Model parameter extraction method and system for 3D scanning contour

Abstract

The invention discloses a 3D scanning contour model parameter extraction method and system, and belongs to the field of robot contour tracking. The method comprises the following steps: firstly, acquiring 3D contour scanning data, and carrying out data preprocessing, including downsampling, data filtering, smoothing, defect identification and compensation and the like, according to the characteristics of original data; then judging the number and the approximate positions of key points according to priori knowledge, and extracting key points of each frame of the contour; the local contour shape is identified according to an iteration nearest point; according to the key point of each frame of the contour and the local contour shape, extracting model parameters of the 3D contour; and finally reconstructing the 3D model according to the model parameters. The 3D scanning contour parameter extraction method overcomes the defects of the 3D contour, such as the negative effects of convex hulls, recesses, abrasion and the like on contour parameter extraction, and also overcomes the problems of target contour dislocation, data missing and the like caused by environment jitter or sensor reasons when the sensor scans data, so that the accuracy of 3D scanning contour parameter extraction is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of robot contour tracking, and more particularly to a method and system for extracting model parameters of a 3D scanning contour. Background Art

[0002] In the intelligent gluing system, a robot is usually used to drive a linear laser sensor to scan the physical object to obtain three-dimensional contour data. The host computer then uses a suitable algorithm to accurately extract the gluing position of the three-dimensional contour point cloud, and finally glue can be applied accurately.

[0003] Therefore, when applying glue, the sensor needs to accurately track the three-dimensional contour, and the host computer needs to process the three-dimensional contour point cloud data in a timely manner and accurately extract certain geometric elements of the three-dimensional contour point cloud, such as the side and center line of the contour.

[0004] There are three main difficulties in this process. First, the trajectory of the foam is complex and curved. When the robot drives the sensor to track the contour, the sensor scanning area is easy to deviate from the target contour, resulting in the inability to accurately obtain the target contour information; second, the foam has no fixed shape and is easy to deform, such as thermal expansion and contraction, which makes it impossible to plan the accurate gluing path in advance; therefore, when tracking the three-dimensional contour, the gluing path must be adjusted in real time according to the point cloud data of the three-dimensional contour; third, the contour cross-section shapes vary, and there are convex hulls, concave hulls, narrowing, glue shortages, glue breakages, etc., which have a greater negative impact on the accurate extraction of the gluing path.

[0005] Due to the above problems, previous methods directly extract feature points of contour point clouds and then reconstruct the model. This is difficult for complex three-dimensional contours. Not only are feature points difficult to extract accurately, but it is also difficult to resist the negative impact of contour point cloud defects, resulting in low accuracy of extracted model parameters. Summary of the invention

[0006] In view of this, the present invention provides a method and system for extracting model parameters of a 3D scanning contour, which can effectively improve the accuracy of 3D scanning contour parameter extraction.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] In one aspect, the present invention discloses a method for extracting model parameters of a 3D scanning profile, comprising the following steps:

[0009] Acquire 3D contour scanning point cloud data, including multiple frames of data, each frame of data including multiple data points, that is, each contour point of the target contour part;

[0010] Preprocessing the 3D contour scanning point cloud data, identifying data defects and compensating for them, to obtain contour data;

[0011] Extracting key points based on the contour data;

[0012] Processing the contour data and the key points using an iterative closest point algorithm to identify a local contour;

[0013] The distances of the key points are calculated, and the geometric parameters of the inverse model are obtained in combination with the local contour.

[0014] Preferably, the data defects include data missing, convex hull, and misalignment.

[0015] Preferably, identifying data deficiencies and compensating for them includes:

[0016] Set a data missing threshold, calculate the mean value of each contour point in the z direction of the target contour part of each frame, and when the mean value is less than the data missing threshold, it is determined that the target contour data is missing, otherwise it is determined to be normal; when the target contour data is missing, compensation is performed according to the normal target contour data of the previous frame;

[0017] Set an SSE threshold, calculate the sum of square errors between the current frame and the adjacent frames in the z direction, and when the sum of square errors exceeds the SSE threshold, determine that a convex hull exists, and offset the current frame data along the z direction by the offset between the current frame data point and the previous frame data point to complete the convex hull compensation;

[0018] Whether there is misalignment is determined by calculating the offset of the data point of the current frame relative to the data point of the previous frame in the y direction, and misalignment compensation is performed on the current frame according to the offset in the y direction.

[0019] Preferably, extracting key points based on the contour data includes:

[0020] The contour data is combined with prior knowledge to extract key points.

[0021] Preferably, the contour data and the key points are processed using an iterative closest point algorithm to identify the local contour, including:

[0022] Constructing target point cloud contour templates of different shapes through the key points;

[0023] Acquire contour point cloud data according to the key points;

[0024] Performing ICP registration on the target point cloud contour template and the contour point cloud data based on the objective function;

[0025] The registration error is calculated, and the shape of the target point cloud contour template with the smallest registration error is the shape of the local contour point cloud.

[0026] Preferably, the objective function is as follows:

[0027]

[0028] Where n is the total number of data points, T′ i is the i-th data point in the target point cloud contour template, S i is the i-th contour point cloud data, R is the rotation matrix, and t is the translation matrix.

[0029] Preferably, the registration error calculation formula is as follows:

[0030]

[0031] Among them, P i To inversely find the i-th point in the model, T indice is the distance P in the theoretical model i The nearest point, indice is found through kdtree, N′ is the total number of corresponding points between the inverse model and the theoretical model, and d(·) is the point P i and point T indice The distance between.

[0032] Preferably, the method further comprises: reconstructing a three-dimensional contour model based on the geometric parameters and the target centerline trajectory.

[0033] On the other hand, the present invention also provides a 3D scanning profile model parameter extraction system, which is used to implement the above 3D scanning profile model parameter extraction method, comprising:

[0034] A data acquisition module is used to obtain 3D contour scanning point cloud data, including multiple frames of data, each frame of data includes multiple data points, that is, each contour point of the target contour part;

[0035] A data processing module is used to pre-process the 3D contour scanning point cloud data, identify data defects and compensate for them, and obtain contour data;

[0036] A key point determination module, used for extracting key points based on the contour data;

[0037] A local contour recognition module, used for processing the contour data and the key points using an iterative closest point algorithm to recognize a local contour;

[0038] The parameter extraction module is used to calculate the distance between each of the key points and obtain the geometric parameters of the inverse model in combination with the local contour.

[0039] Preferably, the system further comprises a three-dimensional model reconstruction unit for reconstructing a three-dimensional contour model based on the geometric parameters and the target centerline trajectory.

[0040] Through the above technical solutions, it can be seen that compared with the prior art, the present invention discloses a model parameter extraction method and system for 3D scanning contours. In the point cloud preprocessing stage, the defects of the three-dimensional point cloud contour are identified and compensated to resist the negative impact of the defects on the model parameter extraction; the key points of the contour are identified according to prior knowledge, which avoids the blindness of contour key point extraction, greatly reduces the workload of key point screening, improves the efficiency and stability of key point extraction, and makes the algorithm have strong stability; the contour point cloud data is directly constructed according to the key points, so that the target point cloud and the source point cloud have better initial values, which effectively avoids the problem that the traditional ICP algorithm is sensitive to the initialization value and easily produces local optimal solutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0042] Figure 1 Flowchart of the model parameter extraction method for 3D scan profile.

[0043] Figure 2 This is a physical picture of the measurement object.

[0044] Figure 3 It is a normal 3D contour point cloud.

[0045] Figure 4 It is a 3D contour point cloud with defects.

[0046] Figure 5 This is the preprocessing result of the normal point cloud.

[0047] Figure 6 This is the preprocessing result of the defect point cloud.

[0048] Figure 7 is the key point information obtained based on prior knowledge.

[0049] Figure 8 Reconstructed model of normal point cloud.

[0050] Fig. 9 Reconstructed model of defect point cloud. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] On the one hand, an embodiment of the present invention discloses a method for extracting model parameters of a 3D scanning profile, such as Figure 1 As shown, the method comprises the following steps:

[0053] S1. Acquire 3D contour scanning point cloud data, including multiple frames of data, each frame of data including multiple data points, that is, each contour point of the target contour part.

[0054] The embodiment of the present invention uses a three-dimensional measurement system based on a line laser sensor to obtain 3D contour scanning point cloud data. The three-dimensional measurement system is mainly composed of a six-axis robot, a controller, a line laser sensor, a measurement object, and a host computer. The six-axis robot uses a FANUC robot, model R-2000iC / 165F; the line laser sensor uses a 3D measurement sensor from XinGe Technology, model sG56N 060x040; the host computer model is Raytheon 911Air, and the CPU is i7-11800H.

[0055] When the robot drives the line laser sensor to scan the measured object, the contour information of the line laser sensor and the position information of the robot are synchronously transmitted to the PC; the host computer can then construct the 3D point cloud contour of the measured object based on these data. Figure 2 As shown, the obtained three-dimensional original contour normal point cloud is as follows Figure 3 As shown, the point cloud with defects is as follows Figure 4 shown.

[0056] S2. Preprocess the 3D contour scanning point cloud data, identify data defects and compensate for them to obtain contour data. The preprocessing results are as follows: Figure 5 , 6 shown.

[0057] Data preprocessing mainly includes two parts:

[0058] First, perform routine data processing, such as downsampling, data filtering, ROI extraction, smoothing, etc.

[0059] Second, defects are identified and compensated for the data. This embodiment takes data missing, convex hull, and misalignment as examples.

[0060] Specifically, the target contour data is missing, which is characterized by incomplete point cloud of the target contour. For the missing target contour data, a certain threshold Z can be set. lim , and calculate the mean value z of each contour point in the z direction of the target contour part of each frame mean ; When z mean Less than Z lim When , it can be judged that the target contour data is missing;

[0061] When compensating, compensation can be performed based on the normal contour data of the previous frame, including:

[0062] Find the adjacent normal frame and extract the center point P of the frame i-1 , and predict the center point coordinates P of the defect frame i ;

[0063] Select the target contour part PC of the previous frame normal point cloud i ;

[0064] Calculate the offset Δi = P i -P i-1 ;

[0065] The target contour point cloud compensated in the i-th frame is PC i =PC i-1 +Δi.

[0066] The main feature of the convex hull is that there is a certain height difference in the z direction between frames. Therefore, the convex hull can be identified based on the sum of square errors SSE in the z direction between the current frame and the adjacent frames. When SSE exceeds a certain threshold, it can be determined that a convex hull exists; SSE can be expressed as:

[0067]

[0068] in, represents the jth point of the i-th frame data, and n is the smaller value of the number of data points in the i-th frame and the i-1-th frame;

[0069] For the compensation of the convex hull, the offset d of the data point of the i-th frame relative to the data point of the i-1th frame can be calculated i , and then offset the i-th frame point cloud data along the z direction, specifically including:

[0070] Calculate the mean value of the z direction of the i-th frame data point and the i-1-th frame data point respectively

[0071] Calculate the offset in the z direction between adjacent frames

[0072] Calculate the compensated point cloud by offset: (Only the z-direction coordinates are updated, the x- and y-direction coordinates remain unchanged).

[0073] The misaligned contour point cloud data mainly has differences in the y direction, so the offset dy in the y direction of the i-th frame data point relative to the i-1-th frame data point can be calculated i To identify;

[0074] Compensation can be based on dy i Compensate the current frame, including:

[0075] Extract the center point P of the target contour i ;

[0076] Calculate the offset in the y direction between adjacent frames in P i Coordinate value in the y direction;

[0077] Compensating for misaligned point clouds (Only the y-direction coordinates are updated, the x- and z-direction coordinates remain unchanged).

[0078] S3. Extract key points based on contour data, including:

[0079] Contour data is combined with prior knowledge to extract key points.

[0080] For the contour data obtained in this embodiment, the contour line of each frame can be roughly divided into two parts, the lower part, the shape of the lower part is mainly a simple quadrilateral, and the shape of the upper part is a triangle, an arc, etc. According to the commonality of these cross-sectional contour lines, 5 key points can be selected, and their approximate positions are as follows: Figure 7 P1 to P5 in the figure. P1 and P5 are two end points of a certain section contour line, P3 is the vertex of the contour line, and P2 and P4 are end points of the upper part of a certain section contour line.

[0081] And according to prior knowledge, P1 and P5 are symmetrical about P3 in the horizontal direction; P2 and P4 are symmetrical about P3 in the horizontal direction.

[0082] S4. Use the iterative closest point algorithm to process the contour data and key points and identify the local contour, including:

[0083] Construct target point cloud contour templates of different shapes through key points;

[0084] Obtain contour point cloud data based on key points;

[0085] Perform ICP registration of the target point cloud contour template and contour point cloud data based on the objective function;

[0086] The registration error is calculated, and the shape of the target point cloud contour template with the smallest registration error is the shape of the local contour point cloud.

[0087] In this embodiment, the source point cloud S (i.e., contour point cloud data) of the part is obtained according to P2, P3, and P4; target point clouds (i.e., target point cloud contour templates) T1, T2, etc. of different shapes (center of circle, triangle, quadrilateral, etc.) are constructed through P2, P3, and P4, so that the target point cloud and the source point cloud have good initial values;

[0088] Based on the objective function, the source point cloud and the target point cloud are registered by the ICP algorithm, and the registration errors are calculated respectively; the objective function formula is as follows:

[0089]

[0090] Where n is the total number of data points, T′ i is the i-th data point in the target point cloud contour template, S i is the i-th contour point cloud data, R is the rotation matrix, and t is the translation matrix.

[0091] The registration error calculation formula is as follows:

[0092]

[0093] Where P i To inversely find the i-th point in the model, T indice is the distance P in the theoretical model i The nearest point, indice is found through kdtree, N′ is the total number of points corresponding to the inverse model and the theoretical model, and d(·) is the point P i and point T indice The distance between.

[0094] The shape of the target point cloud contour template with the smallest registration error is selected as the local contour shape.

[0095] S5. Calculate the distance between each key point and combine it with the local contour to obtain the geometric parameters of the inverse model. The geometric parameters of the inverse model include the distance between key points and the shape of the local contour. The shape of the local contour has been obtained in S4. Assuming that the coordinates of the key points are P1(x1,y1) and P2(x2,y2), the distance d between the key points is calculated by the following formula:

[0096]

[0097] S6. Reconstruct the 3D contour model based on the geometric parameters and the target centerline trajectory.

[0098] Select an appropriate method to extract the target centerline trajectory according to the actual scenario, including but not limited to the following methods:

[0099] 1. Based on geometric shape analysis

[0100] For objects with specific geometric shapes or symmetrical structures, the center line is determined by using its geometric features. For objects with simple geometric shapes, the center line can be directly determined by connecting specific points, while for symmetrical objects, the center line is located by analyzing the intersection of the symmetry plane or symmetry axis.

[0101] 2. Calculation based on algorithms

[0102] (1) Skeleton extraction algorithm: including distance transformation method and topological refinement method.

[0103] The distance transformation method determines the skeleton points by calculating the distance from the contour points to the internal points, and then obtains the centerline trajectory.

[0104] The topology refinement method obtains the skeleton structure, i.e., the center line, by gradually deleting boundary pixels and maintaining connectivity.

[0105] (2) Principal component analysis (PCA) method: Perform principal component analysis on the contour point cloud data, take the direction of the first principal component as the centerline direction, and then determine the centerline trajectory through sampling.

[0106] 3. Based on model fitting

[0107] For objects with known approximate shape models, the centerline is determined by fitting the model. Simple-shaped objects can be directly fitted with a specific fitting algorithm, while complex-shaped objects can be fitted in sections and then combined to obtain the centerline trajectory.

[0108] Combining the geometric parameters and the centerline trajectory, the reconstruction models of the normal point cloud and the defect point cloud in this embodiment are as follows: Figure 8 , 9 As shown. Figure 8-Figure 9 It can be seen that this method can correctly extract model parameters for both normal point clouds and defective point clouds, and the reconstructed model can match the theoretical model well. Three groups of normal point clouds and defective point clouds were selected for experimental testing; the experimental results show that the volume ratio of the reconstructed inverse model to the theoretical model is 0.9805, the intersection ratio is 0.8024, and the registration distance is 0.3143mm.

[0109] On the other hand, the present invention also provides a 3D scanning profile model parameter extraction system, which is used to implement the above 3D scanning profile model parameter extraction method, comprising:

[0110] A data acquisition module is used to obtain 3D contour scanning point cloud data, including multiple frames of data, each frame of data includes multiple data points, that is, each contour point of the target contour part;

[0111] The data processing module is used to pre-process the 3D contour scanning point cloud data, identify data defects and compensate for them, and obtain contour data;

[0112] A key point determination module, used to extract key points based on contour data;

[0113] A local contour recognition module is used to process contour data and key points using an iterative closest point algorithm to recognize local contours;

[0114] The parameter extraction module is used to calculate the distance of each key point and obtain the geometric parameters of the inverse model in combination with the local contour.

[0115] Preferably, the system further comprises a three-dimensional model reconstruction unit for reconstructing the three-dimensional contour model based on the geometric parameters and the target centerline trajectory.

[0116] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0117] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for extracting model parameters of a 3D scanning profile, characterized in that: The following steps are involved: Acquire 3D contour scanning point cloud data, including multiple frames of data, each frame of data including multiple data points, that is, each contour point of the target contour part; Preprocessing the 3D contour scanning point cloud data, identifying data defects and compensating for them, to obtain contour data; Extracting key points based on the contour data; Processing the contour data and the key points using an iterative closest point algorithm to identify a local contour; The distances of the key points are calculated, and the geometric parameters of the inverse model are obtained in combination with the local contour.

2. The method for extracting model parameters of a 3D scanning profile according to claim 1, characterized in that: The data defects include data missing, convex hull, and misalignment.

3. The method for extracting model parameters of a 3D scanning profile according to claim 2, characterized in that: Identify data deficiencies and compensate for them, including: Set a data missing threshold, calculate the mean value of each contour point in the z direction of the target contour part of each frame, and when the mean value is less than the data missing threshold, it is determined that the target contour data is missing, otherwise it is determined to be normal; when the target contour data is missing, compensation is performed according to the normal target contour data of the previous frame; Set an SSE threshold, calculate the sum of square errors between the current frame and the adjacent frames in the z direction, and when the sum of square errors exceeds the SSE threshold, determine that a convex hull exists, and offset the current frame data along the z direction by the offset between the current frame data point and the previous frame data point to complete the convex hull compensation; Whether there is misalignment is determined by calculating the offset of the data point of the current frame relative to the data point of the previous frame in the y direction, and misalignment compensation is performed on the current frame according to the offset in the y direction.

4. The method for extracting model parameters of a 3D scanning profile according to claim 1, characterized in that: Extracting key points based on the contour data includes: The contour data is combined with prior knowledge to extract key points.

5. The method for extracting model parameters of a 3D scanning profile according to claim 1, characterized in that: The contour data and the key points are processed by an iterative closest point algorithm to identify a local contour, including: Constructing target point cloud contour templates of different shapes through the key points; Acquire contour point cloud data according to the key points; Performing ICP registration on the target point cloud contour template and the contour point cloud data based on the objective function; The registration error is calculated, and the shape of the target point cloud contour template with the smallest registration error is the shape of the local contour point cloud.

6. The method for extracting model parameters of a 3D scanning profile according to claim 5, characterized in that: The objective function is as follows: Where n is the total number of data points, T′ i is the i-th data point in the target point cloud contour template, i is the i-th contour point cloud data, R is the rotation matrix, and t is the translation matrix.

7. The method for extracting model parameters of a 3D scanning profile according to claim 5, characterized in that: The registration error calculation formula is as follows: Among them, P i To inversely find the i-th point in the model, T indice is the distance P in the theoretical model i The nearest point, indice is found through kdtree, N′ is the total number of points corresponding to the inverse model and the theoretical model, and d(·) is the point P i and point T indice The distance between.

8. The method for extracting model parameters of a 3D scanning profile according to claim 1, characterized in that: Also includes: A three-dimensional contour model is reconstructed based on the geometric parameters and the target centerline trajectory.

9. A model parameter extraction system for 3D scanning contour, characterized in that: include: A data acquisition module is used to obtain 3D contour scanning point cloud data, including multiple frames of data, each frame of data includes multiple data points, that is, each contour point of the target contour part; A data processing module is used to pre-process the 3D contour scanning point cloud data, identify data defects and compensate for them, and obtain contour data; A key point determination module, used for extracting key points based on the contour data; A local contour recognition module, used for processing the contour data and the key points using an iterative closest point algorithm to recognize a local contour; The parameter extraction module is used to calculate the distance between each of the key points and obtain the geometric parameters of the inverse model in combination with the local contour.

10. A 3D scanning contour model parameter extraction system according to claim 9, characterized in that: It also includes a three-dimensional model reconstruction unit, which is used to reconstruct a three-dimensional contour model based on the geometric parameters and the target centerline trajectory.

Images