High-dynamic surface single-frame measurement method based on multi-view digital speckle correlation method

Abstract

The invention discloses a high-dynamic surface single-frame measurement method based on a multi-view digital speckle correlation method, and belongs to the technical field of multi-view digital speckles, and the method comprises the steps: constructing a speckle projection collection system based on a four-view camera and a photoetching speckle projection module, and collecting a single-frame image of a measured workpiece based on the speckle projection collection system; processing a single-frame image of the measured workpiece based on a multi-scale robust multi-view matching algorithm to obtain an intermediate depth map and a pixel-by-pixel space inclined plane; iterative calculation is carried out on the intermediate parameters of the space inclined plane pixel by pixel based on a digital speckle correlation method, and final parameters of the space inclined plane are obtained; and fusing the final depth map based on the final parameters of the space inclined plane and the left-right consistency superposition test to obtain final point cloud data. According to the invention, a workpiece with a high dynamic surface in a field of view can be quickly reconstructed by using a single-frame image, a high-integrity and high-precision point cloud is obtained, and the measurement precision can reach 0.03-0.07 mm.

Description

Technical Field

[0001] The invention belongs to the technical field of multi-eye digital speckle, and in particular relates to a high-dynamic surface single-frame measurement method based on a multi-eye digital speckle correlation method. Background Art

[0002] In the manufacturing process of sheet metal parts in fields such as aerospace, automotive, etc., it is necessary to perform accurate topography measurement. Despite the many advantages of structured light technology, accurately reconstructing 3D point clouds of high dynamic range surfaces remains a challenging problem. The reflection of high dynamic range surfaces is partially specular and partially diffuse. Specular reflections are only distributed in areas where the surface reflectivity is much higher than other areas. Due to the insufficient measurement range of the camera, the structured light pattern reflected from the high reflectivity position is too bright, which easily causes intensity saturation, while the structured light pattern reflected from the low reflectivity position is very dark and has a low signal-to-noise ratio. The intensity saturation and low signal-to-noise ratio problems seriously affect the accuracy of 3D reconstruction.

[0003] Many scholars have proposed multiple exposure techniques, which can obtain high-quality structured light patterns by changing the exposure time. However, the effectiveness of these techniques is highly dependent on the recognition accuracy of saturated pixels in the image. Many scholars have proposed adaptive fringe projection techniques to address this problem, usually by adjusting the maximum input grayscale to obtain high-quality images. However, the theoretical basis of this method is to assume that the measured surface is smooth, so this method cannot be used to measure geometrically complex objects with high dynamic range surfaces.

[0004] As mentioned above, specular reflections are only distributed in a specific area of ​​the camera viewpoint. Therefore, some scholars increase the number of viewpoints to 2-3, which can change the viewpoint of the camera to avoid capturing the specular reflection area in the image, such as measuring the object from at least three different directions or using a translation stage to adjust the camera's viewpoint. However, this technology cannot guarantee that all saturated areas in one camera view are completely unsaturated in another camera viewpoint. At the same time, the use of a translation stage method will greatly increase the measurement time. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method to solve the problems existing in the above prior art.

[0006] To achieve the above object, the present invention provides a high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method, comprising:

[0007] A speckle projection acquisition system is constructed based on a four-eye camera and a photolithography speckle projection module, and a single-frame image of the workpiece to be measured is acquired based on the speckle projection acquisition system;

[0008] Processing the single frame image of the workpiece to be measured based on a multi-scale robust multi-objective matching algorithm to obtain an intermediate depth map and a pixel-by-pixel spatial oblique plane;

[0009] Iteratively calculating the intermediate parameters of the spatial slant plane pixel by pixel based on the digital speckle correlation method to obtain the final parameters of the spatial slant plane;

[0010] The final depth map is fused based on the final parameters of the spatial inclined plane and the left-right consistency superposition test to obtain final point cloud data.

[0011] Preferably, the process of constructing a speckle projection acquisition system based on a multi-camera comprises:

[0012] Label the four cameras around the camera and install the speckle projection module at the symmetrical center of the camera;

[0013] The labeled quadrocopter is subjected to single-target and dual-target positioning, and then the internal and external parameters of the quadrocopter are optimized by using the bundle adjustment method to generate the speckle projection acquisition system.

[0014] Preferably, the process after the single frame image of the workpiece to be measured is collected by the speckle projection collection system further includes:

[0015] Performing distortion correction processing on the single-frame image of the measured workpiece to obtain a distortion-corrected image;

[0016] Processing the distortion-corrected image based on a median filtering algorithm to obtain a filtered denoised image;

[0017] The filtered image is processed based on a Gaussian filtering algorithm to obtain a preprocessed single-frame image.

[0018] Preferably, the process of obtaining the final depth map and the pixel-by-pixel spatial oblique plane comprises:

[0019] Downsampling the single-frame image of the workpiece under test twice, and initializing the spatial inclined plane within the assumed range of the depth to be measured and the imaging angle;

[0020] Set one image in the single-frame image of the workpiece to be tested as a reference image (each image is used as a reference image once and the following algorithm is processed), and the remaining images are source images, calculate the binocular matching cost of the local matching box of the reference pixel and the corresponding matching box of each source image, and obtain the multi-eye matching cost based on the binocular matching cost weighting;

[0021] Iterate the spatial plane hypothesis according to the multi-view matching cost, and then refine the spatial plane to obtain a final depth map corresponding to the reference image after being downsampled twice and a pixel-by-pixel spatial oblique plane set;

[0022] Repeating view selection, propagation and refinement for multiple times on the depth map corresponding to the reference image and the pixel-by-pixel spatial oblique plane set to generate a coarsest-scale depth map;

[0023] The coarsest-scale depth map is refined by multi-scale consistency matching cost, and then the refined coarsest-scale depth map is upsampled and followed by a depth map refinement module twice to obtain the intermediate depth map and the pixel-by-pixel spatial oblique plane.

[0024] Preferably, the expression of the multi-item matching cost is:

[0025]

[0026] Where W'(I srcj ) represents the weight of each source image corresponding to the reference pixel, m i,j Represents the stereo matching cost calculated between the reference box at the reference pixel and the corresponding box in the source image.

[0027] Preferably, the process of obtaining the final parameters of the spatial inclined plane includes:

[0028] Obtaining an iterative initial frame of a digital speckle correlation method based on the obtained pixel-by-pixel intermediate space slant plane, calculating parameters of the iterative initial frame, and obtaining a pixel-by-pixel iterative initial value;

[0029] The pixel-by-pixel iteration initial value is iteratively calculated by a digital speckle correlation method, and the final parameter of the spatial inclined plane is determined according to a hyperparameter threshold.

[0030] Preferably, the process of iteratively calculating the pixel-by-pixel iterative initial value by digital speckle correlation method further comprises: applying parameter increments to the reference sub-area by ZNSSD correlation function before each iteration;

[0031] The expression of the Δp form of the ZNSSD correlation function is:

[0032]

[0033] in, is the Hessian Matrix, in is the grayscale gradient matrix of the reference sub-area in the x,y direction, is the Jacobian matrix of the shape function.

[0034] Preferably, the process of obtaining the final point cloud data further includes:

[0035] Reprojecting the two-dimensional coordinates in the final depth map back to the world coordinate system, and projecting them into the remaining depth maps to obtain projected sub-pixel coordinates;

[0036] Reprojecting the projected sub-pixel coordinates into a world coordinate system, and projecting the spatial coordinate points reprojected into the world coordinate system back into the reference depth map to obtain sub-pixel coordinates;

[0037] The Euclidean distance between the original coordinates and the sub-pixel coordinates obtains the final consistency distance;

[0038] The final point cloud data is obtained based on the final consistency distance and the spatial inclined plane angle threshold.

[0039] Compared with the prior art, the present invention has the following advantages and technical effects:

[0040] The present invention provides non-contact measurement with high precision, high speed, large field of view, and can use a single frame to quickly reconstruct workpieces with high dynamic surfaces in the field of view to obtain high integrity and high precision point clouds, and its measurement accuracy can reach 0.03-0.07mm. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0042] Figure 1 Schematic diagram of a four-eye speckle projection acquisition system according to an embodiment of the present invention;

[0043] Figure 2 Schematic diagrams of two different random initialization assumptions according to embodiments of the present invention, wherein (a) is a schematic diagram of a front parallel plane with integer value disparity, and (b) is a schematic diagram of an inclined plane with continuous normal vectors and depth used in the present invention;

[0044] Figure 3 A red and black checkerboard iterative pattern diagram of an embodiment of the present invention;

[0045] Figure 4 This is a schematic diagram of reconstructing a three-dimensional point cloud of a standard part according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0047] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0048] Embodiment 1

[0049] This embodiment provides a high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method, including:

[0050] A speckle projection acquisition system is constructed based on a four-eye camera and a photolithography speckle projection module, and a single-frame image of the workpiece to be measured is acquired based on the speckle projection acquisition system;

[0051] Processing the single frame image of the workpiece to be measured based on a multi-scale robust multi-objective matching algorithm to obtain an intermediate depth map and a pixel-by-pixel spatial oblique plane;

[0052] Iteratively calculating the intermediate parameters of the spatial slant plane pixel by pixel based on the digital speckle correlation method to obtain the final parameters of the spatial slant plane;

[0053] The final depth map is fused based on the final parameters of the spatial inclined plane and the left-right consistency superposition test to obtain final point cloud data.

[0054] Specifically:

[0055] Adjust the shooting exposure time of the multi-camera to a reasonable value, and use the well-designed four-camera speckle projection acquisition system to collect a single frame image of the workpiece under test. The edge-guided multi-scale robust multi-camera matching principle is used to obtain the pixel-by-pixel spatial rough intermediate oblique plane corresponding to each camera of the workpiece under test. The pixel-by-pixel spatial oblique plane parameters are mapped to the source image through the homography matrix to initialize the initial iteration parameters of the digital speckle correlation method. The digital speckle correlation method is used for 1-2 iterations, and the final parameters of the spatial oblique plane are determined according to the hyperparameter threshold. Finally, the depth map is fused through the left and right consistency superposition test to obtain the final point cloud for accuracy evaluation and analysis.

[0056] Further optimizing the scheme, the process of constructing a speckle projection acquisition system based on a multi-camera includes:

[0057] Label the four cameras around the camera and install the speckle projection module at the symmetrical center of the camera;

[0058] The labeled quadrocopter is subjected to single-target and dual-target positioning, and then the internal and external parameters of the quadrocopter are optimized by using the bundle adjustment method to generate the speckle projection acquisition system.

[0059] According to a further optimization scheme, the process after the speckle projection acquisition system acquires a single frame image of the workpiece under test also includes:

[0060] Performing distortion correction processing on the single-frame image of the measured workpiece to obtain a distortion-corrected image;

[0061] Processing the distortion-corrected image based on a median filtering algorithm to obtain a filtered denoised image;

[0062] The filtered image is processed based on a Gaussian filtering algorithm to obtain a preprocessed single-frame image.

[0063] Specifically:

[0064] The first step is to label the surrounding four cameras, with the center being the speckle projection module, such as Figure 1 First, all cameras are individually calibrated for single-target alignment; then cameras 2, 3, and 4 are calibrated with camera 1 for dual-target alignment; finally, the bundle adjustment method is used to optimize the internal and external parameters of the multi-camera.

[0065] The second step is to adjust the shooting exposure time of the multi-eye camera to a reasonable value, adjust the focal length of the speckle projection module to make the shooting clear, and use the carefully designed four-eye speckle projection acquisition system to collect a single frame image of the workpiece being measured.

[0066] The third step is that in the image preprocessing stage, in order to improve the image quality and reduce the interference of subsequent processing, this embodiment takes multiple steps to eliminate the noise in the image. First, for the four images, this embodiment performs distortion correction processing to eliminate the image distortion caused by the camera lens or other factors and ensure the geometric accuracy of the image. Subsequently, in order to remove the salt and pepper noise in the image, this embodiment applies the median filtering algorithm for smoothing. The working principle of the median filter is that for each pixel in the image, its value is replaced by the median of all pixel values ​​in the neighborhood of the pixel. Finally, in order to further remove the noise in the image, this embodiment uses Gaussian filtering. Gaussian filtering works by assigning a weighted average of the pixel values ​​in its neighborhood to each pixel in the image, where the weight is determined by a Gaussian function. This filtering method can smooth the image, remove most of the noise, while retaining the overall edge information of the image, and provide clearer image data for subsequent image analysis or recognition tasks.

[0067] The process of obtaining the final depth map and the pixel-by-pixel spatial oblique plane includes:

[0068] Downsampling the single-frame image of the workpiece under test twice, and initializing the spatial inclined plane within the assumed range of the depth to be measured and the imaging angle;

[0069] Set one image in the single-frame image of the workpiece to be tested as a reference image (each image is used as a reference image once and the following algorithm is processed), and the remaining images are source images, calculate the binocular matching cost of the local matching box of the reference pixel and the corresponding matching box of each source image, and obtain the multi-eye matching cost based on the binocular matching cost weighting;

[0070] Iterate the spatial plane hypothesis according to the multi-view matching cost, and then refine the spatial plane to obtain a final depth map corresponding to the reference image after being downsampled twice and a pixel-by-pixel spatial oblique plane set;

[0071] Repeating view selection, propagation and refinement for multiple times on the depth map corresponding to the reference image and the pixel-by-pixel spatial oblique plane set to generate a coarsest-scale depth map;

[0072] The coarsest-scale depth map is refined by multi-scale consistency matching cost, and then the refined coarsest-scale depth map is upsampled and followed by a depth map refinement module twice to obtain the intermediate depth map and the pixel-by-pixel spatial oblique plane.

[0073] Specifically:

[0074] In the first step, the initial image is downsampled twice and the spatial inclined plane is initialized within the assumed range of depth to be measured and imaging angle.

[0075] is the reference image I after downsampling twice ref Each pixel of randomly initializes a spatial oblique plane called a hypothetical plane (including continuous normal vectors and depth). This spatial oblique plane is different from the front parallel plane at the integer value parallax and is closer to the fact that the workpiece is not parallel to the camera imaging plane, such as Figure 2 shown.

[0076] The second step is to calculate the multi-image matching cost (each image is regarded as a reference image to calculate the matching cost).

[0077] Set one image as the reference image and the rest as the source images. Consider a pixel in the reference image as the reference pixel X. ref , since it corresponds to a randomly initialized spatial plane, combined with multi-target calibration parameters, the plane-induced homography principle can be used to calculate the center and shape of the matching box corresponding to the reference pixel local matching box in other images. The proposed adaptive matching cost is used to calculate the binocular matching cost of the reference pixel local matching box and the corresponding matching box of each source image, and then the adaptive weight of each source view is calculated using the binocular matching cost as a clue through the multi-hypothesis joint view selection strategy. Finally, the multi-target matching cost is calculated by weighted averaging. The formula for calculating the multi-target matching cost is as follows:

[0078]

[0079] in Represents the weight of each source image corresponding to the reference pixel, m i,j The binocular matching cost calculated by the reference frame at the reference pixel and the corresponding frame of the source image can be expressed as follows:

[0080]

[0081] Where ZNCC(*,*) represents the normalized cross-correlation matching cost with zero mean.

[0082] The third step is to iterate the spatial plane hypothesis based on the multi-objective matching cost.

[0083] First, the entire image is divided into a red and black checkerboard pattern, such as Figure 3 (a). Red pixels can be updated simultaneously with black pixels, and vice versa. Figure 3 (a) The eight red filled pixels on the left are expanded into four V-shaped areas and four long strip areas, containing 7 pixels and 11 pixels respectively. Figure 3 (b) shows that on this basis, the best multi-view aggregation matching cost (Formula 1) is selected from the pixels in these 8 regions to represent the matching degree of the center point. Finally, the chessboard propagation model is used to update the estimate of the center pixel with the minimum cost of multi-view aggregation.

[0084] The fourth step is to refine the spatial plane.

[0085] Due to the random initialization of each hypothesis, the present embodiment applies a refinement step after each red-black iteration to approximate the real space plane corresponding to each pixel in the larger solution space. The present embodiment obtains a new depth by perturbing the current depth and randomly generating it, as well as a new normal. Subsequently, the present embodiment obtains a new hypothesis by combining these new depths and normals with the current depth and normal. The hypothesis with the minimum aggregation cost is considered to be the final estimate of the pixel. After multiple repetitions of view selection, propagation, and refinement, the final depth map corresponding to the reference image after downsampling twice can be obtained.

[0086] Step 5: Multi-scale geometric consistency iteration.

[0087] Since surface texture is a relative measure, multi-scale information can be used to improve matching accuracy and completeness. Since two downsampling operations have been performed to enhance surface texture, however, reconstruction of thin structure areas requires finer-scale considerations. After repeated view selection, propagation, and refinement, the coarsest-scale depth map is obtained (steps 2 to 4), and the depth map is refined by multi-scale consistency matching cost, as shown in the following formula:

[0088]

[0089] Where Δ i,j is the sub-pixel reprojection error term, and λ is a constant that balances photometric consistency and geometric consistency.

[0090] The depth map is then upsampled twice. After each upsampling, the depth map refinement module is used to balance the accuracy of thin structures and large planes, and the multi-scale consistency matching cost is run again to further refine the depth map. The above two upsampling to depth map refinement can obtain the final depth map and the pixel-by-pixel spatial oblique plane.

[0091] In order to improve the overall matching speed and ensure the reconstruction accuracy of thin structures, a 5×5 matching box is used in the multi-eye patch matching process to obtain a rough intermediate space oblique plane in the near-plane area with low curvature, and an accurate intermediate space oblique plane in the thin structure area with high curvature.

[0092] The process of obtaining the final parameters of the spatial inclined plane includes:

[0093] Obtaining an iterative initial frame of a digital speckle correlation method based on the obtained pixel-by-pixel intermediate space slant plane, calculating parameters of the iterative initial frame, and obtaining a pixel-by-pixel iterative initial value;

[0094] The pixel-by-pixel iteration initial value is iteratively calculated by a digital speckle correlation method, and the final parameter of the spatial inclined plane is determined according to a hyperparameter threshold.

[0095] Furthermore, the process of iteratively calculating the pixel-by-pixel iterative initial value by using the digital speckle correlation method further includes: applying a parameter increment to a reference sub-area by using a ZNSSD correlation function before each iteration;

[0096] In the first step, the iterative initial frame is obtained according to the pixel-by-pixel spatial oblique plane obtained based on the multi-eye patch matching algorithm.

[0097] Through theoretical derivation and experiments, it can be concluded that there is a one-to-one correspondence between the spatial oblique plane and the first-order shape function of the digital speckle correlation method. Therefore, the pixel-by-pixel intermediate spatial oblique plane obtained based on the multi-eye patch matching algorithm can be used to obtain the pixel-by-pixel iteration initial value of the digital speckle correlation method, so as to avoid the problems of too few seed points and obstructed propagation caused by the seed point selection and propagation process of the traditional digital speckle correlation method. At the same time, GPU acceleration can be used to accelerate the iterative process of the digital speckle correlation method.

[0098] The second step is to iterate the parameter calculation of the initial box.

[0099] Since there is a one-to-one correspondence between the spatial inclined plane and the first-order shape function of the digital speckle correlation method, the coordinates of the four corner points of the matching box can be used to initialize the first-order shape function. The expression of the first-order shape function is as follows:

[0100]

[0101] where p1=(u,u x ,u y ,v,v x ,v y ) T is the sub-pixel deviation and gradient parameter corresponding to the first-order shape function. The following two equations can be combined to initialize the first-order shape function:

[0102]

[0103] By combining the above two equations, all the required parameters of the first-order shape function can be solved.

[0104] The third step is the iteration of multi-eye digital speckle correlation method.

[0105] Set an image as the reference image, and use the IC-GN algorithm to iterate the digital speckle correlation method for each source image. The basic idea is to assume that the reference sub-area can better match the target sub-area after deformation, and apply the parameter increment to the reference sub-area before each iteration. The similarity metric ZNSSD (zero normalized square sum difference correlation criterion) is used as the similarity metric between the reference sub-area and the target sub-area, as shown below:

[0106]

[0107] Assuming that the reference sub-area is deformed, the Δp form of its ZNSSD correlation function can be expressed as:

[0108]

[0109] in is the Hessian Matrix, in is the grayscale gradient matrix of the reference sub-area in the x,y direction, is the Jacobian matrix of the shape function. After the mapping parameter increment is obtained from the above formula, it is applied to the reference sub-region to obtain the incremental shape function matrix; then it is inverted and combined with the current shape function matrix to obtain the updated shape function matrix:

[0110] W(x,y;p)=W(x,y;p)·W -1 (x,y;Δp);

[0111] At this point, one iteration is completed.

[0112] The fourth step is to set the iteration threshold.

[0113] Since the larger the matching box, the more accurate the iterative matching of the digital image correlation method, but the matching accuracy of thin structures will also be lost, after one or two iterations, the ZNCC matching cost threshold is set to 0.85 (the larger the value, the more accurate the matching), the Δp threshold is set to 0.05, and the optimal source image matching cost that meets the Δp threshold is used as the final reconstructed matching box. The surrounding points and the center point are reconstructed by triangulation with the reference image and then fitted into a spatial plane. The depth and normal vector of the spatial plane correspond to the updated depth and normal vector of the pixel-by-pixel spatial oblique plane. In the process of digital speckle correlation matching, a matching box of 19×19 is used to maximize the reconstruction accuracy of the near plane at small curvature.

[0114] Step 5: Fusion of the four final depth maps.

[0115] In the depth map fusion step, left-right consistency checks and robust fusion strategies are performed. Specifically, this embodiment reprojects the two-dimensional coordinates in the reference depth image back to the world coordinate system, and projects them into four other depth maps through the homography matrix based on the spatial oblique plane information at the pixel. The depth and normal at the sub-pixel are obtained by an interpolation method. Subsequently, this embodiment reprojects it into the world coordinate system and projects it back into the reference depth map to obtain the sub-pixel coordinates. The final consistency distance is the Euclidean distance between the original coordinates and the reprojected sub-pixel coordinates. Here, a consistency threshold is defined, the relative depth difference is 0.01, the angle between the normals is 30°, and the reprojection error is limited to 0.3 pixels. Any corresponding matches at pixels less than the threshold are counted. When the count is greater than 1, the assumed spatial oblique plane will be retained, and the spatial coordinates corresponding to these assumed 3D points and the normals are averaged as the final result. The 3D spatial coordinates are aggregated to obtain a point cloud on the surface of the workpiece, such as Figure 4 shown.

[0116] The process of obtaining the final point cloud data also includes:

[0117] Reprojecting the two-dimensional coordinates in the final depth map back to the world coordinate system, and projecting them into the remaining depth maps to obtain projected sub-pixel coordinates;

[0118] Reprojecting the projected sub-pixel coordinates into a world coordinate system, and projecting the spatial coordinate points reprojected into the world coordinate system back into the reference depth map to obtain sub-pixel coordinates;

[0119] The Euclidean distance between the original coordinates and the sub-pixel coordinates obtains the final consistency distance;

[0120] The final point cloud data is obtained based on the final consistency distance and the spatial inclined plane angle threshold.

[0121] From the measurement results, it can be seen that the proposed method achieves high-integrity reconstruction on high-dynamic surfaces. Multiple reconstruction measurements are performed on the standard sheet metal parts after fine processing, and the accuracy is evaluated by comparing them with the standard digital model. It can be concluded that the proposed method can achieve a measurement accuracy of 0.03-0.07mm, which meets industrial needs.

[0122] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method, characterized in that: The following steps are involved: A speckle projection acquisition system is constructed based on a four-eye camera and a photolithography speckle projection module, and a single-frame image of the workpiece to be measured is acquired based on the speckle projection acquisition system; Processing the single frame image of the workpiece to be measured based on a multi-scale robust multi-objective matching algorithm to obtain an intermediate depth map and a pixel-by-pixel spatial oblique plane; Iteratively calculating the intermediate parameters of the spatial slant plane pixel by pixel based on the digital speckle correlation method to obtain the final parameters of the spatial slant plane; The final depth map is fused based on the final parameters of the spatial inclined plane and the left-right consistency superposition test to obtain final point cloud data.

2. The high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method according to claim 1 is characterized in that: The process of constructing a speckle projection acquisition system based on a multi-camera comprises: Label the four cameras around the camera and install the speckle projection module at the symmetrical center of the camera; The labeled quadrocopter is subjected to single-target and dual-target positioning, and then the internal and external parameters of the quadrocopter are optimized by using the bundle adjustment method to generate the speckle projection acquisition system.

3. The high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method according to claim 1 is characterized in that: The process after the single frame image of the workpiece to be measured is collected by the speckle projection collection system further includes: Performing distortion correction processing on the single-frame image of the measured workpiece to obtain a distortion-corrected image; Processing the distortion-corrected image based on a median filtering algorithm to obtain a filtered denoised image; The filtered image is processed based on a Gaussian filtering algorithm to obtain a preprocessed single-frame image.

4. The high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method according to claim 1 is characterized in that: The process of obtaining the final depth map and the pixel-by-pixel spatial oblique plane includes: Downsampling the single-frame image of the workpiece under test twice, and initializing the spatial inclined plane within the assumed range of the depth to be measured and the imaging angle; Setting one image in a single frame image of the workpiece to be measured as a reference image and the remaining images as source images, calculating the binocular matching cost of the local matching frame of the reference pixel and the corresponding matching frame of each source image, and obtaining the multi-eye matching cost based on the binocular matching cost weighting; Iterate the spatial plane hypothesis according to the multi-view matching cost, and then refine the spatial plane to obtain a final depth map corresponding to the reference image after being downsampled twice and a pixel-by-pixel spatial oblique plane set; Repeating view selection, propagation and refinement for multiple times on the depth map corresponding to the reference image and the pixel-by-pixel spatial oblique plane set to generate a coarsest-scale depth map; The coarsest-scale depth map is refined by multi-scale consistency matching cost, and then the refined coarsest-scale depth map is upsampled and followed by a depth map refinement module twice to obtain the intermediate depth map and the pixel-by-pixel spatial oblique plane.

5. The high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method according to claim 4 is characterized in that: The expression of the multi-item matching cost is: in Represents the weight of each source image corresponding to the reference pixel, m i,j Represents the binocular matching cost calculated between the reference box at the reference pixel and the corresponding box in the source image.

6. The high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method according to claim 1 is characterized in that: The process of obtaining the final parameters of the spatial inclined plane includes: Obtaining an iterative initial frame of a digital speckle correlation method based on the obtained pixel-by-pixel intermediate space slant plane, calculating parameters of the iterative initial frame, and obtaining a pixel-by-pixel iterative initial value; The pixel-by-pixel iteration initial value is iteratively calculated by a digital speckle correlation method, and the final parameter of the spatial inclined plane is determined according to a hyperparameter threshold.

7. The high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method according to claim 6 is characterized in that: The process of iteratively calculating the pixel-by-pixel iterative initial value by the digital speckle correlation method further comprises: applying a parameter increment to a reference sub-area by a ZNSSD correlation function before each iteration; The expression of the Δp form of the ZNSSD correlation function is: in, is the Hessian Matrix, in is the grayscale gradient matrix of the reference sub-area in the x,y direction, is the Jacobian matrix of the shape function.

8. The high dynamic surface single frame measurement method based on multi-eye digital speckle correlation method according to claim 1 is characterized in that: The process of obtaining the final point cloud data also includes: Reprojecting the two-dimensional coordinates in the final depth map back to the world coordinate system, and projecting them into the remaining depth maps to obtain projected sub-pixel coordinates; Reprojecting the projected sub-pixel coordinates into a world coordinate system, and projecting the spatial coordinate points reprojected into the world coordinate system back into the reference depth map to obtain sub-pixel coordinates; The Euclidean distance between the original coordinates and the sub-pixel coordinates obtains the final consistency distance; The final point cloud data is obtained based on the final consistency distance and the angle threshold of the spatial inclined plane.

Images