A self-insulation concrete wall real stone paint flatness image detection method and system

Abstract

The present application provides a kind of self-insulation concrete wall real stone paint flatness image detection method and system, specifically, composite structured light pattern is projected to the wall surface to be measured, the pattern is logarithmic frequency modulation sinusoidal stripe as high-resolution phase carrier, Gray code of two-dimensional absolute position information is superimposed and encoded in chroma channel, deformed pattern image is synchronously collected, frequency response confidence map is generated based on the image, global energy function containing confidence weight and smoothing constraint is constructed, high-precision absolute phase map is obtained by solving, and is converted into initial three-dimensional point cloud, high-fidelity three-dimensional point cloud is obtained by weighted nonlinear optimization, the three-dimensional point cloud is converted into depth height map, is decomposed to the scale corresponding to the detection standard by the two-dimensional discrete wavelet transform associated with physical scale, the L1 norm of key scale wavelet detail component is calculated, macro fluctuation and surface texture energy coefficient are separated, and the quantitative representation of wall surface macro flatness deviation is realized.

Description

Technical Field

[0001] This application belongs to the field of testing, and in particular relates to a method and system for detecting the flatness of self-insulating concrete wall with stone-like paint. Background Technology

[0002] In the field of building construction, the construction quality of wall finishing projects, especially the smoothness of the surface, is one of the core indicators for evaluating the quality and aesthetic effect of the project. Self-insulating concrete composite walls are widely used due to their excellent thermal insulation performance. However, manufacturing and installation errors in the self-insulating wall substrate, as well as process fluctuations during the application of stone-like paint, can lead to macroscopic undulations, waves, and other smoothness deviations on the wall surface, seriously affecting the aesthetic value and service life of the building. Traditional wall smoothness testing relies on manual sampling using a 2-meter straightedge and feeler gauge. This method is not only labor-intensive and inefficient, but the measurement results are also highly subjective, depending on the experience of the inspectors, and the sparse data provided cannot comprehensively represent the overall smoothness of a large area of ​​the wall.

[0003] Structured light technology projects a specially encoded optical pattern onto the surface of a target object and uses a camera to capture the highly modulated deformation pattern of the object's surface. A decoding algorithm then reconstructs the three-dimensional shape of the object's surface. The surface of real stone paint has rich colored particles and complex, uneven textures, resulting in highly non-uniform optical properties in space. This severely interferes with the imaging quality of the projected pattern, reduces the signal-to-noise ratio in local areas, and thus affects the accuracy and reliability of phase calculations. Existing structured light encoding strategies, whether time-based encoding methods requiring multiple pattern projections or spatial encoding methods requiring a single projection, cannot reliably separate and accurately extract phase noise caused by the dense high-frequency texture of the real stone paint itself and macroscopic phase changes caused by large-scale wall undulations. Conventional filtering methods easily blur the boundary between macroscopic undulations and microscopic textures, leading to distorted flatness assessment results and failing to accurately represent the fundamental macroscopic flatness deviations of the wall that truly reflect construction quality. Summary of the Invention

[0004] This disclosure provides a method and system for detecting the flatness of self-insulating concrete walls with stone-like paint.

[0005] The first aspect of this disclosure provides a method for detecting the flatness of a self-insulating concrete wall with stone-like paint, comprising the following steps:

[0006] A set of composite structured light patterns is projected onto the surface of the self-insulating concrete wall with real stone paint. The composite structured light pattern uses frequency-modulated sinusoidal stripes with a logarithmic spatial frequency as a high-resolution phase information carrier, and Gray code encoding two-dimensional absolute position information is superimposed on the chromaticity channel of the phase information carrier. The image of the deformed pattern modulated by the surface of the wall under test is simultaneously acquired by the imaging device.

[0007] The three-dimensional point cloud data of the wall surface is reconstructed based on the deformed pattern image;

[0008] The three-dimensional point cloud data is converted into a regular gridded depth-height map. A two-dimensional discrete wavelet transform associated with physical scale is performed on the height map to decompose it into multiple scales corresponding to key spatial wavelengths in the wall flatness detection standard. By calculating the L1 norm of the wavelet detail component coefficients within the preset key scale frequency band, the sparse energy coefficients representing macroscopic large-format undulations are separated from the energy coefficients representing the dense high-frequency texture and local singular defects of the real stone paint itself, thereby achieving a quantitative representation of the fundamental macroscopic flatness deviation of the wall.

[0009] The second aspect of this disclosure provides an image detection system for the flatness of a self-insulating concrete wall with stone-like paint, comprising the following modules:

[0010] The acquisition module is used to project a set of composite structured light patterns onto the surface of the self-insulating concrete wall with real stone paint. The composite structured light pattern uses frequency-modulated sinusoidal stripes with a logarithmic spatial frequency as a high-resolution phase information carrier, and Gray code encoded with two-dimensional absolute position information is superimposed on the chromaticity channel of the phase information carrier. The imaging device synchronously acquires the deformed pattern image modulated by the surface of the wall to be tested.

[0011] The construction module is used to reconstruct three-dimensional point cloud data of the wall surface based on the deformed pattern image;

[0012] The calculation module is used to convert the three-dimensional point cloud data into a regular gridded depth-height map, perform a two-dimensional discrete wavelet transform on the height map that is associated with the physical scale, and decompose the height map into multiple scales corresponding to the key spatial wavelengths in the wall flatness detection standard; by calculating the L1 norm of the wavelet detail component coefficients in the preset key scale frequency band, the sparse energy coefficients representing macroscopic large-format undulations are separated from the energy coefficients representing the dense high-frequency texture and local singular defects of the real stone paint itself, so as to achieve a quantitative representation of the macroscopic fundamental flatness deviation of the wall.

[0013] This invention proposes an image detection method for the flatness of self-insulating concrete walls with stone-like paint. It integrates high-resolution phase information and absolute position encoding into a composite structured light pattern through a single projection, and constructs a global optimization strategy weighted by the local measurement signal-to-noise ratio. This strategy suppresses interference from complex textures and uneven reflections on the stone-like paint surface, achieving high-fidelity reconstruction of the wall's three-dimensional morphology. Utilizing multi-scale wavelet transform analysis technology associated with physical scales, it accurately separates large-scale low-frequency fluctuations representing macroscopic flatness deviations from the inherent high-frequency textures and local singular defects of the stone-like paint coating. This enables precise and reliable quantitative detection of the wall's fundamental flatness, improving measurement accuracy on complex textured surfaces. Attached Figure Description

[0014] Figure 1 Flowchart of a method for detecting the flatness of a self-insulating concrete wall with stone-like paint.

[0015] Figure 2 This is a schematic diagram of the spatial frequency distribution curve. Detailed Implementation

[0016] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0017] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0018] Firstly, this disclosure proposes an image detection method for the flatness of self-insulating concrete walls coated with stone-like paint, such as... Figure 1 As shown, it includes the following steps:

[0019] S1. A set of composite structured light patterns is projected onto the surface of the self-insulating concrete wall with stone paint to be tested. The composite structured light pattern uses frequency-modulated sinusoidal stripes with a logarithmic spatial frequency as a high-resolution phase information carrier, and Gray code encoding two-dimensional absolute position information is superimposed on the chromaticity channel of the phase information carrier. The image of the deformed pattern modulated by the surface of the wall to be tested is simultaneously acquired by the imaging device.

[0020] Composite structured light patterns are generated using numerical computation software such as MATLAB or the NumPy library in Python. The brightness channels of these patterns... pixel values In the horizontal direction u, it is obtained from the following formula: Where A is the ambient light intensity and B is the stripe contrast. The initial spatial frequency is given by k, the rate of frequency change is given by k, and the pattern width is given by W. N sets of binary code patterns are generated for construction. Construction of horizontal Gray code and M-group binary code patterns The vertical Gray code is generated and converted using Gray code encoding rules such as standard reflective Gray code. The frequency-modulated sinusoidal stripe pattern is placed in the blue (B) channel of the RGB image, the horizontal Gray code pattern sequence is sequentially placed in the red (R) channel, and the vertical Gray code pattern sequence is sequentially placed in the green (G) channel, merging them into a time-series color composite structured light pattern. This pattern sequence is then projected at high speed onto the wall surface using a digital light processing (DLP) projector. Hardware synchronization is achieved using trigger signals between the camera and the projector, capturing a color image sequence containing deformed stripes and Gray code information frame by frame.

[0021] In one embodiment, a set of composite structured light patterns is projected onto the surface of the self-insulating concrete wall with stone-like paint. The composite structured light patterns use frequency-modulated sinusoidal fringes with a logarithmic spatial frequency variation as high-resolution phase information carriers, and Gray code encoding two-dimensional absolute position information is superimposed on the chromaticity channel of the phase information carriers. The method includes:

[0022] Generate a set of frequency-modulated sinusoidal fringes with spatial frequencies that satisfy a logarithmic function along the x-axis, starting at a frequency of 0.05 periods per pixel and a logarithmic growth factor of 0.001;

[0023] Generate a set of 8-bit horizontal Gray codes and a set of 8-bit vertical Gray codes, and generate corresponding bit-plane patterns.

[0024] The frequency-modulated sine stripe is placed in the L* channel of the CIEL*a*b* color space. The encoded values ​​of each corresponding bit plane of the horizontal Gray code are mapped to the a* channel, and the encoded values ​​of each corresponding bit plane of the vertical Gray code are mapped to the b* channel. When the Gray code value in the bit plane is 0, the corresponding a* or b* channel value is the first mapping value, and when the Gray code value in the bit plane is 1, the corresponding a* or b* channel value is the second mapping value.

[0025] The images of the three channels L*, a*, and b* are synthesized to obtain a set of composite structured light patterns, which are then projected sequentially by a projection device.

[0026] A set of eight composite structured light patterns is generated in the image buffer of a digital micromirror device (DMD) or a liquid crystal on silicon (LCoS) projection system. A shared brightness L* channel image is generated for all patterns. This image is a frequency-modulated sinusoidal fringe pattern varying along the horizontal u-axis, such as... Figure 2 As shown. The intensity value of each pixel (u,v). Determined by the following formula: , where the phase function It is spatial frequency The integral, and the spatial frequency Where u is the horizontal coordinate of a pixel, ranging from, for example, 0 to 1919. This makes the stripes sparse on one side of the image and dense on the other, thus balancing noise resistance in low-frequency regions and detail resolution in high-frequency regions.

[0027] To achieve two-dimensional absolute position decoding in a single measurement, 8-bit standard Gray code is used to encode the horizontal and vertical directions respectively, generating 8 horizontal Gray code bit-plane patterns. to and 8 vertical Gray code position plane patterns to For the i-th composite structured light pattern (i = 1 to 8), the pixel value of the chroma a* channel is... According to the i-th horizontal Gray code position plane Map values: If If it is 0, then Set to -20; if If it is 1, then Set it to 20. Similarly, the chroma value is b * the pixel value of the channel. According to the i-th vertical Gray code position plane Map values: If If it is 0, then Set to -20; if If it is 1, then Set the value to 20. Combine the generated L*, a*, and b* single-channel images into a single CIEL*a*b* color image, and convert this color image to the RGB format required by the projection device. Repeat this process eight times to generate a set of eight composite patterns containing complete 8-bit Gray code information. Project these eight patterns sequentially onto the surface of the wall under test using a projection device such as a DLP projector with a resolution of 1920×1080 and a refresh rate of, for example, 60Hz. Simultaneously, use an industrial camera, such as a CMOS camera with a resolution of 2048×2048, synchronized with the projector to sequentially acquire eight images of the distorted patterns modulated by the wall surface.

[0028] S2, Based on the deformed pattern image, a frequency response confidence map representing the local measured signal-to-noise ratio is generated for each pixel by analyzing the local spectrum of the image.

[0029] Each acquired color deformed image undergoes channel separation to extract the frequency-modulated sinusoidal stripe image of the blue channel. A two-dimensional windowed Fourier transform (STFT) algorithm is applied to this stripe image. By setting a Gaussian window, for example, 31×31 pixels, the Fourier spectrum of the local region is calculated pixel by pixel on the image. Within the local spectrum of each pixel, the position and amplitude of the fundamental frequency peak are located; this peak corresponds to the spatial carrier frequency of the local stripe. The confidence score C(u,v) is the ratio of the fundamental frequency peak amplitude to the sum of the energy of other frequency components in the spectrum, or the ratio of the fundamental frequency peak amplitude to the average level of the local DC component and noise, thus representing the signal-to-noise ratio. The confidence scores of all pixels are combined to form a confidence map of the same size as the original image.

[0030] In one embodiment, the step of generating a frequency response confidence map representing the local measured signal-to-noise ratio for each pixel based on the deformed pattern image by analyzing the local spectrum of the image includes:

[0031] Select any pixel in the deformed pattern image, and apply a Hanning window of size 32×32 pixels centered on the pixel;

[0032] Perform a two-dimensional fast Fourier transform on the data within the window to obtain the local spectrum of the pixel;

[0033] Locate the energy peak of the fundamental frequency signal along the frequency variation axis in the local spectrum diagram;

[0034] The average energy of all spectral components except the DC component in the local spectrum is calculated as the average noise level.

[0035] The signal-to-noise ratio (SNR) is obtained by calculating the ratio of the peak energy of the fundamental frequency signal to the average noise level. This SNR is then used as the confidence level of the pixel. A complete frequency response confidence map is generated by traversing all pixels.

[0036] Pixel-by-pixel local spectral analysis is performed on any one of the eight acquired deformed patterns (usually the first one, or the image obtained by averaging the brightness channels of the eight images). A floating-point confidence map C(u,v) with all elements equal to zero is initialized with the same size as the deformed pattern. Each pixel p(u,v) in the deformed pattern is iterated over. For the current pixel, a rectangular neighborhood window of size 32×32 pixels is defined around it. To reduce spectral leakage, a two-dimensional Hanning window function w(i,j) is applied to the data within this window, in the form of... , where i and j are both in the range of 0 to 31.

[0037] Perform a two-dimensional Fast Fourier Transform (FFT) on the windowed 32×32 image data block to obtain the local complex spectrum of pixel p(u,v). Since the projected fringes vary along the u-axis, the fundamental frequency signal energy will be concentrated on the frequency coordinates. On the axis. Based on the known frequency function of the projection pattern. It can predict the approximate location of the fundamental frequency signal peak at the current pixel position u. .exist On the axis The maximum energy is searched within ±5 frequency units centered on the target. This is the fundamental frequency signal energy. Calculate the average noise level of this local spectrum. Noise energy. The spectrum excluding the DC component S(0,0) and the peak value of the fundamental frequency signal The average energy of all spectral components except for the pixel itself. Calculate the signal-to-noise ratio for that pixel. The SNR value is then stored as the confidence level of that pixel in the corresponding position of the confidence map C(u,v). After traversing all pixels, a complete frequency response confidence map representing the measurement quality of each pixel can be obtained.

[0038] S3, then construct a global energy function with the confidence map as the data item weight and the phase continuity between adjacent pixels as the smoothing constraint, and solve the function to obtain a global high-precision absolute phase map.

[0039] Using the results of the aforementioned windowed Fourier transform, the phase corresponding to the fundamental frequency peak of each pixel is extracted to obtain an initial wrapped phase map wrapped between -π and π. The acquired R and G channel image sequences were thresholded and binarized to decode the horizontal and vertical Gray code sequences, which were then converted into a stripe series sequence table. and A two-dimensional fringe series sequence table K(u,v) is constructed by looking up a table or by direct calculation; the initial absolute phase map is obtained by combining the wrapped phase and the fringe sequence table. Construct an energy function based on a Markov random field (MRF) model. The first item is the data item, where p represents the pixel index, and C(p) is the weight of the corresponding pixel in the confidence map. The phase to be solved is... The first term is the initial absolute phase. The second term is the smoothing term, where N is the neighborhood of pixel p. It is the coupling strength between adjacent pixels. This is the regularization parameter that balances the data terms and the smoothing term. The minimum value of this energy function is found using the GraphCuts algorithm or the iterative heavy-weight least squares method, yielding a globally optimal and spatially continuous high-precision absolute phase map. .

[0040] To obtain an absolute phase without 2π ambiguity through phase unwrapping using a global optimization method, in one embodiment, the construction of a global energy function using the confidence map as data item weights and the phase continuity between adjacent pixels as a smoothing constraint, and solving the function to obtain a high-precision global absolute phase map, includes:

[0041] Define an energy function for each pixel, consisting of a smoothing term weighted by a data term and a regularization coefficient;

[0042] The data item is determined by the confidence value of the pixel, the wrapping phase, the fringe index to be determined, and the coarse phase obtained by Gray code decoding;

[0043] The smoothing term is determined by the spatial distance constraint between adjacent pixels, the confidence value of adjacent pixels, the difference in fringe index, and the maximum cutoff value of the fringe index change.

[0044] The α-expansion image cut optimization algorithm is applied to iteratively minimize the overall energy function of the entire image until the energy function converges and is less than a preset threshold. The optimal fringe index of each pixel is output and the absolute phase is calculated.

[0045] Extracting the wrapped phase from the deformed pattern p represents a pixel, and a rough absolute phase is decoded using the chroma channel information of the eight patterns. This leads to a rough stripe index. The global energy function E(k) is the sum of the energies of all pixels p: ,in It is the precise fringe index of the pixel p to be solved. It is the regularization coefficient for balancing the data terms and the smoothing terms, set between 0.1 and 0.5, and N represents the four neighborhoods of pixel p.

[0046] Data Items Punishment candidate stripe index The deviation from the Gray code decoding result, weighted by the confidence plot C(p): Pixels with high confidence levels have higher weights in their data terms, making the result more similar to the Gray code decoded value. Smoothing term. Adjacent pixels are encouraged to have similar stripe indices, in the form of: Among them, weight d(p,q) is the spatial distance between adjacent pixels. These are parameters that control spatial attenuation, where C(p) and C(q) are the confidence scores of adjacent pixels. By eliminating negatively correlated weighting factors based on image grayscale differences, the smoothing constraint remains strong in high-frequency texture regions on the surface of real stone paint where the image intensity changes drastically due to the mottled material, as long as the frequency response confidence score is normal. This is the cutoff value, for example, set to 2, allowing large fringe index jumps only at real discontinuities such as object boundaries or macroscopic singularities with abrupt depth changes. To find the minimum of this energy function, the α-expansion graph cut optimization algorithm is used. Iteratively, a bipartite graph is constructed for each possible fringe index label α, and the optimal allocation between the current label α and the original labels of the pixel is found using the maximum flow minimum cut algorithm, thereby gradually reducing the total energy. The iteration continues until the relative rate of change of the total energy E(k) is less than [a certain value] after two consecutive iterations. Alternatively, the maximum number of iterations can be reached, such as 50. After optimization, the optimal stripe index for each pixel is output. Global high-precision absolute phase through Calculated.

[0047] S4. The absolute phase map is input into the pre-calibrated system model to calculate and generate an initial three-dimensional point cloud. Then, an objective function with the confidence map as the weight is constructed to perform weighted nonlinear optimization on the coordinates of the initial three-dimensional point cloud, thereby reconstructing high-fidelity three-dimensional point cloud data of the wall surface.

[0048] Using Zhang Zhengyou's camera calibration method, images of a checkerboard calibration board in different poses are captured. The intrinsic parameters and distortion coefficients of the camera are calculated using the `cv::findChessboardCorners` and `cv::calibrateCamera` functions from the OpenCV library. The projector is treated as an inverse camera, projecting a checkerboard pattern and being captured by the calibrated camera, thus calibrating the projector's intrinsic parameters. Then, by capturing an image of the calibration board simultaneously in the fields of view of both the camera and the projector, extrinsic parameters such as the rotation matrix and translation vector between them are calculated to establish a complete system geometric model. Based on this model, the absolute phase value is used... Phase-coordinate mapping is performed, and the 3D coordinates (X, Y, Z) of each camera pixel (u, v) in the world coordinate system are calculated using triangulation principles. All points are then aggregated to form an initial 3D point cloud. The nonlinear objective function is the sum of the reprojection errors of the 3D coordinates (X, Y, Z) of all point clouds, i.e. ,in and It is the projection function of the camera and projector. It is the image observation point. The projector coordinates are calculated based on the phase. The confidence weight is the value corresponding to that point; by minimizing the weighted objective function, the coordinates of all three-dimensional points are iteratively optimized to obtain the corrected high-fidelity three-dimensional point cloud.

[0049] In one embodiment, the construction of an objective function weighted by the confidence map is used to perform weighted nonlinear optimization on the coordinates of the initial 3D point cloud to reconstruct high-fidelity 3D point cloud data of the wall surface, including:

[0050] For each point in the 3D point cloud, a weighted nonlinear least squares objective function is constructed. The objective function consists of a data term weighted by the confidence of the corresponding pixel and a smoothing term weighted by the regularization coefficient.

[0051] The data item is used to penalize the deviation between the predicted phase calculated based on the three-dimensional points in a pre-calibrated system model and the corresponding measured phase.

[0052] The smoothing term is used to penalize the spatial discontinuity between a point and its neighboring 3D points;

[0053] The objective function is iteratively solved to optimize the coordinates of the three-dimensional points. The optimization stops when the change in the objective function is less than a preset threshold or the number of iterations reaches the upper limit. The output is a high-fidelity three-dimensional point cloud that suppresses the error in the low signal-to-noise ratio region.

[0054] The absolute phase map and pre-calibrated system parameters, including the camera intrinsic parameter matrix, are used to determine the system parameters. Distortion coefficient Projector intrinsic parameter matrix Distortion coefficient The initial 3D point cloud was obtained by calculating the rotation matrix R and translation vector T between the camera and the projector. Optimization objective function For all three-dimensional points It is constructed from a set of coordinates, and takes the following form: .

[0055] Data Items It is a weighted photometric reprojection error, used to ensure that the phase of the optimized 3D point in the projector coordinate system is consistent with the phase measured in the camera image. Specifically, it takes the following form: .in, It is a point Pixel coordinates in a camera image The corresponding weights were found from the confidence graph. It is the measured absolute phase. It is a three-dimensional point The process of projecting back to the projector pixel coordinates and calculating the predicted phase involves coordinate system transformation. The predicted phase value is calculated using the pinhole model of the projector and the phase encoding function. Smoothing term. Used to maintain the local smoothness of the surface and prevent overfitting noise in the point cloud, it takes the form of a Laplacian regularization term: ,in It is a point The set of neighborhood points in three-dimensional space. Regularization coefficient. The value is typically taken to be between 0.01 and 0.2. This nonlinear least squares problem is solved iteratively, updating the (X,Y,Z) coordinates of all points in each iteration to reduce... The termination condition for the iteration is set as follows: when the objective function... The decrease value is less than The optimization stops when the number of iterations exceeds 20. The output is 3D point cloud data with reduced measurement errors in low-confidence areas, smoother surface details, and higher accuracy.

[0056] S5, convert the three-dimensional point cloud data into a regular gridded depth height map, perform a two-dimensional discrete wavelet transform on the height map that is associated with the physical scale, and decompose the height map into multiple scales corresponding to the key spatial wavelengths in the wall flatness detection standard.

[0057] A reference plane is determined, for example, by performing plane fitting on point cloud data. A least-squares plane fitting algorithm implemented using Singular Value Decomposition (SVD) is commonly used. The perpendicular distance from each 3D point to the reference plane is calculated, and this distance is the height value. Create a uniform grid on the XY plane with a physical spacing of, for example, 1 mm. Interpolate the irregular point cloud data, such as using radial basis function interpolation or Kriging interpolation, to calculate the height value for each grid node, generating a regular depth-height map. Select an orthogonal wavelet basis with good time-frequency locality, such as the Daubechies4 (db4 wavelet). Use the PyWavelets library in Python or the wavelet toolbox in MATLAB to perform a multi-level two-dimensional discrete wavelet transform (DWT) on the depth-height map. Based on wall flatness inspection standards, such as a 2-meter straightedge, calculate the number of pixels corresponding to a 2-meter length in the depth map. Then the decomposition level j corresponding to the key spatial wavelength is approximately equal to This decomposes the transformation into multiple levels that include the key scale, resulting in a series of approximate component images and detail component images in the horizontal, vertical, and diagonal directions.

[0058] In one embodiment, converting the 3D point cloud data into a regular meshed depth-height map and performing a physical scale-dependent 2D discrete wavelet transform on the height map includes:

[0059] Based on the detection standards and the physical resolution calibrated by the system, the characteristic spatial wavelength used to represent local flatness fluctuations is determined to be 0.128 meters, which corresponds to 256 pixels on the depth-height map;

[0060] Symlet8 was selected as the wavelet basis function.

[0061] The wavelet decomposition layer is set to 8 to ensure that the lowest frequency scale of the decomposition matches the wavelength of the feature space.

[0062] An 8-layer two-dimensional discrete wavelet transform is performed on the regularly gridded depth-height map to obtain one low-frequency approximation component and eight sets of high-frequency detail components.

[0063] The optimized high-fidelity 3D point cloud data is converted into a regular gridded depth-height map H(u,v). This process is achieved by defining a uniform grid on the XY plane of the point cloud and averaging or median filtering the Z coordinate values ​​of all points within each grid cell. The grid resolution must match the physical resolution calibrated by the system. For example, by photographing a calibration board at a known distance, it can be seen that at typical detection distances, one pixel in the image corresponds to 0.5 mm in the physical world. Therefore, the characteristic spatial wavelengths of interest in the flatness inspection standards of the construction industry, such as undulations of 0.128 meters (128 millimeters), correspond to a scale of 256 pixels on the depth-height map.

[0064] To analyze the flatness of the wall at a specific scale, the Symlet8sym8 wavelet, which has good symmetry and tight support, was selected as the basis function. To separate features at the 256-pixel scale, the number of wavelet decomposition levels J was set such that... ≈256, therefore the number of decomposition layers is determined to be J=8. An 8-layer 2D-DWT (two-dimensional discrete wavelet transform) is performed on the depth height map H(u,v). The transform process is iterative: the first layer decomposes H(u,v) into a low-frequency approximate component A1 and three high-frequency detail components level. ,vertical diagonal The second-level decomposition takes A1 as input and further decomposes it into A2 and... This process continues until the 8th layer of decomposition is completed. This yields a minimum-sized low-frequency approximation component A8, and 8 sets of high-frequency detail component coefficient matrices with a total of 24 different scales and orientations. }. The detail components obtained from the j-th level decomposition mainly contain spatial dimensions within... arrive Feature information within a pixel range.

[0065] S6, by calculating the L1 norm of the wavelet detail component coefficients within the preset key scale frequency band, separates the sparse energy coefficients representing macroscopic large-scale undulations from the energy coefficients representing the dense high-frequency textures and local singular defects of the real stone paint itself, thereby achieving a quantitative representation of the macroscopic fundamental flatness deviation of the wall.

[0066] Key dimensions are determined according to building codes, such as the inspection scale corresponding to a 2-meter straightedge, and the detail component coefficient matrix corresponding to this scale is selected from the aforementioned wavelet decomposition results. , , For all wavelet coefficients in the above three detail component matrices Calculate the L1 norm, i.e. The L1 norm is more sensitive to coefficients caused by sparse outliers with large amplitudes, i.e., macroscopic fluctuations. Meanwhile, for wavelet coefficients representing higher-level or finer-scale textures of real stone paint, the energy distribution is more diffuse and the coefficient amplitudes are smaller, resulting in relatively lower L1 norm values. By comparing the L1 norm values ​​of preset key-scale frequency bands with those of higher-frequency scale bands, or by directly using the L1 norm values ​​of key-scale frequency bands as an indicator of smoothness deviation, a threshold based on engineering acceptance standards is set. If the calculated L1 norm value exceeds this threshold, the macroscopic smoothness of the wall at that scale is deemed unqualified. This L1 norm value serves as a quantitative representation of the wall smoothness deviation.

[0067] In one embodiment, the step of separating the sparse energy coefficients representing macroscopic large-format undulations from the energy coefficients representing the dense high-frequency textures and local singular defects of the stone paint by calculating the L1 norm of the wavelet detail component coefficients within a preset key scale frequency band, thereby achieving a quantitative representation of the fundamental macroscopic flatness deviation of the wall, includes:

[0068] The 8th layer decomposition, corresponding to a spatial wavelength of 0.128 meters, was selected as the key scale frequency band representing the macroscopic flatness of the wall.

[0069] The sum of the L1 norms of the detail component coefficients in the horizontal, vertical, and diagonal directions within the frequency band is calculated as an indicator of the macroscopic flatness deviation of the wall.

[0070] Meanwhile, the first to fourth layers of decomposition were selected as the frequency bands representing the high-frequency texture of the real stone paint;

[0071] The sum of the L1 norms of all detail component coefficients within the texture frequency band is calculated as an indicator of texture complexity, thereby separating the indicator representing smoothness from the indicator representing texture.

[0072] Since the 8th layer wavelet decomposition (j=8) detects features with a spatial size of approximately 256 pixels corresponding to a physical scale of 0.128 meters, it precisely corresponds to the large-scale, slowly changing undulations or unevenness of the wall, i.e., the macroscopic flatness problem. Therefore, this layer is defined as the flatness frequency band. The index of macroscopic flatness deviation of the wall... The result is obtained by calculating the sum of the L1 norms of all detail component coefficients within this frequency band: The L1 norm is sensitive to macroscopic fluctuations corresponding to sparse large-amplitude coefficients; the larger the value, the worse the flatness of the wall at the 0.128-meter scale.

[0073] The sand-like texture and uneven spraying of real stone paint are high-frequency features, typically ranging from a few pixels to tens of pixels in spatial scale. Therefore, layers 1 to 4 of the wavelet decomposition (j=1, 2, 3, 4) are selected as the texture frequency bands, corresponding to feature scales of 2-4 pixels, 4-8 pixels, 8-16 pixels, and 16-32 pixels, respectively. This is an index of the complexity of the wall surface texture. The result is obtained by calculating the sum of the L1 norms of all detail component coefficients within this frequency band: This indicator represents the richness of high-frequency details on the wall surface. It is calculated separately. and Two independent indicators successfully separated the macroscopic structural deviations that affect the flatness assessment from the high-frequency textures that are inherent characteristics of the wall finish layer, thus achieving a quantitative assessment of the fundamental flatness of the wall.

[0074] Secondly, this disclosure also provides an image detection system for the flatness of self-insulating concrete wall with stone-like paint, including the following modules:

[0075] The acquisition module is used to project a set of composite structured light patterns onto the surface of the self-insulating concrete wall with real stone paint. The composite structured light pattern uses frequency-modulated sinusoidal stripes with a logarithmic spatial frequency as a high-resolution phase information carrier, and Gray code encoded with two-dimensional absolute position information is superimposed on the chromaticity channel of the phase information carrier. The imaging device synchronously acquires the deformed pattern image modulated by the surface of the wall to be tested.

[0076] The construction module is used to reconstruct three-dimensional point cloud data of the wall surface based on the deformed pattern image;

[0077] The calculation module is used to convert the three-dimensional point cloud data into a regular gridded depth-height map, perform a two-dimensional discrete wavelet transform on the height map that is associated with the physical scale, and decompose the height map into multiple scales corresponding to the key spatial wavelengths in the wall flatness detection standard; by calculating the L1 norm of the wavelet detail component coefficients in the preset key scale frequency band, the sparse energy coefficients representing macroscopic large-format undulations are separated from the energy coefficients representing the dense high-frequency texture and local singular defects of the real stone paint itself, so as to achieve a quantitative representation of the macroscopic fundamental flatness deviation of the wall.

[0078] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0079] The aspects of this application have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0080] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for detecting the flatness of a self-insulating concrete wall with stone-like paint, characterized in that, Includes the following steps: A set of composite structured light patterns is projected onto the surface of the self-insulating concrete wall with real stone paint. The composite structured light pattern uses frequency-modulated sinusoidal stripes with a logarithmic spatial frequency as a high-resolution phase information carrier, and Gray code encoding two-dimensional absolute position information is superimposed on the chromaticity channel of the phase information carrier. The image of the deformed pattern modulated by the surface of the wall under test is simultaneously acquired by the imaging device. The three-dimensional point cloud data of the wall surface is reconstructed based on the deformed pattern image; The three-dimensional point cloud data is converted into a regular gridded depth height map. A two-dimensional discrete wavelet transform associated with physical scale is performed on the height map to decompose the height map into multiple scales corresponding to the key spatial wavelengths in the wall flatness detection standard. By calculating the L1 norm of the wavelet detail component coefficients in the preset key scale frequency band, the sparse energy coefficients representing macroscopic large-format undulations are separated from the energy coefficients representing the dense high-frequency texture and local singular defects of the real stone paint itself, so as to achieve a quantitative representation of the fundamental macroscopic flatness deviation of the wall. The process of reconstructing the three-dimensional point cloud data of the wall surface based on the deformed pattern image includes: Based on the deformed pattern image, a frequency response confidence map representing the local measurement signal-to-noise ratio is generated for each pixel by analyzing the local spectrum of the image. Then, a global energy function is constructed using the confidence map as data item weights and the phase continuity between adjacent pixels as a smoothing constraint. The function is then solved to obtain a high-precision global absolute phase map. The absolute phase map is input into a pre-calibrated system model to calculate and generate an initial 3D point cloud. A weighted nonlinear optimization of the coordinates of the initial 3D point cloud is then performed using an objective function with the confidence map as weights, thereby reconstructing high-fidelity 3D point cloud data of the wall surface. The method involves calculating the L1 norm of wavelet detail component coefficients within a preset key scale frequency band to separate the sparse energy coefficients representing macroscopic large-format undulations from the energy coefficients representing the dense high-frequency textures and local singular defects of the stone paint itself. This achieves a quantitative representation of the fundamental macroscopic flatness deviation of the wall, including: The 8th layer decomposition, corresponding to a spatial wavelength of 0.128 meters, was selected as the key scale frequency band representing the macroscopic flatness of the wall. The sum of the L1 norms of the detail component coefficients in the horizontal, vertical, and diagonal directions within the frequency band is calculated as an indicator of the macroscopic flatness deviation of the wall. Meanwhile, the first to fourth layers of decomposition were selected as the frequency bands representing the high-frequency texture of the real stone paint; The sum of the L1 norms of all detail component coefficients within the texture frequency band is calculated as an indicator of texture complexity, thereby separating the indicator representing smoothness from the indicator representing texture.

2. The method according to claim 1, characterized in that, The process involves projecting a composite structured light pattern onto the surface of the self-insulating concrete wall with stone-like paint. The composite structured light pattern uses frequency-modulated sinusoidal fringes with a logarithmic spatial frequency variation as a high-resolution phase information carrier, and superimposes Gray code encoding two-dimensional absolute position information onto the chromaticity channel of the phase information carrier. The pattern includes: Generate a set of frequency-modulated sinusoidal fringes with spatial frequencies that satisfy a logarithmic function along the x-axis, starting at a frequency of 0.05 periods per pixel and a logarithmic growth factor of 0.001; Generate a set of 8-bit horizontal Gray codes and a set of 8-bit vertical Gray codes, and generate corresponding bit-plane patterns. The frequency-modulated sine stripe is placed in the L* channel of the CIEL*a*b* color space. The encoded values ​​of each corresponding bit plane of the horizontal Gray code are mapped to the a* channel, and the encoded values ​​of each corresponding bit plane of the vertical Gray code are mapped to the b* channel. When the Gray code value in the bit plane is 0, the corresponding a* or b* channel value is the first mapping value, and when the Gray code value in the bit plane is 1, the corresponding a* or b* channel value is the second mapping value. The images of the three channels L*, a*, and b* are synthesized to obtain a set of composite structured light patterns, which are then projected sequentially by a projection device.

3. The method according to claim 1, characterized in that, The step of generating a frequency response confidence map representing the local measurement signal-to-noise ratio for each pixel based on the deformed pattern image by analyzing the local spectrum of the image includes: Select any pixel in the deformed pattern image, and apply a Hanning window of size 32×32 pixels centered on the pixel; Perform a two-dimensional fast Fourier transform on the data within the window to obtain the local spectrum of the pixel; Locate the energy peak of the fundamental frequency signal along the frequency variation axis in the local spectrum diagram; The average energy of all spectral components except the DC component in the local spectrum is calculated as the average noise level. The signal-to-noise ratio (SNR) is obtained by calculating the ratio of the peak energy of the fundamental frequency signal to the average noise level. This SNR is then used as the confidence level of the pixel. A complete frequency response confidence map is generated by traversing all pixels.

4. The method according to claim 1, characterized in that, The construction of a global energy function that uses the confidence map as the data item weights and the phase continuity between adjacent pixels as a smoothing constraint, and the solution of the function to obtain a global high-precision absolute phase map, includes: Define an energy function for each pixel, consisting of a smoothing term weighted by a data term and a regularization coefficient; The data item is determined by the confidence value of the pixel, the wrapping phase, the fringe index to be determined, and the coarse phase obtained by Gray code decoding; The smoothing term is determined by the spatial distance constraint between adjacent pixels, the confidence value of adjacent pixels, the difference in fringe index, and the maximum cutoff value of the fringe index change. The α-expansion image cut optimization algorithm is applied to iteratively minimize the overall energy function of the entire image until the energy function converges and is less than a preset threshold. The optimal fringe index of each pixel is output and the absolute phase is calculated.

5. The method according to claim 1, characterized in that, The construction of an objective function with weights based on the confidence map is used to perform weighted nonlinear optimization on the coordinates of the initial 3D point cloud, thereby reconstructing high-fidelity 3D point cloud data of the wall surface, including: For each point in the 3D point cloud, a weighted nonlinear least squares objective function is constructed. The objective function consists of a data term weighted by the confidence of the corresponding pixel and a smoothing term weighted by the regularization coefficient. The data item is used to penalize the deviation between the predicted phase calculated based on the three-dimensional points in a pre-calibrated system model and the corresponding measured phase. The smoothing term is used to penalize the spatial discontinuity between a point and its neighboring 3D points; The objective function is iteratively solved to optimize the coordinates of the three-dimensional points. The optimization stops when the change in the objective function is less than a preset threshold or the number of iterations reaches the upper limit. The output is a high-fidelity three-dimensional point cloud that suppresses the error in the low signal-to-noise ratio region.

6. The method according to claim 1, characterized in that, The step of converting the 3D point cloud data into a regular meshed depth-height map, and performing a physical scale-dependent 2D discrete wavelet transform on the height map, includes: Based on the detection standards and the physical resolution calibrated by the system, the characteristic spatial wavelength used to represent local flatness fluctuations is determined to be 0.128 meters, which corresponds to 256 pixels on the depth-height map; Symlet8 was selected as the wavelet basis function. The wavelet decomposition layer is set to 8 to ensure that the lowest frequency scale of the decomposition matches the wavelength of the feature space. An 8-layer two-dimensional discrete wavelet transform is performed on the regularly gridded depth-height map to obtain one low-frequency approximation component and eight sets of high-frequency detail components.

7. A system for detecting the flatness of self-insulating concrete walls coated with stone-like paint, characterized in that, Includes the following modules: The acquisition module is used to project a set of composite structured light patterns onto the surface of the self-insulating concrete wall with real stone paint. The composite structured light pattern uses frequency-modulated sinusoidal stripes with a logarithmic spatial frequency as a high-resolution phase information carrier, and Gray code encoded with two-dimensional absolute position information is superimposed on the chromaticity channel of the phase information carrier. The imaging device synchronously acquires the deformed pattern image modulated by the surface of the wall to be tested. The construction module is used to reconstruct three-dimensional point cloud data of the wall surface based on the deformed pattern image; The calculation module is used to convert the three-dimensional point cloud data into a regular gridded depth height map, perform a two-dimensional discrete wavelet transform on the height map that is associated with the physical scale, decompose the height map into multiple scales corresponding to the key spatial wavelengths in the wall flatness detection standard; by calculating the L1 norm of the wavelet detail component coefficients in the preset key scale frequency band, the sparse energy coefficients representing macroscopic large-format undulations are separated from the energy coefficients representing the dense high-frequency texture and local singular defects of the real stone paint itself, so as to realize the quantitative representation of the fundamental macroscopic flatness deviation of the wall; The process of reconstructing the three-dimensional point cloud data of the wall surface based on the deformed pattern image includes: Based on the deformed pattern image, a frequency response confidence map representing the local measurement signal-to-noise ratio is generated for each pixel by analyzing the local spectrum of the image. Then, a global energy function is constructed using the confidence map as data item weights and the phase continuity between adjacent pixels as a smoothing constraint. The function is then solved to obtain a high-precision global absolute phase map. The absolute phase map is input into a pre-calibrated system model to calculate and generate an initial 3D point cloud. A weighted nonlinear optimization of the coordinates of the initial 3D point cloud is then performed using an objective function with the confidence map as weights, thereby reconstructing high-fidelity 3D point cloud data of the wall surface. The method involves calculating the L1 norm of wavelet detail component coefficients within a preset key scale frequency band to separate the sparse energy coefficients representing macroscopic large-format undulations from the energy coefficients representing the dense high-frequency textures and local singular defects of the stone paint itself. This achieves a quantitative representation of the fundamental macroscopic flatness deviation of the wall, including: The 8th layer decomposition, corresponding to a spatial wavelength of 0.128 meters, was selected as the key scale frequency band representing the macroscopic flatness of the wall. The sum of the L1 norms of the detail component coefficients in the horizontal, vertical, and diagonal directions within the frequency band is calculated as an indicator of the macroscopic flatness deviation of the wall. Meanwhile, the first to fourth layers of decomposition were selected as the frequency bands representing the high-frequency texture of the real stone paint; The sum of the L1 norms of all detail component coefficients within the texture frequency band is calculated as an indicator of texture complexity, thereby separating the indicator representing smoothness from the indicator representing texture.

8. The system according to claim 7, characterized in that, The process involves projecting a composite structured light pattern onto the surface of the self-insulating concrete wall with stone-like paint. The composite structured light pattern uses frequency-modulated sinusoidal fringes with a logarithmic spatial frequency variation as a high-resolution phase information carrier, and superimposes Gray code encoding two-dimensional absolute position information onto the chromaticity channel of the phase information carrier. The pattern includes: Generate a set of frequency-modulated sinusoidal fringes with spatial frequencies that satisfy a logarithmic function along the x-axis, starting at a frequency of 0.05 periods per pixel and a logarithmic growth factor of 0.001; Generate a set of 8-bit horizontal Gray codes and a set of 8-bit vertical Gray codes, and generate corresponding bit-plane patterns. The frequency-modulated sine stripe is placed in the L* channel of the CIEL*a*b* color space. The encoded values ​​of each corresponding bit plane of the horizontal Gray code are mapped to the a* channel, and the encoded values ​​of each corresponding bit plane of the vertical Gray code are mapped to the b* channel. When the Gray code value in the bit plane is 0, the corresponding a* or b* channel value is the first mapping value, and when the Gray code value in the bit plane is 1, the corresponding a* or b* channel value is the second mapping value. The images of the three channels L*, a*, and b* are synthesized to obtain a set of composite structured light patterns, which are then projected sequentially by a projection device.

Images