A point cloud-based belt lap joint defect detection method

Abstract

This invention discloses a point cloud-based method for detecting belt layer overlap defects, mainly comprising the following steps: S1: A 3D camera acquires point clouds of the belt layer material on the production line; S2: Planar correction and uniform sampling; S3: After calculating the centroid coordinates, the bottom point cloud is segmented; S4: The scanned point cloud data is fitted to obtain the coordinate parameters of the centroid, and the bottom point cloud and the belt point cloud are cut according to the belt thickness feature to obtain the belt point cloud; S5: The plane of the segmented belt point cloud is fitted to obtain the parameters of the fitted plane, and the point set outside the threshold is calculated through the plane parameters. The remaining points are segmented by Euclidean clustering to separate various cluster sets, and invalid clusters are filtered out according to the belt overlap angle range; S6: The boundary point set of the overlap point cloud is calculated and projected onto the depth map to obtain the overlap distribution range. This invention method is simple, easy to implement, efficient and practical. This invention method is suitable for the belt layer production process of various tires, has high robustness, and low false detection rate and misdetection rate.

Description

Technical Field

[0001] This invention relates to the fields of intelligent manufacturing, defect detection, and image processing, specifically to a method for detecting defects in belt layer overlap based on point clouds. Background Technology

[0002] As one of the most important components of a car, the quality of tires not only affects the performance of the car, but also the personal safety of the driver. The belt layer refers to the material layer that is wrapped around the tire body along the circumferential direction of the center line of the tread base of radial tires and bias-ply tires. It is an important component of the tire that bears the force, and plays the role of absorbing impact and tightening the tire body.

[0003] Since the quality of the belt layer material joints in the tire production process is mainly controlled manually and is subject to the subjective influence of workers, it is easy to miss or mis-inspect, resulting in overlapping of the belt layer joints, which affects the uniformity and safe life of the tire.

[0004] Currently, methods for detecting overlap defects and foreign objects are widely used in tire production. However, in the detection of overlaps in belt layer joints, some current methods identify the overlap area using depth or RGB images. Due to the limitations of the depth and RGB value ranges of two-dimensional images, the detection of overlap features of belt layer material is not accurate enough, easily leading to false positives and false negatives. The mainstream detection method uses X-ray inspection, but this method is mainly used to inspect finished tires and cannot reduce the defect rate in the production process, nor can it reduce the waste of materials and processes. At the same time, the inspection equipment is complex in structure and too expensive, and X-rays have an impact on the health and safety of on-site personnel, requiring additional protection for personnel. Summary of the Invention

[0005] The purpose of this invention is to provide a point cloud-based method for detecting defects in belt layer overlap, in order to solve the problems mentioned in the background art, such as the inability to reduce the defect rate in the production process, the inability to reduce the waste of materials and processes, the complexity and high cost of the detection equipment, and the impact of X-rays on the health and safety of on-site personnel, requiring additional protection for personnel.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a point cloud-based method for detecting overlap defects in belt layers, comprising: S1, acquiring point clouds of the belt strip on the belt layer production line using a 3D camera; S2, plane correction and uniform sampling; S3, segmenting the bottom point cloud after determining the centroid coordinates; S4, segmenting the overlap point cloud after fitting a plane; S5, filtering out interfering point clouds after clustering and segmentation; S6, determining the boundary point set of the overlap point cloud and projecting it onto a depth map to obtain the overlap distribution range.

[0007] The above-mentioned method for detecting overlap defects in belt layers based on point clouds specifically includes the following steps:

[0008] S1: Point cloud data acquisition: The point cloud data is acquired using our company's line laser scanning measurement sensor and includes the three-dimensional information of the actual material strip and the three-dimensional information of its bottom surface, such as... Figure 4 As shown;

[0009] S2: Planar correction and uniform sampling: Planar correction is achieved through Euclidean rotation and translation matrix, while uniform sampling reduces the amount of point cloud data and preserves the shape features of the point cloud by replacing all point clouds within a voxel with points close to the center point in the voxel grid.

[0010] S3: After finding the centroid coordinates, segment the bottom point cloud: The centroid of the point cloud is essentially the average of the x, y, and z coordinates of all points in the point cloud. The bottom point cloud can be segmented using the height difference between the material strip and the bottom surface through the centroid.

[0011] S4: Segmenting the overlapping point cloud after fitting the plane: Based on the point cloud of the material strip obtained in step S3, due to the flexible characteristics of the material strip itself, fluctuations will occur during the operation of the material strip. When collecting data, the fluctuation point cloud and the overlapping point cloud are mixed together. Therefore, when segmenting the point cloud of the material strip according to the fitted plane parameters, the fluctuation point cloud and the overlapping point cloud will be segmented at the same time.

[0012] S5: Filtering interfering point clouds after clustering and segmentation: Based on the point cloud obtained in step S4, it can be divided into n point cloud clusters through Euclidean clustering. Calculate the bounding box of each cluster. The center line of the point cloud can be obtained through the corner points of the bounding box. Calculate the angle between the center line of the cluster and the normal vector of the running direction. Filter out non-overlapping point clouds through the angle of the material belt interface.

[0013] S6: Find the boundary point set of the overlapping point cloud and project it onto the depth map to obtain the overlapping distribution range: Since the normal of the boundary points is less than the right angle, the boundary points of the point cloud can be calculated based on the change of the normal vector of each point; Due to the mapping relationship between the 3D points and the 2D depth map, the boundary point cloud can be projected onto the depth map to obtain the overlapping distribution range.

[0014] Compared with the prior art, the beneficial effects of the present invention are:

[0015] This point cloud-based method for detecting belt layer overlap defects involves fitting a plane of the segmented belt point cloud to obtain the parameters of the fitted plane. The parameters are then used to calculate the set of points outside a threshold. Euclidean clustering is performed on the remaining points to create various cluster sets. Invalid clusters are filtered out based on the belt overlap angle range. The boundary point set of the overlap point cloud is then calculated and projected onto a depth map to obtain the overlap distribution range. This method is simple, efficient, and practical, suitable for various tire belt layer production processes, exhibiting high robustness and low false detection and misdetection rates. Attached Figure Description

[0016] Figure 1The flowchart is a method for detecting overlap defects in belt layers based on point clouds according to the present invention.

[0017] Figure 2 This is a schematic diagram of the device for scanning point clouds according to the present invention;

[0018] Figure 3 This is a schematic diagram of the point cloud of the belt-layer material scanned according to the present invention;

[0019] Figure 4 The image used in this invention is a scanned image of the actual belt layer material depth.

[0020] Figure 5 The data used in this invention example is real belt layer material point cloud data obtained by scanning;

[0021] Figure 6 For the comparison of point cloud data before and after horizontal correction in this invention, red represents the original point cloud data, and green represents the point cloud data after horizontal correction.

[0022] Figure 7 This refers to the sparse point cloud data obtained by the uniform sampling method in this invention.

[0023] Figure 8 This is a diagram showing the effect of cutting the bottom surface after fitting the plane according to the present invention. Red represents the point cloud of the bottom surface, and green represents the point cloud of the material strip.

[0024] Figure 9 This is an image showing the effect of cutting the overlapping defect point cloud after fitting the plane according to the present invention. Green represents the material strip point cloud, and white represents the overlapping defect point cloud.

[0025] Figure 10 This is the point cloud of overlapping defects cut out by the present invention;

[0026] Figure 11 The diagram shows the distribution of point cloud data for the bottom surface point cloud, material strip point cloud, and overlap defect point cloud of this invention. Red represents the bottom surface point cloud, green represents the material strip point cloud, and white represents the overlap defect point cloud. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see Figure 1-11 The present invention provides a technical solution: a method for detecting overlap defects in belt layers based on point clouds.

[0029] like Figure 1As shown, the present invention includes the following steps:

[0030] S1: Use an industrial camera to collect point cloud data P0 of the belt layer, which includes point cloud P1 of the belt layer material and point cloud P2 of the bottom surface.

[0031] S2-1: Due to the complexity of the actual situation on the production line, the installation of the data acquisition equipment or the inclination of the production line itself may cause the material belt to deviate during the conveying process. Therefore, the collected point cloud data needs to be corrected for planar orientation.

[0032] S2-2: Due to the excessive density of the collected point cloud data, uniform sampling is performed on P0, and a 3D voxel mesh is created on P0, with one point taken from each voxel. Replace all the points within the voxel to obtain the point cloud P0';

[0033] S2-3: Calculate the fitting plane S1 of the point cloud P0', and obtain the normal vector of S1:

[0034]

[0035] S2-4: Find the normal vector of the fitting plane S1 Normal vector to the horizontal plane S0 The included angle v1 between them;

[0036] S2-5: Rotate the point cloud P0` by an angle v1 using the rotation and translation matrix T to obtain the horizontally corrected planar point cloud P0``;

[0037]

[0038] S3: Find the centroid of P0`` Since the material belt is attached to the conveyor belt and runs on it, there is a height difference between the material belt and the bottom surface of the conveyor belt. Based on the z value of the centroid m0 of the point cloud P0``, a through-filter is performed to divide P0`` into the belt point cloud P1 and the bottom point cloud P2.

[0039] S4-1: Find the fitting plane S2 of the bundled point cloud P1, and calculate the centroid of P1:

[0040]

[0041] S4-2: Calculate the distance between each point and plane S2, and set a distance threshold:

[0042]

[0043] S4-3: Where, Z max Z represents the average thickness of the belt layer. m1Given the z-value of the centroid m1 of P1, delete points within a distance threshold D;

[0044] S5-1: Perform Euclidean clustering on the remaining points to create various cluster sets N. j (P i );

[0045] S5-2: For each cluster set N j (P i Find the OBB bounding box, obtain the coordinates of each pole of the bounding box and the width and height of each cluster;

[0046] S5-3: Calculate the angle β° between the normal phase and the y-axis. Since the overlap angle during strip production is α°, the clustering of non-overlapping areas is filtered out by the angle α.

[0047] S5-4: False detections and misdetections may occur during the detection of overlaps. Interference elements need to be filtered out, such as wrinkles, waves, and displacement of the belt, which can cause interference.

[0048] S5-5: After filtering out interfering elements, the detection result is obtained, namely the point cloud Res1 of the overlapping part;

[0049] S6-1: Calculate the OBB bounding box of the point cloud Res1 and find the length and width of the bounding box;

[0050] S6-2: Based on the normal estimation of the boundary points of the point cloud, search for the neighboring region with radius r for any point in the point cloud Res1, and denote the points in the neighboring region as N(P) i Find N(P) i normal vector of ) ,calculate Phasor with Res1 method The included angle The included angle at the edge points Greater than the threshold When a point is identified as an edge point, the boundary point set Res2 of the resulting point cloud Res1 is obtained by statistical analysis.

[0051] S6-3: Based on the pinhole imaging principle, each three-dimensional point P in the point set Res2 is... i Projecting onto a 2D depth map, obtain the corresponding pixel coordinates in the depth map. This will give you the overlapping area on the two-dimensional map;

[0052] The method of this invention is simple, easy to implement, efficient and practical. It is suitable for the production process of belt layers in various tires, has high robustness, and low false detection rate and misdetection rate.

[0053] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

[0054] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A point cloud based belt splicing defect detection method, comprising: S1. A 3D camera captures point clouds of the material strip on the belt layer production line. S2, planar correction and uniform sampling; S3. After finding the centroid coordinates, segment the bottom point cloud; S4. After fitting the plane, segment the overlapping point cloud; S5. After clustering and segmentation, filter out the interfering point cloud; S6. Find the boundary point set of the overlapping point cloud and project it onto the depth map to obtain the overlapping distribution range. Its characteristic is that it specifically includes the following steps: S1: Collect point cloud data The point cloud data was obtained using our company's line laser scanning measurement sensor and includes the three-dimensional information of the actual material strip and the three-dimensional information of the bottom surface. S2: Planar correction and uniform sampling Planar correction is achieved through Euclidean rotation and translation matrices, while uniform sampling reduces the amount of point cloud data and preserves the shape features of the point cloud by replacing all point clouds within a voxel with points close to the center point in the voxel lattice. S3: Divide the bottom point cloud after finding the centroid coordinates. The centroid of a point cloud is essentially the average of the x, y, and z coordinates of all points in the point cloud. The centroid can be used to divide the bottom point cloud by utilizing the height difference between the material strip and the bottom surface. S4: Segmentation of overlapping point cloud after fitting plane Based on the point cloud of the material belt obtained in step S3, due to the flexible characteristics of the material belt itself, fluctuations will occur during the operation of the material belt. When collecting data, the fluctuation point cloud and the overlapping point cloud are mixed together. Therefore, when segmenting the point cloud of the material belt according to the fitted planar parameters, the fluctuation point cloud and the overlapping point cloud will be segmented at the same time. S5: Filtering interfering point clouds after clustering and segmentation Based on the point cloud obtained in step S4, it can be divided into n point cloud clusters by Euclidean clustering. The bounding box of each cluster is calculated. The center line of the point cloud can be obtained through the corner points of the bounding box. The angle between the center line of the cluster and the normal vector of the running direction is calculated. Non-overlapping point clouds are filtered out by the angle of the material belt interface. S6: Find the boundary point set of the overlapping point cloud, project it onto the depth map to obtain the overlapping distribution range. Since the normals of the boundary points are all less than right angles, the boundary points of the point cloud can be calculated based on the changes in the normal vector of each point. Due to the mapping relationship between the three-dimensional points and the two-dimensional depth map, the boundary point cloud can be projected onto the depth map to obtain the overlapping distribution range. It also includes the following steps: (1) Input the point cloud data of the belt layer material detected on the production line, where each point cloud data has location information; (2) Perform noise reduction preprocessing and uniform sampling preprocessing on the point cloud data of the belt layer material; (3) Perform plane fitting on the real point cloud data to obtain the fitting plane of the real material strip; (4) Calculate the angle between the plane normal phase and the vertical normal phase based on the obtained fitting plane, and perform plane correction on the pre-processed belt layer material point cloud; (5) Based on the centroid of the corrected point cloud, cut the bottom plane according to the mass position; (6) Compute the points after cutting to fit a plane and obtain the parameters of the fitting plane; (7) Cut the material strip point cloud according to the obtained fitted plane, obtain the point cloud of the raised part of the material strip, and perform clustering processing. Calculate the bounding box of each point cloud block obtained by clustering, obtain the angle of each point cloud block to the horizontal axis, and filter the point cloud of the non-joint area through the angle range of the joint. (8) Finally, the point cloud data of the overlapping area is obtained. The boundary of the point cloud is extracted based on the normal, and the boundary point set of the overlapping area is obtained. The overlapping area range is obtained by projecting it onto the two-dimensional depth map. Two plane fitting operations are required: the point cloud after horizontal correction and the point cloud of the cut strip. The horizontally corrected point cloud needs to calculate the angle between the normal vector of the horizontal plane and the normal vector of the fitting plane of the true point cloud.

2. The method for detecting overlap defects in belt layers based on point clouds according to claim 1, characterized in that: The collection of real material strip point cloud data on the production line requires the collection of: point cloud of the bottom surface of the production line and point cloud of the material strip.

3. The method for detecting overlap defects in belt layers based on point clouds according to claim 1, characterized in that: The point cloud of the material strip originates from the material strip produced on the production line, and the line laser scanning equipment splices the point cloud in a horizontal direction.

4. The method for detecting overlap defects in belt layers based on point clouds according to claim 1, characterized in that: The point cloud on the bottom surface of the production line originates from the structure of the material-carrying belt on the production line. There are two types: conveyor belt and roller. Both are represented as planar point clouds in the point cloud data.

Images