Stamping part measuring method and system based on double-frequency fringe projection and spectral material identification

Abstract

The invention discloses a stamping part measuring method and system based on double-frequency fringe projection and spectral material recognition. The method comprises the steps that phase shift patterns of low-frequency fringes and high-frequency fringes are generated; a deformation stripe image of the surface of the measured stamping part is collected; calculating a wrapped phase according to the deformed fringe image; performing dual-frequency phase unwrapping; a multispectral image of the stamping part is collected, and the relative reflectivity is calculated; performing principal component analysis dimension reduction on the multi-wavelength reflectivity data; identifying the material of the stamping part according to the spectral features, and dynamically adjusting fringe projection parameters based on an identification result; mapping the unwrapped phase into a three-dimensional coordinate; performing adaptive bilateral filtering on the point cloud data; and extracting the characteristics of the stamping part, calculating the size and comparing. By applying the method, the system has the following beneficial effects that high-precision and high-efficiency measurement of stamping parts made of various materials is realized by fusing a dual-frequency phase-shift fringe projection three-dimensional shape reconstruction algorithm and a structured spectral reflectivity analysis material recognition algorithm.

Description

Technical Field

[0001] The present invention relates to the field of machine vision measurement technology, and in particular to a stamping part measurement method and system using dual-frequency fringe projection and spectral material recognition. Background Art

[0002] Stampings are critical components widely used in modern industry, including automotive, aviation, and electronics. Their dimensional accuracy directly impacts product quality and assembly performance. As the manufacturing industry continues to demand higher precision, accurate measurement of stamping part dimensions is becoming increasingly important. Traditionally, stamping part dimensional measurement relies primarily on contact measuring tools (such as calipers and micrometers) or coordinate measuring machines (CMMs). These methods suffer from low measurement efficiency, difficulty implementing online inspection, and the inability to simultaneously measure multiple critical dimensions.

[0003] With the development of Industry 4.0, non-contact optical measurement technology has gradually become the mainstream direction for stamping part dimensional measurement due to its high efficiency, high precision, and non-destructive testing characteristics. Structured light 3D measurement technology has been widely used due to its high measurement accuracy, fast speed, and wide range of applications. Phase-shifted fringe projection technology, in particular, achieves submillimeter 3D measurement accuracy by projecting a series of fringe patterns with fixed phase differences.

[0004] However, existing optical measurement technology faces multiple challenges when applied to stamping part measurement. First, stamping parts are typically made of metal, which has a highly reflective surface and is prone to specular reflection. This can cause overexposure or spotting during optical measurement, affecting measurement accuracy. Second, stamping parts have complex shapes and often contain discontinuous features such as holes, slots, and steps. These areas are difficult to measure and prone to data loss. Furthermore, interference factors such as lighting changes and vibration in industrial environments can also affect measurement stability.

[0005] The mainstream optical measurement technologies currently on the market include phase-shifted fringe projection, speckle projection, and optical coding. While phase-shifted fringe projection offers high measurement accuracy, it struggles with phase extraction when measuring highly reflective surfaces. Speckle projection offers strong interference immunity but relatively low accuracy. Optical coding, while simple and fast, has limited ability to process surface discontinuities. Furthermore, existing measurement systems generally lack the ability to identify the material being measured and are unable to optimize measurement parameters based on different material properties, resulting in poor performance when measuring complex stamping parts.

[0006] Therefore, there is an urgent need to develop an intelligent measurement method and system that can effectively handle stamping parts made of various materials, especially stamping parts with highly reflective surfaces and complex shapes, to improve measurement accuracy, efficiency and scope of application. Summary of the Invention

[0007] The purpose of the present invention is to provide a stamping part measurement method and system based on dual-frequency fringe projection and spectral material recognition. Through an algorithm fusion mechanism, high-precision and high-efficiency measurement of stamping parts made of various materials can be achieved, overcoming the limitations of existing technologies in processing highly reflective surfaces and complex shapes.

[0008] A key aspect of this invention is the organic combination of a dual-frequency phase-shifted fringe projection 3D topography reconstruction algorithm and a structured spectral reflectance analysis material identification algorithm. This method dynamically adjusts the reconstruction parameters based on the material identification results, enabling adaptive measurement of stamping parts made of different materials. The system first collects spectral reflectance data from stamping parts using multi-wavelength illumination for material identification. Based on the identification results, it then dynamically optimizes fringe projection parameters, camera parameters, and post-processing algorithm parameters. Finally, it performs the optimized dual-frequency phase-shifted fringe projection 3D topography reconstruction to obtain a high-precision 3D model of the stamping part and key dimension measurements.

[0009] The dual-frequency phase-shifted fringe projection algorithm is the basis of the present invention. It uses two frequency phase-shifted fringe patterns: low-frequency fringe patterns (8-16 stripes / screen width) and high-frequency fringe patterns (64-128 stripes / screen width). The mathematical model of the fringe pattern is: pn (x,y)=I0[1+m·cos(2πfx+φ0+2πn / N)], where I0 is the average light intensity, m is the fringe contrast, f is the fringe frequency, φ0 is the initial phase, n is the number of phase shift steps, and N is the total number of phase shift steps. The system projects N phase-shifted fringes for each frequency (typically N=3, 4, or 5), with a phase difference of 2π / N.

[0010] When the stripe pattern is projected onto the surface of the stamped part, the stripes are deformed due to the uneven surface topography. The camera collects these deformed stripe images from a specific angle, which can be expressed as: I n (x,y)=a(x,y)+b(x,y)·cos[φ(x,y)+2πn / N], where a(x,y) is the background light intensity, b(x,y) is the reflected modulated light intensity, and φ(x,y) is the phase containing the object height information.

[0011] Based on the N-step phase-shifted image, the system calculates the wrapping phase: φ w (x,y)=tan^(-1)[∑(n=0 to N-1)I n (x,y)·sin(2πn / N) / ∑(n=0 to N-1)I n (x,y)·cos(2πn / N)]. For the four-step phase shift method (N=4), the calculation can be simplified to: w (x,y)=tan^(-1)[(I3-I1) / (I0-I2)]. Due to the periodicity of the inverse trigonometric function, the calculated phase value φ wIt is wrapped in the range of -π to π, with a phase jump of 2π.

[0012] The innovation of this invention is the dual-frequency phase unwrapping. H ) and low frequency (f L ) fringes, achieving robust phase unwrapping: φ(x,y)=φ H (x,y)+2πk(x,y), where k(x,y) is an integer order determined by the low-frequency phase: k(x,y)=round[(f H / f L ·φ L (x,y)-φ H (x,y)) / (2π)]. This dual-frequency unwrapping method improves the phase unwrapping reliability on complex surfaces and discontinuous areas.

[0013] The unwrapped phase is mapped to three-dimensional coordinates using the triangulation principle. First, the relationship between the phase value φ(x,y) and the pixel displacement d(x,y) is: d(x,y) = φ(x,y)·p / (2π), where p is the pixel width of a single periodic stripe of the projector. Then the three-dimensional coordinates are calculated: Z(x,y) = B / (d(x,y)-d0), X(x,y) = Z(x,y)(x-x0) / f x ,Y(x,y)=Z(x,y)(y-y0) / f y , where B is the baseline distance between the camera and the projector, d0 is the pixel displacement corresponding to the reference plane, (x0, y0) is the coordinate of the camera principal point, and f x and f y is the equivalent focal length parameter of the camera.

[0014] Structured spectral reflectance analysis and material identification are another core of this invention. The system is equipped with a multi-wavelength LED light source (including 460nm blue light, 520nm green light, 590nm yellow light, 630nm red light, and 850nm near-infrared light) to form structured spectral lighting. The spectral characteristics are expressed as: S(λ) = ∑I i ·δ(λ-λ i ), where I i is the intensity of the i-th LED, λ i is its central wavelength.

[0015] The system collects images of stamping parts under different wavelengths of illumination and calculates the relative reflectivity: R λ (x,y)=I λ (x,y) / I λ ^0(x,y), where I λ (x, y) is the image of the stamping under wavelength λ, I λ^0(x,y) is an image of a standard white plate under the same conditions. Each pixel obtains an N-dimensional reflectivity vector: R(x,y)=[R λ 1(x,y),R λ 2(x,y),...,R λ N(x,y)].

[0016] To reduce computational complexity, the system uses principal component analysis (PCA) to reduce the dimensionality of high-dimensional spectral data: S = Φ^T·R, where Φ is the principal component eigenvector matrix and S is the eigenvector after dimensionality reduction. Based on these spectral features, the system uses support vector machines (SVM) to classify materials: f(S) = sign(∑α i y i K(S,S i )+b), where the radial basis function kernel is used: K(S,S')=exp(-γ||S-S'||^2).

[0017] Material recognition results are used to dynamically adjust measurement parameters such as fringe contrast, average light intensity, and camera exposure time: m adaptive =m base ·f reflect (M), I 0adaptive =I 0base ·g reflect (M), t exp =t0·α M , where M is the material parameter, f reflect 、g reflect and α M This is an adjustment function based on material properties. The adaptive mechanism enables the system to automatically select the optimal measurement parameters for stamping parts made of different materials, significantly improving measurement accuracy and stability.

[0018] Another key point of the present invention is the morphology-material information fusion mechanism. The system calculates the spatial gradient of material parameters to predict edge positions: probability (x,y)=||▽M(x,y)||, and increase the local projection density in the edge area. At the same time, construct the morphology-material fusion feature vector: Feature fusion =[S,H], where S is the spectral feature vector and H is the morphological feature vector, which is used to improve the accuracy of feature point recognition and size measurement.

[0019] For areas with highly reflective materials, the present invention develops a phase local optimization strategy: φ optimized (x,y)=argmin φ [∑(i∈N(x,y))w i ·(φ(i)-φ estimated (i))2 +λ·R(φ)], where N(x,y) is the point (x,y) Neighborhood, w i is the weight coefficient, R(φ) is the regularization term, and λ is the balance parameter. The local optimization strategy effectively solves the problem of difficult phase extraction in highly reflective areas.

[0020] The acquired original 3D point cloud data is post-processed by adaptive bilateral filtering: Z'(x,y)=∑G s (p,q)G r (Z(x+p,y+q)-Z(x,y),M)Z(x+p,y+q) / ∑G s (p,q)G r (Z(x+p,y+q)-Z(x,y),M), where G s and G r are Gaussian kernels in the spatial domain and the value domain respectively, G r The parameters of are associated with the material properties M, which realizes adaptive filtering based on material properties, which can effectively suppress noise while retaining edge and shape features.

[0021] Compared with the existing technology, the present invention has the following advantages: First, by integrating dual-frequency phase-shifted fringe projection and spectral reflectance analysis, high-precision measurement of stamping parts of various materials is achieved, especially the measurement accuracy of highly reflective metal surfaces is significantly improved; second, the parameter adaptive adjustment mechanism based on material recognition results enables the system to automatically optimize measurement parameters for different materials, improving the adaptability and stability of the system; third, the morphology-material information fusion mechanism and the special processing strategy for highly reflective areas solve the difficulties of traditional methods in processing complex shapes and edge features; finally, the adaptive bilateral filtering algorithm improves the quality of three-dimensional data and provides a basis for subsequent feature extraction and size measurement. The present invention can be widely used in the quality control of stamping parts in the automotive, aviation, electronics and other industries, and has significant application value and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a schematic diagram of the overall architecture of the system of the present invention; Figure 2 It is a flow chart of the dual-frequency phase-shifted fringe projection algorithm of the present invention; Figure 3 This is a flow chart of the structured spectral reflectance analysis material identification algorithm of the present invention; Figure 4 It is a schematic diagram of the algorithm fusion and parameter optimization mechanism of the present invention; Figure 5 This is a typical three-dimensional measurement rendering of a stamping part in the present invention, including the classification results of different material areas and the comparison of measurement accuracy; Figure 6 This is a comparison diagram of the effects before and after the treatment of the high-reflective area of the present invention; Figure 7 This is a comparison chart of the measurement accuracy of the system of the present invention and the traditional method on stamping parts made of different materials; Figure 8 It is a schematic diagram of the multi-angle data acquisition and fusion process of the system of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] like Figure 1 As shown, the intelligent visual stamping part size measurement system that integrates dual-frequency phase-shifted fringe projection and spectral reflectance analysis of the present invention mainly includes a multi-wavelength structured light source, a high-resolution industrial camera, a three-dimensional morphology reconstruction module, a material recognition module, a parameter adaptation module, a point cloud processing module, and a feature extraction and size measurement module.

[0025] The multi-wavelength structured light source can simultaneously realize the two functions of stripe projection and multi-wavelength illumination. In this embodiment, the structured light source adopts digital micromirror device (DMD) technology to realize programmable light source projection. It is equipped with an LED array with wavelengths of 460nm, 520nm, 590nm, 630nm and 850nm, which can produce phase-shifted stripes of different frequencies and multi-wavelength illumination. The resolution of DMD is 1920×1080 pixels, which can generate high-quality stripe patterns. The structured light source is also equipped with a polarization control device for suppressing mirror reflections, as well as an optical path switching mechanism to achieve rapid switching between structured light mode and spectral illumination mode, with a switching time of less than 50ms.

[0026] The high-resolution industrial camera utilizes a CMOS sensor with a resolution of 4096 x 3072 pixels and a maximum frame rate of 60 fps. It is equipped with a high-quality lens and an adjustable aperture. The baseline distance B between the camera and the structured light source is adjustable from 300 to 500 mm and can be optimized based on the measurement field of view and accuracy requirements. The camera is mounted on a precision slide drive mechanism, enabling multi-angle data acquisition to cover all surfaces of the stamping part and address visual obstructions.

[0027] 3D shape reconstruction module such as Figure 2 As shown in Figure 1, it realizes the three-dimensional topography reconstruction of dual-frequency phase-shifted fringe projection. First, two phase-shifted fringe patterns of low frequency and high frequency are generated. The mathematical model of the fringe pattern is: I pn (x,y)=I0[1+m·cos(2πfx+φ0+2πn / N)]; Where I0 is the average light intensity, which controls the projection brightness; m is the fringe contrast, which controls the black-white contrast of the fringe, and is generally set to 0.5-1. This value will be dynamically reduced for highly reflective surfaces; f is the fringe frequency, which is generally 8-16 fringe / screen width for low frequency and 64-128 fringe / screen width for high frequency; φ0 is the initial phase, which is generally set to 0; n represents the number of phase shift steps, which is 0, 1, ..., N-1; and N is the total number of phase shift steps, which is generally 3, 4, or 5.

[0028] In this embodiment, f L = 12 low-frequency stripes / screen width and f H =96 high-frequency stripes / screen width, frequency ratio f H / f L =8. A four-step phase shift method (N=4) is used, with the phases being 0, π / 2, π, and 3π / 2, respectively. When the fringes are projected onto the surface of the stamped part, they are deformed due to the change in surface height. The deformed fringes image captured by the camera can be expressed as: I n (x,y)=a(x,y)+b(x,y)·cos[φ(x,y)+2πn / N]; Among them, a(x,y) is the background light intensity, which is affected by the ambient light and surface reflection characteristics; b(x,y) is the reflected modulated light intensity, which is related to the surface reflectivity; φ(x,y) is the phase containing the object height information, which needs to be extracted.

[0029] Based on the four-step phase shift image, calculate the wrapped phase: φ w (x,y)=tan^(-1)[(I3-I1) / (I0-I2)]; Due to the periodicity of the inverse trigonometric function, the calculated phase value φ w It is wrapped in the range of -π to π, with a 2π periodic jump. In order to obtain a continuous absolute phase, a dual-frequency phase unwrapping method is used: φ(x,y)=φ H (x,y)+2πk(x,y); Where k(x,y) is an integer order, determined by the low-frequency phase: k(x,y)=round[(f H / f L ·φ L (x,y)-φ H (x,y)) / (2π)]; This dual-frequency unwrapping method can effectively handle surface discontinuities and highly reflective areas, significantly improving the reliability of phase unwrapping. In order to further ensure the accuracy of unwrapping, the present invention designs a frequency ratio f H / fL Meet f H / f L <N / 2 condition, unambiguous phase unwrapping can be achieved. In this embodiment, f H / f L = 8, N = 4, meet f H / f L <2N condition, a certain degree of phase error can be tolerated.

[0030] The mapping relationship between the unwrapped phase and the three-dimensional coordinates is based on the principle of triangulation. First, calculate the pixel displacement corresponding to the phase: d(x, y) = φ(x, y)·p / (2π); where p is the pixel width of a single-period stripe of the projector, p = number of horizontal pixels of the projector / stripe frequency. For example, for a DMD projector with a resolution of 1920×1080, when the high frequency is 96 stripes / screen width, p = 1920 / 96 = 20 pixels. Then, according to the system calibration parameters, calculate the three-dimensional coordinates: Z(x, y) = B / (d(x, y) - d0); X(x, y) = Z(x, y)(x - x0) / f x Y(x, y) = Z(x, y)(y - y0) / f y ; where B is the baseline distance between the camera and the projector, generally 300 - 500 mm; d0 is the pixel displacement corresponding to the reference plane; (x0, y0) is the camera principal point coordinates; f x and f y are the equivalent focal length parameters of the camera. These parameters are obtained through system calibration. The calibration process uses a standard checkerboard calibration board and the Zhang's calibration method to obtain the camera internal parameters and external parameters.

[0031] The material recognition module realizes structured spectral reflectance analysis for material recognition, as Figure 3 shown. This module uses the difference in the reflection characteristics of different materials for light of different wavelengths to identify the material on the surface of the stamping part. The system first illuminates the stamping part with multi-wavelength LEDs to generate structured spectral illumination covering the visible to near-infrared band. The spectral characteristics can be expressed as: S(λ) = ∑I i ·δ(λ - λ i ); where I i is the intensity of the i-th LED, λ iis its central wavelength, and δ is the Dirac function. In this example, five LEDs with central wavelengths are used: 460nm blue light for detecting metal surface oxide layers; 520nm green light for general surface reflectance analysis; 590nm yellow light, which has a characteristic reaction to copper, brass, etc.; 630nm red light, which has good discrimination ability for steel materials; and 850nm near-infrared light, which can penetrate surface contamination and reflect the characteristics of the substrate.

[0032] The system collects images under five different wavelength illumination conditions to obtain a set of multispectral image sequences I λ (x,y). To eliminate the effects of illumination non-uniformity and camera response differences, calculate the relative reflectivity: R λ (x,y)=I λ (x,y) / I λ ^0(x,y); Among them I λ ^0(x,y) is an image of a standard white plate (with a reflectivity close to 100%) under the same lighting conditions. Each pixel position obtains a 5-dimensional reflectivity vector: R(x,y)=[R 460 (x,y),R 520 (x,y),R 590 (x,y),R 630 (x,y),R 850 (x,y)]; Since multispectral data has high dimensionality and redundancy, principal component analysis (PCA) is used for dimensionality reduction: S = Φ^T·R; Where Φ is the principal component eigenvector matrix, and S is the eigenvector after dimensionality reduction. Generally, the first two to three principal components are selected, which can retain approximately 95% of the spectral information. The principal component analysis process involves calculating the mean μ and covariance matrix C of the reflectance vector, performing eigenvalue decomposition C = ΦΛΦ^T on the covariance matrix, and selecting the k eigenvectors corresponding to the largest eigenvalues to form the reduced dimensionality matrix Φ.

[0033] Based on the spectral features after dimension reduction, the system uses support vector machines (SVM) to classify materials: f(S)=sign(∑α i y i K(S,S i )+b); Among them S i is the support vector, y i is the class label, α i is the Lagrange multiplier, b is the bias term, and K is the kernel function. The radial basis function kernel is used: K(S,S')=exp(-γ||S-S'||^2); γ is a kernel parameter that controls the complexity of the decision boundary. Its optimal value is determined through cross-validation and is generally within the range of 0.1-10. The system has established classification models for more than ten common stamping material categories, including stainless steel, ordinary steel, aluminum alloy, copper alloy, and galvanized sheet. Classification accuracy is high.

[0034] In addition to classifying and identifying material types, the system also establishes a correlation model between the reflectance spectrum and the physical properties of the material: M=f(S); Where M is a vector of material parameters, including surface roughness, reflectivity, and surface treatment type, and f is a mapping function implemented using multivariate nonlinear regression or a neural network. This model enables the system to further distinguish between different surface treatments on the same material, such as polishing, brushing, sandblasting, and electroplating.

[0035] like Figure 4 As shown in the figure, the parameter adaptation module is the core module of the intelligent system. This module dynamically optimizes the measurement parameters based on the material recognition results. Based on the material characteristics M, the system adaptively adjusts the following parameters: Fringe projection parameters: fringe contrast m and average light intensity I0 are optimized through material-related adjustment functions: m adaptive =m base ·f reflect (M); I 0adaptive =I 0base ·g reflect (M); where f reflect and g reflect is an adjustment function based on the material property M. For example, for highly reflective materials (such as polished aluminum alloy), the system will reduce the fringe contrast and average light intensity to avoid overexposure and improve the signal-to-noise ratio; for low-reflective materials (such as sandblasted steel), the system will increase the fringe contrast and average light intensity to obtain a clearer fringe pattern.

[0036] Camera parameters: exposure time t exp , Gain and Aperture are also adjusted based on the material properties: t exp =t0·α M ; Where t0 is the reference exposure time, α M is the material correlation coefficient. For example, for highly reflective materials α M Take 0.3-0.5, for low reflective material α M Take 1.2-1.5.

[0037] Phase Unwrapping: Different unwrapping algorithms are applied to different material regions. Conventional dual-frequency unwrapping is used for smooth continuous areas, a combination of temporal and spatial phase unwrapping is used for highly reflective areas, and a phase-preserving partitioned unwrapping algorithm is used for material boundary areas.

[0038] Filter parameters: Parameters σ of adaptive bilateral filtering s and σ r (M) Adjust according to material characteristics, increase the filtering strength in smooth areas to suppress noise, and reduce the filtering strength in edge areas to preserve detailed features.

[0039] One of the key points of this invention is the fusion mechanism of shape and material information. The system calculates the spatial gradient of material parameters to predict edge positions: Edge probability (x,y)=||▽M(x,y)||; Where ▽M(x,y) is the spatial gradient of the material parameters, which can effectively detect the boundary of material changes, which is generally also the boundary of geometric shapes. The system increases the local projection density in the detected edge area and constructs the morphology-material fusion feature vector: Feature fusion =[S,H]; Where S is the spectral feature vector, and H is the topographic feature vector (such as surface curvature, normal vector, etc.). This fusion feature contains both material information and geometric information, enabling more accurate identification of stamping part features.

[0040] For areas with highly reflective materials, the present invention proposes local phase optimization: φ optimized (x,y)=argmin φ [∑(i∈N(x,y))w i ·(φ(i)-φ estimated (i))^2+λ·R(φ)]; Where N(x,y) is the neighborhood of point (x,y), w i is the weight coefficient, which is related to the reliability of the pixel; estimated (i) is the initial phase estimate; R(φ) is the regularization term used to maintain phase smoothness; and λ is a balance parameter that controls the weight ratio between the data term and the regularization term. This local optimization strategy effectively addresses the difficulty of extracting phase in highly reflective areas by incorporating spatial correlation and prior knowledge.

[0041] The point cloud processing module performs adaptive bilateral filtering on the acquired original 3D point cloud data: Z'(x,y)=∑G s (p,q)G r(Z(x+p,y+q)-Z(x,y),M)Z(x+p,y+q) / ∑G s (p,q)G r (Z(x+p,y+q)-Z(x,y),M) Among them: G s (p,q)=exp(-(p 2 +q 2 ) / (2σ s 2 )) is the Gaussian kernel in the spatial domain, which controls the spatial range of the filter; G r (Δz,M)=exp(-Δz 2 / (2σ r 2 (M))) is the range Gaussian kernel, which controls the allowed height difference range; σ s and σ r (M) is the parameter that controls the filtering range, where σ r Related to the material property M.

[0042] This adaptive bilateral filtering can dynamically adjust the filtering strength according to the material characteristics, effectively suppress noise, and preserve edge and shape features. For example, for a smooth polished metal surface, increasing σ s and reduce σ r , enhance the filtering effect; for brushed metal with textured surface, reduce σ s and increase σ r , preserving surface details.

[0043] The feature extraction and dimensional measurement module extracts key features of stamped parts and calculates their dimensions based on processed 3D point cloud data. Feature extraction includes edge detection (based on principal curvature and normal vector changes), surface extraction (using a region growing algorithm), and feature recognition (such as holes, slots, and steps). Dimension calculation is based on the distance and angular relationships between extracted feature points, lines, and surfaces. The geometric dimensions are then compared with the CAD model to generate a dimensional deviation report.

[0044] like Figure 8 As shown, to address perspective occlusion, the system also implements multi-angle data acquisition and fusion capabilities. The camera and structured light source are mounted on precision slides, enabling scanning of stamped parts from various angles. The multiple sets of acquired 3D data are aligned using the Iterative Closest Point (ICP) algorithm and then fused into a complete 3D model using a weighted average method: Z fused (x,y)=∑w i (x,y)·Z i (x,y) / ∑w i (x,y).

[0045] where Z i (x, y) is the height value obtained at different angles, w i (x,y) represents a weighting factor, which depends on viewing angle, material, and measurement uncertainty. For example, areas with surface normals close to the camera's viewing direction receive a higher weight, while areas with high reflectivity, resulting in greater measurement uncertainty, receive a lower weight. Weighted fusion ensures the system acquires complete and accurate 3D data.

[0046] The system also includes a self-learning module that records optimal measurement parameters for stamping parts made of different materials and continuously optimizes the material-topography correlation model based on actual measurement results. This module uses neural networks or reinforcement learning algorithms to continuously improve system performance based on historical measurement data, giving the system adaptive learning capabilities.

[0047] The present invention has been verified in multiple application scenarios, such as Figure 5 and Figure 7 As shown, it has significant performance advantages compared with traditional methods: Measurement Accuracy: The system's measurement accuracy can reach 0.01-0.02mm, far superior to the 0.05-0.1mm of traditional methods. This system maintains high-precision measurements, especially on highly reflective metal surfaces and edge features, where traditional methods often experience data loss or severe errors.

[0048] Scope of application: The system can handle stamping parts of various materials, including highly reflective metals (such as polished aluminum alloys and stainless steel), low-reflective metals (such as sandblasted steel), and composite materials. It can also accurately measure complex shapes such as deep holes, fine grooves, sharp edges, and other difficult areas.

[0049] Measurement efficiency: The measurement time for a typical stamping part is 15-20 seconds, which is 10-20 times faster than traditional contact methods and 2-3 times faster than conventional optical methods. The system also supports automatic batch measurement, further improving production efficiency.

[0050] Environmental adaptability: The system maintains stable performance in various industrial environments and has good adaptability to ambient light changes, temperature fluctuations and mechanical vibrations.

[0051] This invention can be used to control the quality of stamping parts in the automotive, aviation, and electronics industries, improving production efficiency and product quality. Future applications could be expanded, such as integration into flexible production lines for online testing, integration with digital twin technology for full lifecycle quality traceability, and expansion to high-precision measurement of other workpiece types.

[0052] To better understand the present invention, the present invention provides the following calculation process to verify the specific process of the stamping parts intelligent measurement method and system that integrates dual-frequency phase-shifted fringe projection and spectral reflectance analysis described in the present invention: 1. Initial parameter setting: For a measurement example of a highly reflective aluminum alloy stamping part, set the following initial parameters: System hardware parameters: DMD projector resolution: 1920×1080 pixels; industrial camera resolution: 4096×3072 pixels, pixel size: 3.45μm; baseline distance between camera and projector (B): 400mm; camera principal point coordinates (x0, y0): (2048, 1536) pixels; camera equivalent focal length parameters fx=fy=4640 pixels; reference plane pixel displacement (d0): 5.0 pixels; working distance: 500mm.

[0053] Fringe projection parameters: low-frequency fringes fL = 12 / screen width; high-frequency fringes fH = 96 / screen width; frequency ratio fH / fL = 8; phase shift step number N = 4; standard fringe contrast mbase = 0.8; standard average light intensity I0base = 180 (0-255 grayscale levels).

[0054] Multi-wavelength illumination parameters: LED wavelength and initial intensity: 460nm: I460=150; 520nm: I520=180; 590nm: I590=170; 630nm: I630=160; 850nm: I850=200; baseline exposure time t0=30ms.

[0055] Algorithm parameters: Adaptive bilateral filtering parameters: Spatial domain Gaussian kernel parameter σ s =2.0; range Gaussian kernel reference parameter σ rbase =0.2; phase local optimization parameters: neighborhood size: 5×5 pixels; balance parameter λ=0.1; SVM kernel function parameter γ=1.0.

[0056] 2. Specific calculation process 1. Material recognition and parameter adaptive adjustment First, image acquisition is performed under multi-wavelength spectral illumination. The system sequentially activates LED light sources at 460nm, 520nm, 590nm, 630nm, and 850nm, capturing images of highly reflective aluminum alloy stampings at each wavelength. Simultaneously, a reference image of a standard whiteboard is captured under the same conditions.

[0057] Take a typical point (x,y)=(1500,2000) on the stamping surface, and the image grayscale value at each wavelength is: I460(1500,2000)=195; I520(1500,2000)=220; I590(1500,2000)=235; I630(1500,2000)=245; I850(1500,2000)=228; The corresponding grayscale value of the whiteboard reference image: I460 0 (1500,2000)=230; I520 0 (1500,2000)=240; I590 0 (1500,2000)=245; I630 0 (1500,2000)=250; I850 0 (1500,2000)=235; According to the relative reflectivity calculation formula, the relative reflectivity is calculated for each wavelength: R460(1500,2000)=I460(1500,2000) / I460 0 (1500,2000)=195 / 230=0.848; R520(1500,2000)=I520(1500,2000) / I520 0 (1500,2000)=220 / 240=0.917; R590(1500,2000)=I590(1500,2000) / I590 0 (1500,2000)=235 / 245=0.959; R630(1500,2000)=I630(1500,2000) / I630 0 (1500,2000)=245 / 250=0.980; R850(1500,2000)=I850(1500,2000) / I850 0 (1500,2000)=228 / 235=0.970; Form the reflectivity vector for this point: R(1500,2000)=[0.848,0.917,0.959,0.980,0.970].

[0058] To perform principal component analysis (PCA), the system builds a covariance matrix based on a large number of pre-collected samples: C=[0.0058,0.0042,0.0037,0.0033,0.0028;0.0042,0.0063,0.0045,0.0039,0.0031;0.0037,0.0045,0. 0072,0.0055,0.0039;0.0033,0.0039,0.0055,0.0068,0.0047;0.0028,0.0031,0.0039,0.0047,0.0075].

[0059] Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues and eigenvectors: Eigenvalues: λ=[0.0237,0.0053,0.0024,0.0013,0.0009] Take the first two eigenvectors to form a dimensionality reduction matrix: Φ=[0.4107,0.5216;0.4532,0.2768;0.4721,-0.1345;0.4603,-0.4582;0.4326,-0.6573].

[0060] Perform dimensionality reduction: S=Ф T ·RS=[0.4107,0.4532,0.4721,0.4603,0.4326;0.5216,0.2768,-0.1345,-0.4582,-0.6573] [0.848, 0.917, 0.959, 0.980, 0.970] T .

[0061] Compute the first principal component: S1=0.4107×0.848+0.4532×0.917+0.4721×0.959+0.4603×0.980+0.4326×0.970=0.3483+0.4156+0.4527+0.4511+0.4196=2.0873.

[0062] Compute the second principal component: S2=0.5216×0.848+0.2768×0.917-0.1345×0.959-0.4582×0.980-0.6573×0.970=0.4423+0.2539-0.1290-0.4490-0.6376=-0.5194.

[0063] The eigenvector is obtained: S=[2.0873,-0.5194].

[0064] Input the feature vector to the pre-trained SVM classifier: SVM decision function: f(S)=sign(∑α i y i K(S,S i )+b); For a support vector S i =[2.1,-0.48], calculate the kernel function value: K(S,S i )=exp(-γ||SS i || 2 )

[0065] =exp(-1.0×||(2.0873,-0.5194)-(2.1,-0.48)|| 2 ) =exp(-1.0×√((2.0873-2.1) 2 +(-0.5194-(-0.48)) 2 ) 2 ) =exp(-1.0×√((-0.0127) 2 +(-0.0394) 2 ) 2 ) =exp(-1.0×√(0.00016+0.00155) 2 ) =exp(-1.0×(0.0395) 2 )=exp(-0.00156)=0.9984.

[0066] By calculating the weighted sum of multiple support vectors, the final SVM classifier output is polished aluminum alloy with material parameters M = [0.95, 0.02], where the first component represents reflectivity and the second component represents surface roughness (Ra, micrometers).

[0067] Based on the identified material parameters, the system dynamically adjusts the measurement parameters: Stripe Contrast: m adaptive =m base ·f reflect (M)=0.8×(1-0.7M[0])=0.8×(1-0.7×0.95)=0.8×(1-0.665)=0.8×0.335=0.268.

[0068] Average light intensity: I 0adaptive =I 0base ·greflect (M)=180×(1-0.6M[0])=180×(1-0.6×0.95)=180×(1-0.57)=180×0.43=77.4≈77.

[0069] Exposure time: t exp =t0·αM=30×(0.5-0.2M[0])=30×(0.5-0.2×0.95)=30×(0.5-0.19)=30×0.31=9.3ms.

[0070] Adaptive bilateral filtering parameters: σ r =σ rbase ·(1+3M[1])=0.2×(1+3×0.02)=0.2×1.06=0.212.

[0071] The above parameter adjustments reflect the adaptation to highly reflective materials: reducing stripe contrast and light intensity to avoid overexposure, shortening exposure time to reduce saturated areas, and slightly increasing filtering parameters to preserve subtle surface features.

[0072] 2. Dual-Frequency Phase-Shifted Fringe Projection and 3D Reconstruction Using the optimized parameters, the system projects low-frequency and high-frequency phase-shifted fringes and acquires deformed fringe images.

[0073] Four-step phase-shift image of low-frequency stripes (fL=12 stripes / screen width) Grayscale value at point (1500,2000): I L0 (1500,2000)=125; I L1 (1500,2000)=82; I L2 (1500,2000)=36; I L3 (1500,2000)=76.

[0074] Four-step phase-shift image of high-frequency stripes (fH=96 stripes / screen width) Grayscale value at point (1500,2000): I H0 (1500,2000)=128; I H1 (1500,2000)=86; I H2 (1500,2000)=38; I H3 (1500,2000)=77.

[0075] Calculate the low-frequency wrapping phase: φ Lw (1500,2000)=tan -1 [(I L3 -I L1 ) / (I L0 -I L2 )] =tan -1 [(76-82) / (125-36)]=tan -1 [(-6) / 89]=tan -1 [-0.0674]=-0.0673rad.

[0076] Calculate the high-frequency wrapping phase: φ Hw (1500,2000)=tan -1 [(I H3 -I H1 ) / (I H0 -I H2 )] =tan -1 [(77-86) / (128-38)]=tan -1 [(-9) / 90]=tan -1 [-0.1]=-0.0997rad.

[0077] Perform two-frequency phase unwrapping: k(1500,2000)=round[(f H / f L ·φL w -φH w ) / (2π)] =round[(96 / 12·(-0.0673)-(-0.0997)) / (2π)] =round[(-0.5384-(-0.0997)) / (2π)] =round[(-0.4387) / (2π)]=round[-0.0698]=0 Absolute phase: φ(1500,2000)=φH w +2πk=-0.0997+2π·0=-0.0997rad.

[0078] Convert to pixel displacement: d(1500,2000)=φ(1500,2000)·p / (2π) =-0.0997·(1920 / 96) / (2π)=-0.0997·20 / 6.2832=-0.3174 pixels Calculate 3D coordinates: Z(1500,2000)=B / (d(1500,2000)-d0) =400 / (-0.3174-5.0)=400 / (-5.3174)=-75.2248mm.

[0079] A negative value indicates that the point is below the reference plane. In actual use, the absolute value is 75.2248 mm.

[0080] X(1500,2000)=Z(1500,2000)·(1500-x0) / f x =-75.2248·(1500-2048) / 4640=-75.2248·(-548) / 4640=-75.2248·(-0.1181)=8.8841mm; Y(1500,2000)=Z(1500,2000)·(2000-y0) / f y =-75.2248·(2000-1536) / 4640=-75.2248·464 / 4640=-75.2248·0.1=-7.5225mm.

[0081] Therefore, the three-dimensional coordinates of the point (1500, 2000) are (8.8841, -7.5225, 75.2248) mm.

[0082] 3. Phase optimization processing of highly reflective areas For the point (2500, 1800) in the highly reflective area, the directly calculated phase is unreliable.

[0083] Initial phase estimates for point (2500, 1800) and its 5×5 neighborhood (partially listed): φ estimated (2498,1798)=-0.2314rad φ estimated (2499,1798)=-0.2103rad φ estimated (2500,1798) = invalid data (oversaturated) φ estimated (2501,1798) = invalid data (oversaturated) φ estimated (2502,1798)=-0.1872rad... φestimated (2500,1800) = invalid data (oversaturated)... Apply the phase local optimization algorithm: φ optimized (x,y)=argmin φ [∑(i∈N(x,y))w i ·(φ(i)-φ estimated (i)) 2 +λ·R(φ)]; Since the data of point (2500,1800) is invalid, its phase needs to be estimated from the neighborhood. Set the weight coefficient w i Related to pixel reliability, reliable data points have a weight of 1.0 and invalid data points have a weight of 0.0. The regularization term R(φ) can be the second-order derivative of the phase to ensure smoothness.

[0084] By solving the minimum value of the objective function, the optimized phase is obtained: optimized (2500,1800)=-0.1923rad.

[0085] This result is more accurate than the value obtained by simply averaging the neighborhood (-0.1911 rad), mainly because it takes into account the local continuity of the surface.

[0086] 4. Shape-Material Information Fusion For a material boundary region: For example, at point (3000, 2200), calculate the spatial gradient of the material parameters: M[0](2999,2200)=0.95(reflectivity) M[0](3001,2200)=0.72(reflectivity) M[1](2999,2200)=0.02 (surface roughness) M[1](3001,2200)=0.20 (surface roughness) Calculate the reflectivity gradient:

[0087] =(M[0](3001,2200)-M[0](2999,2200)) / 2=(0.72-0.95) / 2=-0.115.

[0088] Calculate the surface roughness gradient: =(M[1](3001,2200)-M[1](2999,2200)) / 2=(0.20-0.02) / 2=0.09.

[0089] Similarly, the y-direction gradient is calculated: =0.002 =-0.003 Gradient norm of material parameters: ||▽M(3000,2200)|| =√(( ) 2 +( ) 2 +( ) 2 +( ) 2 ) =√((-0.115) 2 +(0.002) 2 +(0.09) 2 +(-0.003) 2 ) =√(0.013225+0.000004+0.0081+0.000009)=√0.021338=0.1461.

[0090] Calculate marginal probability: Edge probability (3000,2200)=||▽M(3000,2200)||=0.1461.

[0091] Based on a set threshold (e.g., 0.1), the point is identified as a material boundary point, and the system uses denser fringe projection and more cautious phase unwrapping in this area.

[0092] Constructing the shape-material fusion feature vector: Feature fusion =[S,H] Where S is the spectral eigenvector [2.072, -0.315], H is the morphological eigenvector, and the surface curvature k at this point is 0.0025 mm. -1 , normal vector n=[0.03,-0.02,0.999].

[0093] The complete fused feature vector: Feature fusion (3000,2200)=[2.072,-0.315,0.0025,0.03,-0.02,0.999] The fused feature vector is used for subsequent precise edge positioning and feature recognition.

[0094] 5. Adaptive bilateral filtering Apply adaptive bilateral filtering to the acquired 3D point cloud data, Take the point (1500, 2000) as an example: define a 5×5 filter window; the original height value Z(x+p, y+q) within the part of the window is: Z(1498,1998)=75.243mm; Z(1499,1998)=75.238mm; Z(1500,1998)=75.232mm; Z(1501,1998)=75.235mm; Z(1502,1998)=75.241mm...Z(1500,2000)=75.225mm... Calculate the spatial domain Gaussian kernel: For point (1498,1998): p=1498-1500=-2; q=1998-2000=-2; G s (-2,-2)=exp(-(-2) 2 +(-2) 2 ) / (2×2.0 2 ))=exp(-8 / 8)=exp(-1)=0.3679.

[0095] For the point (1499,1998): G s (-1,-2)=exp(-(((-1) 2 +(-2) 2 ) / (2×2.0 2 )) =exp(-5 / 8)=exp(-0.625)=0.5353, The spatial kernel values of other points are calculated similarly.

[0096] Calculate the range Gaussian kernel, taking into account the material-related parameters σ r =0.212: For the point (1498,1998): ΔZ=Z(1498,1998)-Z(1500,2000) =75.243-75.225=0.018mm.

[0097] G r (0.018)=exp(-(0.018) 2 / (2×0.212 2 )) =exp(-0.000324 / 0.08996)=exp(-0.0036)=0.9964.

[0098] For the point (1499,1998): ΔZ=Z(1499,1998)-Z(1500,2000) =75.238-75.225=0.013mm.

[0099] G r (0.013)=exp(-(0.013) 2 / (2×0.212 2 )) =exp(-0.000169 / 0.08996)=exp(-0.0019)=0.9981.

[0100] Calculate the filtered height value: Z'(1500,2000)=∑G s (p,q)·G r (ΔZ)·Z(1500+p,2000+q) / ∑G s (p,q)·G r (ΔZ); Molecular part (partial enumeration): 0.3679×0.9964×75.243+0.5353×0.9981×75.238+...=75.231×∑G s (p,q)·G r (ΔZ).

[0101] The final filtered height value Z'(1500,2000)=75.231mm is compared with the original value 75.225mm. The noise is effectively suppressed while the surface features are preserved.

[0102] 6. Multi-angle data collection and fusion For specific areas on stamped parts (such as the edge of a hole), the system acquires data from three different angles: Angle 1 (front view): Z1(3500,1500)=62.536mm, measurement uncertainty σ1=0.015mm; Angle 2 (15° to the left): Z2(3500,1500)=62.512mm, measurement uncertainty σ2=0.023mm; Angle 3 (20° to the right): Z3(3500,1500)=62.548mm, measurement uncertainty σ3=0.018mm; Define weights based on measurement uncertainty: w1=1 / σ1 2 =1 / (0.015) 2=4444.44; w2=1 / σ2 2 =1 / (0.023) 2 =1890.36w3=1 / σ3 2 =1 / (0.018) 2 =3086.42.

[0103] Calculate the weighted average fusion height value: Z fused =(w1×Z1+w2×Z2+w3×Z3) / (w1+w2+w3) =(4444.44×62.536+1890.36×62.512+3086.42×62.548) / (4444.44+1890.36+3086.42) =(277935.54+118165.78+192551.57) / 9421.22=588652.89 / 9421.22=62.535mm.

[0104] The fused data is more accurate than the data from a single angle, and the measurement uncertainty is reduced to: σ fused =1 / √(w1+w2+w3)=1 / √9421.22=0.010mm.

[0105] 7. Feature extraction and dimension measurement Assume there is a circular hole on a stamped part. 30 edge points are extracted from the point cloud data. The following are the 3D coordinates of some of these edge points (unit: mm): P1=(45.23,60.12,62.54); P2=(45.36,60.78,62.52); P3=(45.65,61.32,62.53)...P 30 =(44.86,59.53,62.55).

[0106] Use the least squares method to fit a circle and solve the system of equations: ∑(x i -x c ) 2 +(y i -y c ) 2 =r 2 (i=1,2,...,30); The fitting result gives the coordinates of the circle center (x c ,y c ,z c)=(45.05,60.02,62.53)mm, radius r=1.50mm.

[0107] Therefore, the diameter of the hole D = 2r = 3.00 mm, and the standard error (fitting residual) σ fit =0.012mm. Compared with the standard value of 3.02mm in the CAD model, the deviation is -0.02mm, which is better than the tolerance requirement of ±0.03mm.

[0108] 8. Results Analysis Through the above calculation process, we can draw the following conclusions: Accuracy of material identification: The system successfully identified the test piece as polished aluminum alloy, with a reflectivity of 0.95 and a surface roughness Ra of 0.02μm. These parameters closely matched the actual material characteristics, validating the effectiveness of the spectral reflectance analysis algorithm.

[0109] Adaptive parameter optimization: Based on the identified material properties, the system reduced fringe contrast from 0.8 to 0.268, average light intensity from 180 to 77, and exposure time from 30ms to 9.3ms. These adaptive adjustments effectively avoided overexposure and data loss on highly reflective surfaces, laying the foundation for subsequent 3D reconstruction.

[0110] Reliability of dual-frequency phase unwrapping: The calculation process shows that the dual-frequency phase unwrapping algorithm successfully obtains the absolute phase value and maintains high stability even on highly reflective surfaces. The key to phase unwrapping is to reasonably set the frequency ratio (8 in this example) to satisfy f H / f L <2N condition to ensure unambiguous unpacking.

[0111] High-reflective area processing effect: For oversaturated high-reflective areas, the phase local optimization algorithm successfully restored the phase data, filling the data gaps in the traditional method. The optimized phase value φ optimized (2500,1800)=-0.1923rad has a high reliability, which verifies the adaptability of the algorithm to highly reflective areas.

[0112] Advantages of shape-material fusion: At the material boundary, the gradient-based edge probability calculation successfully detects the material change area. fusion The integration of spectral and morphological features improves the accuracy of feature recognition, especially in areas with material changes.

[0113] Effect of adaptive bilateral filtering: The filtering process optimizes the height value of point (1500, 2000) from 75.225mm to 75.231mm, suppressing measurement noise while retaining surface detail features. Material-related filter parameter adjustment (σ r =0.212) ensures the matching of filtering intensity and surface characteristics.

[0114] Improved multi-angle data fusion: Weighted fusion of data from three perspectives reduces measurement uncertainty from the best single-view value of 0.015mm to 0.010mm, improving accuracy by approximately 33% while also enhancing data integrity.

[0115] Measurement accuracy verification: The circular hole diameter measurement result is 3.00mm, with a deviation of only -0.02mm from the standard value of 3.02mm, and a fitting residual of 0.012mm, which is far superior to the ±0.05mm accuracy level of traditional methods and meets the measurement requirements of precision stamping parts.

[0116] In summary, this system performs exceptionally well in measuring highly reflective aluminum alloy stampings. Through adaptive parameter adjustment guided by material recognition, it solves the measurement challenges of highly reflective surfaces, a problem traditional methods struggle to address. The integration of dual-frequency phase-shifted fringe projection and spectral reflectance analysis creates a significant synergistic effect, enabling the system to measure stampings made of a variety of complex materials. Its high precision, efficiency, and adaptability demonstrate broad application value and potential.

[0117] The foregoing is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A stamping parts measurement method based on dual-frequency fringe projection and spectral material recognition, characterized in that: include: Generate a phase-shifted pattern of low-frequency stripes and high-frequency stripes, where the frequency of the low-frequency stripes is 8-16 per screen width and the frequency of the high-frequency stripes is 64-128 per screen width; Collect deformation stripe images on the surface of the stamping part being tested; Calculate the wrapping phase based on the deformed fringe image; Perform dual-frequency phase unwrapping; Multispectral images of stamping parts are collected through multi-wavelength spectral illumination to calculate relative reflectivity; Perform principal component analysis to reduce the dimensionality of multi-wavelength reflectance data; Identify the material of stamping parts based on spectral characteristics and dynamically adjust fringe projection parameters based on the identification results; Map the unwrapped phase into three-dimensional coordinates; Perform adaptive bilateral filtering on the acquired three-dimensional point cloud data; Extract characteristic points, lines, and surfaces of stamping parts, calculate key dimensions, and compare them with the standard model.

2. The measuring method according to claim 1, wherein: The mathematical model of the stripe pattern is: pn (x,y)=I0[1+m·cos(2πfx+φ0+2πn / N)], where I0 is the average light intensity, m is the fringe contrast, f is the fringe frequency, φ0 is the initial phase, n is the number of phase shift steps, and N is the total number of phase shift steps; The image mathematical model is: I n (x,y)=a(x,y)+b(x,y)·cos[φ(x,y)+2πn / N], where a(x,y) is the background light intensity, b(x,y) is the reflected modulated light intensity, and φ(x,y) is the phase containing the object height information; The formula for calculating the wrapping phase is: w (x,y)=tan^(-1)[∑(n=0 to N-1)I n (x,y)·sin(2πn / N) / ∑(n=0 to N-1) I n (x,y)·cos(2πn / N)]; The unwrapping formula is: φ(x,y)=φ H (x,y)+2πk(x,y), where k(x,y)=round[(f H / f L ·φ L (x,y)-φ H (x,y)) / (2π)],f H and f L are the fringe frequencies of high and low frequencies, respectively, H (x,y) and φ L (x,y) are the high-frequency and low-frequency wrapping phases, respectively; The relative reflectivity calculation formula is: R λ (x,y)=I λ (x,y) / I λ ^0(x,y), where I λ (x, y) is the image of the stamping under wavelength λ, I λ ^0(x,y) is the image of a standard whiteboard under the same conditions; The principal component analysis dimensionality reduction formula is: S = Φ^T·R, where Φ is the principal component eigenvector matrix and R is the original spectral reflectance vector; Dynamically adjust the fringe projection parameters to: m adaptive =m base ·f reflect (M), I 0adaptive =I 0base ·g reflect (M), where M is the material parameter, f reflect 、g reflect and α M is an adjustment function based on material properties; The three-dimensional coordinate formula is: Z(x,y)=B / (d(x,y)-d0), X(x,y)=Z(x,y)(x-x0) / f x ,Y(x,y)=Z(x,y)(y-y0) / f y , where B is the baseline distance between the camera and the projector, d0 is the pixel displacement corresponding to the reference plane, (x0, y0) is the coordinate of the camera principal point, and f x and f y is the equivalent focal length parameter of the camera; The adaptive bilateral filtering formula is: Z'(x,y)=∑G s (p,q)G r (Z(x+p,y+q)-Z(x,y),M)Z(x+p,y+q) / ∑G s (p,q)G r (Z(x+p,y+q)-Z(x,y),M), where G s and G r are the spatial domain and value range Gaussian kernels respectively.

3. The measuring method according to claim 2, wherein: The method further includes performing phase local optimization on a high-reflective material area, and the optimization formula is: φ optimized (x,y)=argmin φ [∑(i∈N(x,y))w i ·(φ(i)-φ estimated (i)) 2 +λ·R(φ)], where N(x,y) is the neighborhood of point (x,y), w i is the weight coefficient, R(φ) is the regularization term, and λ is the balance parameter.

4. The measuring method according to claim 2, wherein: The method also includes performing shape-material information fusion to predict edge positions: probability (x,y)=||▽M(x,y)||, where ▽M(x,y) is the spatial gradient of the material parameters, and the morphology-material fusion feature vector is constructed: Feature fusion =[S,H], where S is the spectral feature vector and H is the morphological feature vector.

5. The measuring method according to claim 2, wherein: In the material identification, the support vector machine classification model is used: f(S)=sign(∑α i y i K(S,S i )+b), where S i is the support vector, y i is the class label, α i is the Lagrange multiplier, b is the bias term, and K is the radial basis function kernel: K(S,S')=exp(-γ||S-S'||^2).

6. The measurement method according to claim 2, wherein: In the dual-frequency phase unwrapping, the ratio of the high frequency to the low frequency satisfies: f H / f L < N / 2 to achieve unambiguous phase unwrapping.

7. The measurement method according to claim 2, wherein: The wavelengths used in the multi-wavelength spectrum lighting include 460nm blue light, 520nm green light, 590nm yellow light, 630nm red light and 850nm near infrared light. The lighting spectrum characteristics are expressed as: S(λ)=∑I i ·δ(λ-λ i ), where I i is the intensity of the i-th wavelength channel, λ i is the center wavelength.

8. A measurement system for stamping parts using the dual-frequency fringe projection and spectral material recognition method according to any one of claims 1 to 7, characterized in that: include: Multi-wavelength structured light source is used to generate phase-shifted fringe patterns and multi-wavelength illumination. The mathematical model of the generated phase-shifted fringe pattern is: pn (x,y)=I0[1+m·cos(2πfx+φ0+2πn / N)]; High-resolution industrial cameras for acquiring deformed fringe images and multispectral images; 3D shape reconstruction module, used to calculate the wrapping phase: φ based on the deformed fringe image w (x,y)=tan^(-1)[∑(n=0to N-1)I n (x,y)·sin(2πn / N) / ∑(n=0 to N-1)I n (x,y)·cos(2πn / N)], perform dual-frequency phase unwrapping: φ(x,y)=φ H (x,y)+2πk(x,y), where k(x,y)=round[(f H / f L ·φ L (x,y)-φ H (x,y)) / (2π)] and map the phase into three-dimensional coordinates; Material recognition module, used to calculate relative reflectance based on multispectral images: R λ (x,y)=I λ (x,y) / I λ ^0(x,y), perform principal component analysis to reduce the dimension: S = Φ^T·R, and identify the material of the stamping part; Parameter adaptive module, used to dynamically adjust fringe projection and image acquisition parameters according to material recognition results; Point cloud processing module for performing adaptive bilateral filtering: Z'(x,y)=∑G s (p,q)G r (Z(x+p,y+q)-Z(x,y),M)Z(x+p,y+q) / ∑G s (p,q)G r (Z(x+p,y+q)-Z(x,y),M); Feature extraction and dimension measurement module, used to extract stamping part features and calculate key dimensions.

9. The measurement system according to claim 8, characterized in that: The multi-wavelength structured light source adopts digital micromirror device (DMD) technology to realize programmable light source projection and wavelength switching, and the switching time is less than 50ms.

10. The measurement system according to claim 8, characterized in that: The system also includes a self-learning module for recording the optimal measurement parameters of stamping parts made of different materials and establishing a mapping relationship between spectral characteristics and material parameters: M=f(S), where M is the material parameter vector, S is the spectral characteristic vector, and f is a multivariate nonlinear regression model or a neural network model.

Images