness point pair set and the seed point pair set, and delete these two points from their respective boundary point sets.

[0063] Step S304: Find new seed thickness point pairs until all point pairs in the seed set are used up. Specifically: After detecting all thickness point pairs starting with vector pq as the seed, new seed point pairs need to be obtained. First, vector pq needs to be deleted from the seed set. Then, the point pairs in the seed set are sorted according to their distance, and the point pair with the smallest distance is used as the new seed thickness point pair. The search for thickness pairs is repeated until there are no more seeds in the seed set.

[0064] Step S305: The average distance between all thickness point pairs is taken as the final thickness. Specifically, when the number of thickness point pairs exceeds a preset threshold, the thickness can be estimated, and the average distance between all thickness point pairs is calculated as the final thickness value. Then, a hierarchical statistical histogram can be constructed to count the quantity of scrap steel at each thickness.

[0065] It should be noted that in step S302 above, if the depth mutation method or the curvature mutation method is used...

[0066] To extract edge feature data for separate thickness estimation, in order to determine that the found point is on the opposite side, a graph structure must be constructed to satisfy the requirement that the distance on the graph is relatively long, which is similar to the principle mentioned in step S204.

[0067] It should be noted that any modifications made according to specific embodiments of the present invention do not depart from the spirit of the present invention and the scope of the claims.

Claims

1. An automatic method for measuring metal thickness based on three-dimensional point cloud data, characterized in that, Includes the following steps: S101. Input the raw ordered point cloud data; S102. Extract the edge features of the point cloud to obtain the edge point data set D; S103. Calculate the principal component direction for each point in the edge point data set D; S104. Perform Euclidean clustering on the edge point data set D to obtain a point set; S105. Find the initial thickness point pair for the point set; S106. Based on the initial thickness point pairs, other thickness point pairs are found using the region growing method; S107. Calculate the average value of the thickness point pairs as the final thickness value; The specific steps for finding the initial thickness point pair are as follows: 1) Find thickness point pairs among similar feature points; 2) Find the initial corresponding point q based on a point p in the point set; 3) Place the found initial seed point pairs into the seed point pair set; The method described above for finding the initial corresponding point q includes the following steps: 1) Form a plane with the direction of point p and point p, and find the set of points a in the neighborhood whose distance to the plane is less than the distance threshold; 2) For points in set a, point q must satisfy both a small Euclidean distance from point p and a large graph distance from point p; The specific details of the methods for regional growth mentioned above are as follows: 1) Using the initial point pair vector pq as the normal vector, find the plane plane1 that passes through point p and the plane plane2 that passes through point q; 2) Select the point sets w1 and w2 that are closest to plane1 and plane2 respectively; 3) Take a point p2 from the point set w1, and construct the point normal plane plane3 along the direction of point p2; 4) Find the set of points within the preset distance of the point set w2 to the plane plane3, and find point q2; 5) Find the thickness point pairs with point pair pq as the seed point pair, and find new seed point pairs from them; 6) Repeat the above steps until the seed set is empty.

2. The automatic metal thickness measurement method based on three-dimensional point cloud data according to claim 1, characterized in that, The metal in question is scrap steel.

3. The automatic metal thickness measurement method based on three-dimensional point cloud data according to claim 1, characterized in that, The methods for extracting point cloud edge features include methods based on depth abrupt change values ​​or methods based on curvature abrupt change values.

Images