Power supply case structural member size automatic measurement method based on full-field three-dimensional vision

Abstract

This invention presents an automatic measurement method for power supply chassis structural components based on full-field 3D vision. It utilizes a phase-shifting fringe projection measurement method, employing multi-brightness fringe projection synthesis to obtain a single-viewpoint point cloud of the power supply chassis structural component from a single measurement viewpoint. Based on this, point clouds from all measurement viewpoints are acquired. These point clouds are then stitched together using randomly positioned markers on the measurement turntable to obtain a complete point cloud of the power supply chassis structural component. This complete point cloud data is then denoised to obtain the measurement point cloud of the power supply chassis structural component. A model point cloud is obtained based on the CAD model of the power supply chassis structural component. The measurement point cloud and the model point cloud are registered to a unified coordinate system. Based on the shape features extracted from the power supply chassis structural component model, local measurement point clouds corresponding to these shape features are extracted from the measurement point cloud. These local measurement point clouds are fitted to obtain geometric parameters, and the measured dimensional information of the shape features is calculated based on these geometric parameters. This invention is highly adaptable, accurate, and fast, enabling automated inspection of structural components.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional measurement and relates to an automatic measurement method for the dimensions of power supply chassis structural components based on full-field three-dimensional vision. Background Technology

[0002] Onboard equipment structural components, such as power supply chassis, are crucial satellite parts primarily used in highly integrated spacecraft electronic systems. The inspection efficiency and accuracy requirements for these components are high. Power supply chassis structural components are typically high-precision machined parts, with numerous individual pieces. Currently, manual measurement with calipers is the primary method. However, when measuring a large number of dimensions, manual measurement is time-consuming and inaccurate, failing to meet inspection requirements. While contact measurement methods such as coordinate measuring machines offer high accuracy, they require programming the measurement path, resulting in high offline programming costs and hindering practical production needs. Summary of the Invention

[0003] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides an automatic measurement method for the dimensions of power supply chassis structural components based on full-field three-dimensional vision. This solves the technical problem that existing manual and contact-based measurements are inefficient and inaccurate, failing to meet the actual inspection requirements of power supply chassis structural components.

[0004] This invention proposes an automatic measurement method for the dimensions of power supply chassis structural components based on full-field 3D vision, including:

[0005] After multi-brightness stripe projection and synthesis, the 3D vision measurement system acquires a single measurement viewpoint point cloud of a power supply chassis structural component from a single measurement viewpoint.

[0006] Based on the method in the previous step, obtain the point cloud of all measurement viewpoints of the power supply chassis structure. Use the random markers on the measurement turntable to stitch together the point cloud of all measurement viewpoints to obtain the complete point cloud containing the power supply chassis structure. Then, perform noise reduction processing on the complete point cloud data containing the power supply chassis structure to obtain the measurement point cloud of the power supply chassis structure.

[0007] The model point cloud is obtained from the CAD model of the power supply chassis structure. The measured point cloud is registered with the model point cloud to unify the measured point cloud and the model point cloud into the same coordinate system. Then, based on the shape features extracted from the CAD model of the power supply chassis structure, the local measured point cloud corresponding to the shape feature is extracted from the measured point cloud. The local measured point cloud is fitted to obtain geometric parameters. The measured size information of the shape feature is calculated based on the geometric parameters.

[0008] Furthermore, after multi-brightness stripe projection and synthesis, the 3D vision measurement system acquires a single-viewpoint point cloud of the power supply chassis structural component from a single measurement viewpoint, specifically including:

[0009] Multiple sets of phase-shifted fringe patterns are constructed by using multiple three-frequency four-step phase-shifted fringe patterns under different fringe projection brightness. These multiple sets of phase-shifted fringe patterns are projected from dark to bright onto the surface of the power supply chassis structural component. The dual cameras of the three-dimensional vision measurement system synchronously acquire the fringe patterns modulated by the power supply chassis structural component, thereby generating multiple sets of mask images.

[0010] At each brightness level, the number of unsaturated pixels of the same pixel in the mask image in the multiple sets of amplitude phase shift fringe patterns is counted. When the number exceeds the saturation threshold, the corresponding mask image is valid; otherwise, it is invalid.

[0011] For the pixels of the effective mask image, only the mask image pixels corresponding to the brightness level with the highest modulation are retained as the final mask image, and the phase-shifted stripe image is synthesized using the final mask image;

[0012] The phase-shifted fringe image is processed to obtain a single-viewpoint point cloud.

[0013] Furthermore, the step of obtaining the point cloud of all measurement viewpoints of the power supply chassis structural components according to the method in the previous step specifically includes:

[0014] Set the measurement viewpoint of the 3D vision measurement system so that the working distance of the 3D vision measurement system is within the depth of field of the dual cameras, and ensure that at least 3 pairs of marker points are captured simultaneously within the field of view of the dual cameras, and that the centers of the 3 pairs of marker points are not collinear. The dual cameras measure the point cloud data of the top surface of the power supply chassis structure from above, and then supplement the measurement of the point cloud data of the side surface of the power supply chassis structure from the four directions of front, back, left and right.

[0015] Furthermore, the step of stitching together the point clouds of all measurement viewpoints using randomly positioned markers on the measurement turntable to obtain a complete point cloud including the power supply chassis structural components specifically includes:

[0016] Step 1: Extract the center of the marker point in the acquired image and perform epipolar matching. Calculate the spatial position of the center of the marker point. For the two sets of point clouds acquired from two adjacent measurement viewpoints, use the least squares method to solve the rigid transformation matrix between the two sets of point clouds to obtain the stitching result of the two sets of adjacent point cloud data.

[0017] Step 2: Following the method in Step 1, continuously stitch the point clouds of two adjacent viewpoints together based on marker points, and use the ICP algorithm for further registration until all viewpoint point clouds are stitched together; then apply Gaussian filtering to the point clouds in the stitched overlapping area to obtain a complete point cloud including the power supply chassis structure.

[0018] Furthermore, the Gaussian filtering of the point cloud in the stitched overlapping region specifically includes:

[0019] Set the voxel grid side length, divide the point cloud data into several voxel grids of the same size for all single measurement viewpoints, calculate the centroid of the point cloud in each voxel grid, and replace the points in the voxel grid with the centroid.

[0020] Iterate through the point cloud of all measured viewpoints and calculate the distance d between each point and its k nearest neighbors. i Let i = 1, 2, ..., k, where the subscript i is the index of the k nearest neighbor. Calculate the average distance. and d i Standard deviation std, remove All nearest neighbors, where t is an adjustable parameter.

[0021] Furthermore, the step of denoising the complete point cloud data containing the power supply chassis structural component to obtain the measurement point cloud of the power supply chassis structural component specifically includes:

[0022] Principal component analysis is used to obtain the three principal directions of the complete point cloud containing the power supply chassis structural components. The normal vector perpendicular to the turntable plane is determined to define the turntable plane. The point cloud of the turntable plane is removed according to the dimensions of the power supply chassis structure to finally obtain the measurement point cloud of the power supply chassis structural components.

[0023] Furthermore, obtaining the model point cloud based on the CAD model of the power supply chassis structural components specifically includes:

[0024] The CAD model is converted into STL format to obtain information about the triangular facets on the surface of the CAD model. Each triangular facet is randomly upsampled to fill the point cloud data, and finally the CAD model is discretized into a model point cloud.

[0025] Furthermore, the registration of the measured point cloud and the model point cloud unifies them into the same coordinate system. Then, based on the shape features extracted from the CAD model of the power supply chassis structural component, the corresponding local measured point cloud is extracted from the measured point cloud. The local measured point cloud is fitted to obtain geometric parameters, and the measured dimensional information of the shape feature is calculated based on the geometric parameters. Specifically, this includes:

[0026] Based on the FRG algorithm, the measured point cloud of the power supply chassis structural component is registered with the model point cloud for the first time, so that the measured point cloud and the model point cloud are unified under the same coordinate system.

[0027] The measurement point cloud and the model point cloud are registered a second time based on the ICP algorithm;

[0028] Based on the shape feature type of the CAD model, the point cloud of the shape feature is segmented to obtain the local point cloud corresponding to the shape feature, and the local point cloud is fitted to obtain the geometric dimension parameters of the shape feature.

[0029] Based on the spatial location of the shape features extracted from the CAD model, the corresponding shape features are located on the measurement point cloud. Then, a pass-through filter is used to remove the point cloud outside the shape features to segment the local measurement point cloud of the corresponding shape features.

[0030] The local measurement point cloud of the shape feature is first denoised, and then the geometric parameters of the shape feature are obtained by fitting based on the random sampling consensus algorithm and the global least squares algorithm. The measured size information of the shape feature is calculated based on the geometric parameters.

[0031] Furthermore, the method further includes: comparing the measured dimensional information of the shape features with the design dimensions extracted from the design model, determining whether the processing is qualified, and providing a report.

[0032] Furthermore, the power supply chassis can be replaced with other on-board equipment.

[0033] This invention relates to an automatic measurement method for the dimensions of power supply chassis structural components based on full-field 3D vision. It performs 3D dimension detection on the power supply chassis structural components based on the input 3D point cloud model and labeled dimensions, and generates a detection report, thereby improving detection efficiency and accuracy. This provides a strong guarantee for the processing quality of power supply chassis structural components in spacecraft electronic systems. Furthermore, this method can also be applied to the measurement of the dimensions of payloads for onboard computers, onboard communication equipment, and other onboard devices, demonstrating good versatility. Attached Figure Description

[0034] Figure 1 A schematic diagram of the automatic measurement method for scanning power supply chassis structural components based on three-dimensional vision provided by the present invention;

[0035] Figure 2 A schematic diagram of a phase-shifting fringe projection measurement system based on binocular stereo vision;

[0036] Figure 3 A schematic diagram of the automatic measurement system for scanning power supply chassis structural components based on three-dimensional vision provided by the present invention;

[0037] Figure 4 A flowchart illustrating the registration process between measured point clouds and model point clouds;

[0038] Figure 5 The results are as follows: (a) is the point cloud of the complete part, (b) is the feature associated with distance dimension 1 between the two planes, and (c) is the feature associated with distance dimension 2 between the two planes.

[0039] Figure 6 The results of segmentation of the dimensional features of industrial parts are shown in (a) and (b) respectively. (a) is the CAD model with dimensions marked, and (b) is the point cloud with the segmented features.

[0040] Figure 7 The results are the feature fitting results for point clouds. (a) is the feature fitting result associated with distance dimension 1 between two planes, and (b) is the feature fitting result associated with distance dimension 2 between two planes. Detailed Implementation

[0041] To make the objectives and technical solutions of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0042] This invention proposes an automatic measurement method for the dimensions of power supply chassis structural components based on full-field 3D vision. Onboard computers, onboard communication equipment, and other payloads typically have a regular, cube-like shape. The method of this invention can also be applied to measuring the dimensions of payloads for onboard computers, onboard communication equipment, and other similar devices, demonstrating good versatility. Based on this invention, automatic scanning and dimensional measurement of such onboard equipment structural components can be achieved. The following detailed explanation uses an onboard power supply chassis structural component as an example to illustrate the real-time method of this invention.

[0043] This invention provides an automatic measurement method for the dimensions of power supply chassis structural components based on full-field 3D vision, such as... Figure 1 As shown, it includes:

[0044] The automatic measurement method of this invention is implemented using a three-dimensional vision measurement system, as described in the reference. Figure 2-3 The 3D vision measurement system consists of a turntable, a robotic arm, a phase-matching full-field measurement sensor, and a computer.

[0045] Specifically, the upper surface of the turntable has a dedicated flexible fixture, with randomly distributed circular markers affixed to it. A phase-matching full-field measurement sensor, driven by a robotic arm, rotates in conjunction with the turntable to acquire a 3D point cloud of the power supply chassis structural components. A computer is used to store the data and perform control and computational functions.

[0046] The phase-matched full-field measurement sensor consists of dual cameras, a stripe projector, a cooling fan, a data interface, and a laser projector. The camera and stripe projector are selected based on factors such as the size of the object being measured, camera field of view, and volume. The working distance is determined by the measurement range and the projector's field of view angle. The maximum size of the object being measured can be 500mm × 500mm. Since single-field-of-view measurement places high demands on the projector and probe baseline, and increasing the field of view increases the dot pitch, affecting measurement accuracy, multi-field-of-view measurement is adopted. A measurement range of 260mm × 220mm is recommended for each field of view. To cover the field of view, a medium-sized short-throw projector with a relatively large field of view is selected. The working distance of the sensor is calculated based on the field of view angle and the lateral range of the field of view, resulting in 500mm. Then, the camera resolution is selected according to the measurement requirements. When the object being measured has significant curvature variations and many step edges, a high-resolution camera can ensure the accuracy of detailed measurements; a 2448×2048 resolution camera is selected. Secondly, the lens is selected based on the camera parameters. The lens focal length can be calculated based on the camera target size, sensor working distance, and measurement range, with a 16mm focal length being preferred. Simultaneously, the lens's optical resolution must be greater than the camera's resolution. Finally, simulations are performed based on the system parameters to calculate the random measurement error and determine if the measurement accuracy requirements are met. A cooling fan is used to cool the camera and stripe projector, suppressing temperature drift; a data interface is used for data transmission between the sensor and the computer; and a laser projector is used to manually assist in determining whether the sensor is at the designed working distance. When the laser points on the surface of the measured object coincide, that surface position is considered to be at the designed working distance.

[0047] The implementation process of the automatic measurement method for the dimensions of power supply chassis structural components based on full-field 3D vision using a 3D vision measurement system is as follows:

[0048] S1. Based on phase-matching 3D vision measurement, local point clouds of power supply chassis structural components are obtained. Using the phase-shifting fringe projection measurement principle, and through multi-brightness fringe projection and synthesis technology, a single-field point cloud of the highly reflective surface of the power supply chassis structural components within a measurement angle field of view of the sensor is obtained.

[0049] Specifically, the phase-matched full-field measurement sensor projects phase-shifted sinusoidal fringes onto the object under test using a fringe projector. Two cameras capture the fringes, and after phase unwrapping and phase decomposition, the three-dimensional shape of the object is calculated using the structured light triangulation principle or binocular stereo vision principle. This sensor utilizes the phase-shifted fringe projection measurement principle, combined with multi-brightness synthesis technology, to achieve precise single-field-of-view measurement of highly reflective surfaces of power supply chassis structural components. The specific process is as follows:

[0050] S11. A three-frequency, four-step phase-shift fringe pattern is adopted, with fringe periods of 15, 16, and 17 pixels respectively. Four phase-shift fringe patterns with a step size of π / 2 are projected in each period. The fringe pattern can be represented as:

[0051]

[0052] Among them, I i (x,y) represents the i-th phase-shifted fringe pattern, (x,y) represents the coordinates in the camera image plane coordinate system, I′(x,y) represents the average brightness of the background light, and I″(x,y) represents the modulation degree of the fringe pattern. This indicates the wrap-around phase.

[0053] S12. Project the phase-shifted fringe pattern onto the surface of the test object. For the surface of the power supply chassis structural component with different reflectivity, design a total of 7 sets of three-frequency four-step phase-shifted fringe patterns with different projection brightness. The fringe projector projects the 7 sets of phase-shifted fringe patterns from dark to bright onto the surface of the power supply chassis structural component. The dual cameras synchronously acquire the fringe patterns modulated by the surface of the test object.

[0054] S13. Generate a set of mask images M k (x,y), count the number of unsaturated pixels of the same pixel (x,y) in N phase-shifted fringe patterns at each brightness level, denoted as m. k (x, y), assign values ​​to the mask image:

[0055]

[0056] A mask value of 1 indicates that the mask is valid, while a value of 0 indicates that the mask is invalid.

[0057] Calculate the tonal value of the stripe pattern:

[0058]

[0059] Where k is the brightness level number, I" k (x,y) represents the modulation depth at brightness level k, and i is the phase shift number, which ranges from 0 to 3. Let (x,y) represent the fringe pattern brightness acquired at brightness level k, and (x,y) represent the coordinates in the camera image plane coordinate system.

[0060] For pixels with more than 1 valid mask value, only the mask value corresponding to the brightness level with the highest modulation is retained as valid, and the mask values ​​for other brightness levels are set to 0.

[0061]

[0062] The effective mask image pixels are used as the final mask image. Finally, a set of high dynamic range phase-shift stripe images is synthesized using the mask image.

[0063]

[0064] S14. Based on the synthesized high dynamic phase-shifted fringe image, perform phase principal value solving, multi-frequency heterodyne phase expansion, epipolar correction, phase matching, and three-dimensional reconstruction to obtain a single-viewpoint cloud.

[0065] S2. Based on the principle of acquiring single-viewpoint point clouds, a rotating turntable, in conjunction with a robotic arm path planning system, drives the probe to measure power supply chassis structural components at different angles, covering all surfaces of the power supply chassis structural components. The measurements are then stitched together based on marker points on the turntable to obtain a complete, dense point cloud, such as... Figure 3 As shown, point cloud data preprocessing is performed to address noise and invalid points. The specific steps are as follows:

[0066] S21. The camera's measurement viewpoint is set based on the dimensions of the power supply chassis structure. The camera's working distance must be within its depth of field to capture clear images of the marker points and achieve higher accuracy in extracting the center coordinates. Since the power supply chassis structure obstructs the marker points on the turntable, at least three pairs of marker points must be captured simultaneously within the field of view of both cameras, and the center coordinates of these three pairs of marker points must not be collinear. Coordinating with the turntable and robotic arm movements, the two cameras measure the point cloud data of the top surface of the power supply chassis structure from above, and then supplement the measurement with point cloud data from the sides of the power supply chassis structure from the front, back, left, and right.

[0067] S22. The markers are fixedly pasted on the turntable, so their relative spatial positions remain unchanged. Based on this characteristic, the center of the markers in the dual-camera images is first extracted and epipolar matching is performed. The spatial position of the center of the markers is calculated. For the two sets of point clouds acquired from two adjacent measurement viewpoints, the rigid transformation matrix between the two sets of point clouds is solved using the least squares method based on the extracted center coordinates of the markers. This yields a coarse stitching result of the two sets of adjacent point cloud data, making the two sets of point clouds roughly unified under the same coordinate system.

[0068] S23. Perform coarse stitching of point clouds from two adjacent viewpoints based on marker points, and use the ICP algorithm for precise registration until the stitching of all viewpoint point clouds is completed.

[0069] S24. To address the problem of excessively dense and cluttered point clouds in overlapping regions, the voxel mesh method is used to eliminate overlapping points while preserving geometric features. Gaussian filtering is employed to filter the point clouds for noise in the original point cloud data.

[0070] Specifically, the voxel grid method sets the side length of the voxel grid, divides the point cloud data into several voxel grids of the same size, calculates and retains the centroid of the point cloud data in the grid, and replaces the data points in the grid with the centroid, thus achieving the effect of simplifying the point cloud data. Specifically, the K-neighborhood method is used to traverse the point clouds of all measurement viewpoints, calculate the distance d between each point and its k nearest neighbors in its k-neighborhood. i(i = 1, 2, ..., k, where the index i is the index of the k nearest neighbor), average distance and d i Standard deviation std, remove All nearest neighbors, where t is an adjustable parameter.

[0071] S25. To address irrelevant point cloud noise from the turntable, principal component analysis (PCA) is used to obtain the three principal directions and centroid of the point cloud. In the 3D point cloud, the eigenvalues ​​of the covariance matrix represent the amplitude of the largest variance in the data change, and the eigenvectors represent the direction of the largest variance. The direction perpendicular to the turntable plane is the direction of the eigenvector corresponding to the largest eigenvalue. A single-point random sampling consensus algorithm is used to determine a plane as the turntable plane by randomly selecting a point from the measured point cloud and the normal vector of the turntable plane. This removes the point cloud of the turntable plane to obtain the point cloud of the measured object, thus obtaining the measurement point cloud of the power supply chassis structural components. To further remove irrelevant point cloud noise from the complete point cloud data, based on the single-point random sampling consensus algorithm, a plane is determined by randomly selecting a point and the normal vector from the point clouds of all measurement viewpoints. For multiple randomly selected points, multiple corresponding planes are calculated, and the distance judgment threshold θ of each plane is obtained. h Points within the plane are used as the turntable plane point cloud. Since the turntable plane has the most points and its dimensions are known, the turntable plane point cloud can be determined based on the number of points extracted from multiple planes and the plane dimensions. The turntable plane point cloud is then removed to obtain the point cloud of the measured object, thus obtaining the measurement point cloud of the power supply chassis structural components. The threshold θ is used for judgment. h It can be adapted to the actual application scenarios and experience.

[0072] S3. Registration of measurement point cloud and model point cloud based on FRG algorithm and ICP algorithm, the process is as follows: Figure 4 As shown, point cloud segmentation, feature fitting, and size calculation are performed based on size-related features. The results are then compared and identified with the size information extracted from the chassis CAD model to achieve size measurement.

[0073] S31. Convert the CAD model into STL format, obtain the information of the triangular facets on the model surface, randomly upsample each triangular facet to fill the point cloud data of the triangular facets, and finally discretize the STL model into a point cloud.

[0074] S32. Based on the FRG algorithm, the measured point cloud of the test part and the point cloud of the CAD model are coarsely registered for the first time, so that the point cloud of the model under different coordinate systems and the measured point cloud are roughly unified under the same coordinate system, providing a good initial position for subsequent ICP fine registration and ensuring the effectiveness of the fine registration algorithm.

[0075] S33. Based on the ICP algorithm, the measured point cloud and the CAD model are precisely registered a second time to improve the registration accuracy and achieve the best fit.

[0076] S34. Based on the dimension-related feature type (e.g., plane, cylinder, etc.) provided on the CAD model, segment the point cloud near the feature to obtain the local point cloud corresponding to the feature, and fit the local point cloud to obtain the geometric dimension parameters of the feature.

[0077] The dimensions extracted from the CAD model include linear dimensions, radius dimensions, and diameter dimensions. Linear dimensions can be calculated from point-to-surface distances and point-to-line distances, while radius and diameter dimensions can be directly read from the fitted equations.

[0078] Among them, point cloud segmentation based on size-related features quickly locates the positions of size-related features on the measured point cloud based on the spatial location information of size-related feature information extracted from the 3D CAD model. Then, a pass-through filter is used to quickly remove the parts outside the features, achieving the purpose of feature separation. Figure 5 As shown. After segmenting the complete measured point cloud, the segmentation result is as follows. Figure 6 As shown.

[0079] The fitting of point cloud surface feature parameters first involves denoising to reduce errors, followed by feature fitting based on Random Sample Consensus (RANSAC) and Total Least Squares (TLS). RANSAC discards some invalid points, using data that best meets the requirements to find a model for fitting, thus denoising the point cloud and retaining valid data. TLS is an improvement on Least Squares, considering interference factors in the coefficient matrix during fitting and taking into account errors on the x, y, and z axes. This results in a more accurate fitting than Least Squares. The fitting results are shown below. Figure 7 As shown.

[0080] S35. Based on geometric parameters, calculate the corresponding measured dimension information in the measured point cloud, compare it with the design dimension information extracted from the CAD design model, determine whether the processing is qualified, and output a report.

Claims

1. An automatic measurement method for the dimensions of power supply chassis structural components based on full-field 3D vision, characterized in that, include: After multi-brightness stripe projection and synthesis, the 3D vision measurement system acquires a single measurement viewpoint point cloud of a power supply chassis structural component from a single measurement viewpoint. Specifically, this includes: constructing multiple sets of phase-shifted fringe patterns using multiple three-frequency four-step phase-shifted fringe patterns at different fringe projection brightness levels; projecting these multiple sets of phase-shifted fringe patterns from dark to bright onto the surface of a power supply chassis structural component; synchronously acquiring the fringe patterns modulated by the power supply chassis structural component using dual cameras of a three-dimensional vision measurement system, thereby generating multiple sets of mask images; at each brightness level, counting the number of unsaturated pixels for the same pixel in the mask image within the multiple sets of phase-shifted fringe patterns; if the number of pixels exceeds a saturation threshold, the corresponding mask image is valid; otherwise, it is invalid; for the pixels in the valid mask images, only the pixels corresponding to the brightness level with the highest modulation intensity are retained as the final mask image; the final mask image is used to synthesize a phase-shifted fringe image; and the phase-shifted fringe image is processed to obtain a single measurement viewpoint point cloud. Based on the method in the previous step, all measurement viewpoint point clouds of the power supply chassis structural component are obtained. Then, random marker points on the measurement turntable are used to stitch together all measurement viewpoint point clouds to obtain a complete point cloud containing the power supply chassis structural component. This complete point cloud data containing the power supply chassis structural component is then denoised to obtain the measurement point cloud of the power supply chassis structural component. Specifically, the step of stitching together all measurement viewpoint point clouds using random marker points on the measurement turntable to obtain a complete point cloud containing the power supply chassis structural component includes: Step 1: Extracting the center of the marker points in the acquired image and performing epipolar matching; calculating the spatial position of the marker point center; and for two sets of point clouds acquired from two adjacent measurement viewpoints, using the least squares method to solve for the rigid transformation matrix between the two sets of point clouds to obtain... The stitching result of two sets of adjacent point cloud data; Step 2: According to the method in Step 1, the point clouds of two adjacent viewpoints are continuously stitched based on marker points, and further registered using the ICP algorithm until the stitching of all viewpoint point clouds is completed; and Gaussian filtering is applied to the point cloud of the stitched overlapping area to obtain a complete point cloud containing the power supply chassis structure; The step of denoising the complete point cloud data containing the power supply chassis structure to obtain the measurement point cloud of the power supply chassis structure specifically includes: using principal component analysis to obtain the three principal directions of the complete point cloud containing the power supply chassis structure, determining the normal vector perpendicular to the turntable plane to determine the turntable plane; removing the turntable plane point cloud according to the size of the power supply chassis structure, and finally obtaining the measurement point cloud of the power supply chassis structure. The model point cloud is obtained from the CAD model of the power supply chassis structure. The measured point cloud is registered with the model point cloud to unify the measured point cloud and the model point cloud into the same coordinate system. Then, based on the shape features extracted from the CAD model of the power supply chassis structure, the local measured point cloud corresponding to the shape feature is extracted from the measured point cloud. The local measured point cloud is fitted to obtain geometric parameters. The measured size information of the shape feature is calculated based on the geometric parameters.

2. The method according to claim 1, characterized in that, The step of obtaining the point cloud of all measurement viewpoints for the power supply chassis structural components according to the method in the previous step specifically includes: Set the measurement viewpoint of the 3D vision measurement system so that the working distance of the 3D vision measurement system is within the depth of field of the dual cameras, and ensure that at least 3 pairs of marker points are captured simultaneously within the field of view of the dual cameras, and that the centers of the 3 pairs of marker points are not collinear. The dual cameras measure the point cloud data of the top surface of the power supply chassis structure from above, and then supplement the measurement of the point cloud data of the side surface of the power supply chassis structure from the four directions of front, back, left and right.

3. The method according to claim 1, characterized in that, The Gaussian filtering of the point cloud in the spliced ​​overlapping region specifically includes: Set the voxel grid side length, divide the point cloud data into several voxel grids of the same size for all single measurement viewpoints, calculate the centroid of the point cloud in each voxel grid, and replace the points in the voxel grid with the centroid. Iterate through the point cloud of all measured viewpoints and calculate the distance between each point and its k nearest neighbors. Footnote i It is the index of the k nearest neighbor, calculate the average distance. and Standard deviation Remove All nearest neighbors, where t is an adjustable parameter.

4. The method according to claim 1, characterized in that, The process of obtaining the model point cloud based on the CAD model of the power supply chassis structural components specifically includes: The CAD model is converted into STL format to obtain information about the triangular facets on the surface of the CAD model. Each triangular facet is randomly upsampled to fill the point cloud data, and finally the CAD model is discretized into a model point cloud.

5. The method according to claim 4, characterized in that, The process of registering the measured point cloud with the model point cloud unifies them into the same coordinate system. Then, based on the shape features extracted from the CAD model of the power supply chassis structure, local measured point clouds corresponding to these shape features are extracted from the measured point cloud. Geometric parameters are obtained by fitting these local measured point clouds, and the measured dimensions of the shape features are calculated based on these geometric parameters. Specifically, this includes: Based on the FRG algorithm, the measured point cloud of the power supply chassis structural component is registered with the model point cloud for the first time, so that the measured point cloud and the model point cloud are unified under the same coordinate system. The measurement point cloud and the model point cloud are registered a second time based on the ICP algorithm; Based on the shape feature type of the CAD model, the point cloud of the shape feature is segmented to obtain the local point cloud corresponding to the shape feature, and the local point cloud is fitted to obtain the geometric dimension parameters of the shape feature. Based on the spatial location of the shape features extracted from the CAD model, the corresponding shape features are located on the measurement point cloud. Then, a pass-through filter is used to remove the point cloud outside the shape features to segment the local measurement point cloud of the corresponding shape features. The local measurement point cloud of the shape feature is first denoised, and then the geometric parameters of the shape feature are obtained by fitting based on the random sampling consensus algorithm and the global least squares algorithm. The measured size information of the shape feature is calculated based on the geometric parameters.

6. The method according to claim 1, characterized in that, The method further includes: comparing the measured dimensional information of the shape features with the design dimensions extracted from the design model, determining whether the processing is qualified, and giving a report.

7. The method according to any one of claims 1-6, characterized in that, The power supply chassis can be replaced with other on-board equipment.

Images