A high dynamic range real-time composite three-dimensional thickness detection method

Abstract

The present application relates to a kind of high dynamic range real-time composite three-dimensional thickness detection method, belong to industrial detection technical field.Aiming at the defects of traditional thickness measurement method relying on manual operation, edge detail loss, insufficient precision and weak adaptability to different specifications workpiece, the present application proposes a detection scheme of two-dimensional and three-dimensional collaborative fusion: through structured light projection encoding image, extract two-dimensional edge feature and integrate into three-dimensional reconstruction process, extract edge after three-dimensional data segmentation processing, project to two-dimensional plane, complete thickness measurement after smoothing, extreme detection and distance calculation.This method does not need physical contact workpiece, can realize the automatic accurate measurement of complex curved surface ripple workpiece, effectively solve the problem of low efficiency and poor adaptability of traditional method, adapt to different specifications workpiece detection needs, provide efficient and reliable quality inspection technical support for industrial mass production.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional vision inspection and precision measurement technology, specifically relating to a high dynamic range real-time composite three-dimensional thickness detection method, which is particularly suitable for high-precision and high-efficiency three-dimensional thickness measurement of complex corrugated workpieces. Background Technology

[0002] In the industrial manufacturing field, accurately acquiring the wall thickness data of complex corrugated workpieces is a key prerequisite for ensuring the reliability of equipment operation. Traditional measurement methods are divided into two categories: contact and non-contact. The former, represented by coordinate measuring machines and mechanical probes, is prone to scratching the workpiece surface or causing deformation, and has low measurement efficiency, making it difficult to adapt to batch inspection needs. The latter includes electromagnetic methods, optical interferometry, etc., but faces bottlenecks such as loss of edge details, insufficient measurement accuracy, and weak adaptability to working conditions, and cannot meet the accurate inspection needs of complex corrugated structures. To address this, this invention designs a three-dimensional fusion detection method. Through core technologies such as structured light projection, two-dimensional-three-dimensional collaborative constraints, point cloud processing, and extreme value detection, it realizes automated and accurate measurement of the thickness of complex corrugated workpieces, effectively solving the core problems of low accuracy, poor efficiency, and insufficient adaptability of traditional methods. Summary of the Invention

[0003] This invention presents a high dynamic range real-time composite three-dimensional thickness detection method, which can be applied to the rapid and accurate thickness detection of various complex corrugated workpieces. This method overcomes the problems of traditional contact measurement, which is prone to damaging the workpiece and has low efficiency, as well as the loss of edge details and insufficient accuracy of traditional non-contact measurement.

[0004] One DLP projector for projecting striped images; One industrial camera for acquiring images of distorted workpieces; One computer for data processing; A platform for mounting industrial cameras and projectors; One testing platform for placing the workpiece to be tested; This invention presents a high dynamic range real-time composite three-dimensional thickness detection method, the steps of which are as follows: Step 1: Place the complex curved corrugated workpiece to be tested on the inspection table and acquire a stripe-free image of the workpiece using an industrial camera; process the acquired image to extract the edge information of the workpiece; use a combination of adaptive gradient threshold and connected component constraint to filter out the effective pixel set of the workpiece edge, as shown in formula (1): Formula (1) Among them, let For pixel coordinates, For stripeless images Pixel gradient magnitude reflects the degree of change in pixel brightness. For adaptive gradient thresholding, Connectivity constraints are used to remove noise points. For pixel validity marking function, This is the final set of valid pixels; subsequent calculations will only be performed on this set. It is a two-dimensional integer space, representing the set of coordinates of all pixels in the image.

[0005] Step 2: Based on the effective pixel set selected in Step 1, the structured light 3D reconstruction technology is used to achieve accurate mapping from the 2D image to the 3D point cloud through structured light encoding and phase calculation; a sinusoidal fringe pattern with encoded information is projected onto the workpiece surface through a DLP projector, and the projected fringe image model is shown in formula (2): Formula (2) in, It is the first The grayscale values ​​of the pixels in the striped image with phase shift. It is the background grayscale value of the image. It is the modulation gray value of the encoded light. Let be the principal value of the phase to be determined.

[0006] While the projector projects the stripes, the industrial camera in step 1 simultaneously acquires the distorted image of the modulated workpiece; for the effective pixel set obtained in step 1, the absolute phase of each point is solved by the multi-frequency heterodyne method, and the wrapping phase is first smoothed by formula (3): Formula (3) in, To wrap the phase, n Represents the number of phase shift steps. Let the phase expansion order be , It is an absolute phase.

[0007] To eliminate phase ambiguity, the multi-frequency heterodyne method is used to solve for the phase expansion order: for wavelength and The two sets of fringes were used to obtain their enclosing phase using the phase-shifting method. and The equivalent wavelength is calculated using formula (4), and the phase difference is solved based on the equivalent wavelength. .

[0008] Formula (4) After phase expansion is completed by three-frequency interference fringes, the three-dimensional coordinates of each point of the object can be solved by combining epipolar constraints and triangulation principles, as shown in formulas (5) and (6), thus realizing the accurate mapping of the two-dimensional image of the workpiece to the three-dimensional point cloud.

[0009] Formula (5) Formula (6) in, This represents the x-coordinate of the corresponding point in the projector's pixel coordinate system. For camera pixels The absolute phase value, The period of the vertical stripes, This refers to the horizontal resolution of the projector. These are the camera's scaling factor, internal parameters, and external parameters, respectively. These are the projector's scaling factor, internal parameters, and external parameters, respectively; by solving equations (5) and (6), the three-dimensional coordinates of the unique world coordinate system can be obtained. .

[0010] Step 3: Based on the complete 3D point cloud generated in Step 2, select the key cross-sections for thickness measurement and extract the cross-section edge information. The specific operation is as follows: use the algorithm to segment the 3D point cloud and extract the point cloud plane of the target cross-section. The validity judgment of the segmentation process is shown in Formula (7).

[0011] Formula (7) in, To determine the minimum number of iterations required, round up. The algorithm confidence score represents the probability of finding the optimal model. The proportion of outliers in the data. The minimum number of samples required to fit the target geometric model.

[0012] The method for determining whether a point is an interior point is formula (8): Formula (8) in, For the first Points to the fitting plane Euclidean distance; The distance threshold is determined by the data noise level or point cloud resolution. For a point in space, It is an interior point set.

[0013] Let the filtered planar point cloud be ,in For any target point, the total number of points in the cloud is given. , Let be a three-dimensional real space, representing the three-dimensional coordinate space of the point cloud; its K-nearest neighbor set , KThe preset number of nearest neighbors is used to construct the local neighborhood of the target point. The point cloud boundary points are determined based on the difference in the distribution of normal vectors between the neighborhoods of points and edge points in the plane, and edge point clouds are filtered to select the target point. unit vector and neighboring points point to The direction vector of the normal vector The spatial angle is given by formula (9).

[0014] Formula (9) in, The dot product of two vectors. Its value quantifies the degree of spatial matching between the normal vector and the pointing direction of the neighborhood.

[0015] Step 4: Denoise the edge point cloud acquired in Step 3 by designing filters and local smoothing, as shown in Formula (10) and Formula (11), orthogonally project the three-dimensional edge point cloud onto the two-dimensional plane and sort it to form an ordered point set, and perform Gaussian smoothing preprocessing with physical window constraints.

[0016] Formula (10) Formula (11) in, The joint weighting coefficients for the target point and its nearest neighbors. For angle threshold, For regularization terms, The target point is the Euclidean distance between its nearest neighbors, ensuring that the nearest points have a higher weight.

[0017] Formula (12) Formula (13) in, A one-dimensional Gaussian kernel function with physical window constraints. It is the first The smoothed Y-coordinate of each point For the first The original coordinates of each point Therefore Centered on The original coordinates of all points within the neighborhood window of the range. , This is an indicator function; it takes the value 1 if the condition within the parentheses is met, and 0 otherwise. It is the standard deviation of the Gaussian kernel that controls the smoothness. Implement Gaussian weights for neighboring points Effective constraints on physical windows.

[0018] Step 5: The algorithm is used to identify the peaks and valleys of the point cloud curve after step 4. Finally, the wall thickness of the complex corrugated workpiece is measured by calculating the Euclidean distance across the curve. The judgment rule is as shown in formula (14).

[0019] Formula (14) in, The maximum value of the Y-coordinate within the neighborhood. It is the minimum value of the Y-coordinate within the neighborhood; and These are the X-axis and Y-axis coordinates of the target point in the smoothed point set, respectively. This represents the magnitude difference of the Y-coordinate within the window; The preset minimum amplitude threshold is set if the target point Y value is the maximum value within the window, and the amplitude difference of the Y values ​​within the window is... Greater than the minimum amplitude threshold Then it is determined to be a peak. If the target point Y value is the minimum value within the window and the amplitude difference meets the threshold requirement, then it is determined to be a trough. .

[0020] Let the set of coordinates of the peaks and troughs of curve A be... , The set of coordinates of the peaks and troughs of curve B is , , This represents the total number of peaks on the curve. This represents the total number of troughs in the curve. s The corresponding point index is used; for the corresponding peak points and corresponding trough points that match the index in the two curves, the two-dimensional Euclidean distance calculation method is used, as shown in formula (15).

[0021] Formula (15) in, This refers to the measured wall thickness at the crest of the corrugated surface workpiece. The wall thickness at the trough of the corrugated workpiece is measured, and the thickness of the complex corrugated workpiece is calculated.

[0022] The beneficial effects of this invention are: to realize automated high-precision thickness measurement in the industrial batch quality inspection process of complex curved corrugated workpieces, non-contact detection, and to solve the problems of traditional measurement methods that rely on manual operation, loss of edge details, insufficient measurement accuracy and weak adaptability. Attached Figure Description

[0023] To more clearly illustrate the embodiments and technical solutions of the present invention, the accompanying drawings used will be briefly described below.

[0024] Figure 1 Flowchart for wall thickness inspection of complex corrugated workpieces; (a) Edge extraction from images of complex corrugated workpieces; (b) Mapping a two-dimensional image to a three-dimensional point cloud; (c) The target plane for segmenting the 3D point cloud of the workpiece; (d) Extract the edges of the planar point cloud; (e) Search for the extreme values ​​of the curve and calculate the thickness. Detailed Implementation

[0025] This invention presents a high dynamic range real-time composite three-dimensional thickness detection method, which can be applied to the rapid and accurate thickness detection of various complex corrugated workpieces. This method overcomes the problems of traditional contact measurement, which is prone to damaging the workpiece and has low efficiency, as well as the loss of edge details and insufficient accuracy of traditional non-contact measurement.

[0026] This invention presents a high dynamic range real-time composite three-dimensional thickness detection method, the steps of which are as follows: Step 1: Place the complex curved corrugated workpiece to be tested on the inspection table and acquire a stripe-free image of the workpiece using an industrial camera; process the acquired image to extract the edge information of the workpiece; use a combination of adaptive gradient threshold and connected component constraint to filter out the effective pixel set of the workpiece edge, as shown in formula (1): Formula (1) Among them, let For pixel coordinates, For stripeless images Pixel gradient magnitude reflects the degree of change in pixel brightness. For adaptive gradient thresholding, Connectivity constraints are used to remove noise points. For pixel validity marking function, This is the final set of valid pixels; subsequent calculations will only be performed on this set. It is a two-dimensional integer space, representing the set of coordinates of all pixels in the image.

[0027] Step 2: Based on the effective pixel set selected in Step 1, the structured light 3D reconstruction technology is used to achieve accurate mapping from the 2D image to the 3D point cloud through structured light encoding and phase calculation; a sinusoidal fringe pattern with encoded information is projected onto the workpiece surface through a DLP projector, and the projected fringe image model is shown in formula (2): Formula (2) in, It is the first The grayscale values ​​of the pixels in the striped image with phase shift. It is the background grayscale value of the image. It is the modulation gray value of the encoded light. Let be the principal value of the phase to be determined.

[0028] While the projector projects the stripes, the industrial camera in step 1 simultaneously acquires the distorted image of the modulated workpiece; for the effective pixel set obtained in step 1, the absolute phase of each point is solved by the multi-frequency heterodyne method, and the wrapping phase is first smoothed by formula (3): Formula (3) in, To wrap the phase, n Represents the number of phase shift steps. Let the phase expansion order be , It is an absolute phase.

[0029] To eliminate phase ambiguity, the multi-frequency heterodyne method is used to solve for the phase expansion order: for wavelength and The two sets of fringes were used to obtain their enclosing phase using the phase-shifting method. and The equivalent wavelength is calculated using formula (4), and the phase difference is solved based on the equivalent wavelength. .

[0030] Formula (4) After phase expansion is completed by three-frequency interference fringes, the three-dimensional coordinates of each point of the object can be solved by combining epipolar constraints and triangulation principles, as shown in formulas (5) and (6), thus realizing the accurate mapping of the two-dimensional image of the workpiece to the three-dimensional point cloud.

[0031] Formula (5) Formula (6) in, This represents the x-coordinate of the corresponding point in the projector's pixel coordinate system. For camera pixels The absolute phase value, The period of the vertical stripes, This refers to the horizontal resolution of the projector. These are the camera's scaling factor, internal parameters, and external parameters, respectively. These are the projector's scaling factor, internal parameters, and external parameters, respectively; by solving equations (5) and (6), the three-dimensional coordinates of the unique world coordinate system can be obtained. .

[0032] Step 3: Based on the complete 3D point cloud generated in Step 2, select the key cross-sections for thickness measurement and extract the cross-section edge information. The specific operation is as follows: use the algorithm to segment the 3D point cloud and extract the point cloud plane of the target cross-section. The validity judgment of the segmentation process is shown in Formula (7).

[0033] Formula (7) in, To determine the minimum number of iterations required, round up. The algorithm confidence score represents the probability of finding the optimal model. The proportion of outliers in the data. The minimum number of samples required to fit the target geometric model.

[0034] The method for determining whether a point is an interior point is formula (8): Formula (8) in, For the first Points to the fitting plane Euclidean distance; The distance threshold is determined by the data noise level or point cloud resolution. For a point in space, It is an interior point set.

[0035] Let the filtered planar point cloud be ,in For any target point, the total number of points in the cloud is given. , Let be a three-dimensional real space, representing the three-dimensional coordinate space of the point cloud; its K-nearest neighbor set , K The preset number of nearest neighbors is used to construct the local neighborhood of the target point. The point cloud boundary points are determined based on the difference in the distribution of normal vectors between the neighborhoods of points and edge points in the plane, and edge point clouds are filtered to select the target point. unit vector and neighboring points point to The direction vector of the normal vector The spatial angle is given by formula (9).

[0036] Formula (9) in, The dot product of two vectors. Its value quantifies the degree of spatial matching between the normal vector and the pointing direction of the neighborhood.

[0037] Step 4: Denoise the edge point cloud acquired in Step 3 by designing filters and local smoothing, as shown in Formula (10) and Formula (11), orthogonally project the three-dimensional edge point cloud onto the two-dimensional plane and sort it to form an ordered point set, and perform Gaussian smoothing preprocessing with physical window constraints.

[0038] Formula (10) Formula (11) in, The joint weighting coefficients for the target point and its nearest neighbors. For angle threshold, For regularization terms, The target point is the Euclidean distance between its nearest neighbors, ensuring that the nearest points have a higher weight.

[0039] Formula (12) Formula (13) in, A one-dimensional Gaussian kernel function with physical window constraints. It is the first The smoothed Y-coordinate of each point For the first The original coordinates of each point Therefore Centered on The original coordinates of all points within the neighborhood window of the range. , This is an indicator function; it takes the value 1 if the condition within the parentheses is met, and 0 otherwise. It is the standard deviation of the Gaussian kernel that controls the smoothness. Implement Gaussian weights for neighboring points Effective constraints on physical windows.

[0040] Step 5: The algorithm is used to identify the peaks and valleys of the point cloud curve after step 4. Finally, the wall thickness of the complex corrugated workpiece is measured by calculating the Euclidean distance across the curve. The judgment rule is as shown in formula (14).

[0041] Formula (14) in, The maximum value of the Y-coordinate within the neighborhood. It is the minimum value of the Y-coordinate within the neighborhood; and These are the X-axis and Y-axis coordinates of the target point in the smoothed point set, respectively. This represents the magnitude difference of the Y-coordinate within the window; The preset minimum amplitude threshold is set if the target point Y value is the maximum value within the window, and the amplitude difference of the Y values ​​within the window is... Greater than the minimum amplitude threshold Then it is determined to be a peak. If the target point Y value is the minimum value within the window and the amplitude difference meets the threshold requirement, then it is determined to be a trough. .

[0042] Let the set of coordinates of the peaks and troughs of curve A be... , The set of coordinates of the peaks and troughs of curve B is , , This represents the total number of peaks on the curve. This represents the total number of troughs in the curve. s The corresponding point index is used; for the corresponding peak points and corresponding trough points that match the index in the two curves, the two-dimensional Euclidean distance calculation method is used, as shown in formula (15).

[0043] Formula (15) in, This refers to the measured wall thickness at the crest of the corrugated surface workpiece. The wall thickness at the trough of the corrugated workpiece is measured, and the thickness of the complex corrugated workpiece is calculated.

[0044] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A high dynamic range real-time composite three-dimensional thickness detection method, characterized in that, Includes the following steps: Step 1: Place the complex curved corrugated workpiece to be tested on the inspection table and acquire a stripe-free image of the workpiece using an industrial camera; process the acquired image to extract the edge information of the workpiece; use a combination of adaptive gradient threshold and connected component constraint to filter out the effective pixel set of the workpiece edge, as shown in formula (1): Official (1) Among them, let For pixel coordinates, For stripeless images Pixel gradient magnitude reflects the degree of change in pixel brightness. For adaptive gradient thresholding, Connectivity constraints are used to remove noise points. For pixel validity marking function, This is the final set of valid pixels; subsequent calculations will only be performed on this set. It is a two-dimensional integer space, representing the set of coordinates of all pixels in the image; Step 2: Based on the effective pixel set selected in Step 1, the structured light 3D reconstruction technology is used to achieve accurate mapping from the 2D image to the 3D point cloud through structured light encoding and phase calculation; a sinusoidal fringe pattern with encoded information is projected onto the workpiece surface through a DLP projector, and the projected fringe image model is shown in formula (2): Official (2) in, It is the first The grayscale values ​​of the pixels in the striped image with phase shift. It is the background grayscale value of the image. It is the modulation gray value of the encoded light. Let be the principal value of the phase to be determined; While the projector projects the stripes, the industrial camera in step 1 simultaneously acquires the distorted image of the modulated workpiece; for the effective pixel set obtained in step 1, the absolute phase of each point is solved by the multi-frequency heterodyne method, and the wrapping phase is first smoothed by formula (3): Official (3) in, To wrap the phase, n Represents the number of phase shift steps. Let the phase expansion order be , It is an absolute phase; To eliminate phase ambiguity, the multi-frequency heterodyne method is used to solve for the phase expansion order: for wavelength and The two sets of fringes were used to obtain their enclosing phase using the phase-shifting method. and The equivalent wavelength is calculated using formula (4), and the phase difference is solved based on the equivalent wavelength. ; Official (4) After phase unwrapping is completed using the three-frequency interference fringes, the three-dimensional coordinates of each point on the object can be solved by combining the epipolar constraint and the triangulation principle, as shown in formulas (5) and (6), thus achieving accurate mapping from the two-dimensional image of the workpiece to the three-dimensional point cloud: Official (5) Official (6) in, This represents the x-coordinate of the corresponding point in the projector's pixel coordinate system. For camera pixels The absolute phase value, The period of the vertical stripes, This refers to the horizontal resolution of the projector. These are the camera's scaling factor, internal parameters, and external parameters, respectively. These are the projector's scaling factor, internal parameters, and external parameters, respectively; by solving equations (5) and (6), the three-dimensional coordinates of the unique world coordinate system can be obtained. . Step 3: Based on the complete 3D point cloud generated in Step 2, select the key cross-sections for thickness measurement and extract the edge information of the cross-sections. The specific operation is as follows: use an algorithm to segment the 3D point cloud and extract the point cloud plane of the target cross-section. The validity judgment of the segmentation process is shown in Formula (7): Official (7) in, To determine the minimum number of iterations required, round up. The algorithm confidence score represents the probability of finding the optimal model. The proportion of outliers in the data. The minimum number of samples required to fit the target geometric model; The method for determining whether a point is an interior point is formula (8): Official (8) in, For the first Points to the fitting plane Euclidean distance; The distance threshold is determined by the data noise level or point cloud resolution. For a point in space, It is an interior point set; Let the filtered planar point cloud be ,in For any target point, the total number of points in the cloud is given. , Let be a three-dimensional real space, representing the three-dimensional coordinate space of the point cloud; its K-nearest neighbor set , K The preset number of nearest neighbors is used to construct the local neighborhood of the target point. The point cloud boundary points are determined based on the difference in the distribution of normal vectors between the neighborhoods of points and edge points in the plane, and edge point clouds are filtered to select the target point. unit vector and neighboring points point to The direction vector of the normal vector The spatial angle is given by formula (9); Formula (9) in, The dot product of two vectors. Its value quantifies the degree of spatial matching between the normal vector and the direction of the neighborhood; Step 4: Denoise the edge point cloud acquired in Step 3 by designing filters and local smoothing, as shown in Formula (10) and Formula (11), orthogonally project the three-dimensional edge point cloud onto the two-dimensional plane and sort it to form an ordered point set, and perform Gaussian smoothing preprocessing with physical window constraints. Formula (10) Official (11) in, The joint weighting coefficients for the target point and its nearest neighbors. For angle threshold, For regularization terms, The target point is the Euclidean distance between its nearest neighbors, ensuring that the nearest points have a higher weight. Official (12) Official (13) in, A one-dimensional Gaussian kernel function with physical window constraints. It is the first The smoothed Y-coordinate of each point For the first The original coordinates of each point Therefore Centered on The original coordinates of all points within the neighborhood window of the range. , This is an indicator function; it takes the value 1 if the condition within the parentheses is met, and 0 otherwise. It is the standard deviation of the Gaussian kernel that controls the smoothness. Implement Gaussian weights for neighboring points Effective constraints on physical windows; Step 5: The algorithm is used to identify the peaks and valleys of the point cloud curve after Step 4. Finally, the wall thickness of the complex corrugated workpiece is measured by calculating the Euclidean distance across the curve. The judgment rule is as shown in formula (14). Official (14) in, The maximum value of the Y-coordinate within the neighborhood. It is the minimum value of the Y-coordinate within the neighborhood; and These are the X-axis and Y-axis coordinates of the target point in the smoothed point set, respectively. This represents the magnitude difference of the Y-coordinate within the window; The preset minimum amplitude threshold is set if the target point Y value is the maximum value within the window, and the amplitude difference of the Y values ​​within the window is... Greater than the minimum amplitude threshold Then it is determined to be a peak. If the target point Y value is the minimum value within the window and the amplitude difference meets the threshold requirement, then it is determined to be a trough. ; Let the set of coordinates of the peaks and troughs of curve A be... , The set of coordinates of the peaks and troughs of curve B is , , This represents the total number of peaks on the curve. This represents the total number of troughs in the curve. s The corresponding point index is used; for the corresponding peak points and corresponding trough points that match the index in the two curves, the two-dimensional Euclidean distance calculation method is used, as shown in formula (15); Official (15) in, This refers to the measured wall thickness at the crest of the corrugated surface workpiece. The wall thickness at the trough of the corrugated workpiece is measured, and the thickness of the complex corrugated workpiece is calculated. The operation ends.

Images