Guided filter and SVD orthogonal least squares wall putty layer flatness identification method

Abstract

The application discloses a guided filter and SVD orthogonal least square wall putty layer flatness identification method, and belongs to the technical field of flatness, and comprises the following steps: S1, wall putty layer three-dimensional point cloud data acquisition: complete point cloud data of the wall putty layer to be detected is acquired through a three-dimensional scanning device, and original point cloud data set is obtained; S2, point cloud structure preserving denoising processing: the original point cloud data obtained is subjected to structure preserving denoising processing based on guided filter, and a denoised point cloud data set is obtained; S3, SVD orthogonal least square plane fitting: based on the singular value decomposition orthogonal least square method, the denoised point cloud data is subjected to global plane fitting, and an average plane is obtained; S4, global flatness calculation: the average plane is translated to the bottom of the point cloud, a zero reference surface is constructed, and the vertical distance of each point in the denoised point cloud data to the zero reference surface is calculated, the application can effectively denoise while retaining geometric characteristics, the fitting result is stable and reliable, and the quantitative reference is unified.

Description

Technical Field

[0001] This invention relates to the field of flatness calculation technology, and in particular to a method for identifying the flatness of wall putty layers using guided filtering and SVD orthogonal least squares. Background Technology

[0002] The smoothness of the wall putty layer is one of the core indicators for assessing the quality of building construction projects. Traditional testing methods mainly rely on the manual "right-edge method," where workers use a 2-meter right-edge and feeler gauge for sampling inspection. This method has significant drawbacks: it is labor-intensive, inefficient, and the measurement results heavily depend on the worker's skill and subjective judgment. Furthermore, the sampling points are sparse and scattered, failing to provide comprehensive information on the overall density and smoothness of the wall putty layer, resulting in poor repeatability of the measurement results.

[0003] With the development of 3D scanning technology, non-contact detection methods based on point cloud data have gradually become mainstream. This method acquires complete point cloud data of the wall putty layer through high-speed scanning, achieving a leap from "point" to "surface," and can "accurately capture its overall posture and trend" and perform "global flatness calculation." However, existing technologies still have the following shortcomings in the data processing stage:

[0004] 1. For point cloud denoising, traditional smoothing algorithms (such as Gaussian filtering and mean filtering) cannot effectively distinguish between noise and real geometric structures when suppressing noise. They easily blur key features such as wall edges and corners, leading to a decrease in the accuracy of subsequent geometric analysis.

[0005] 2. Regarding plane fitting and flatness quantification, existing technologies mostly use the RANSAC algorithm for plane fitting. This algorithm is a non-deterministic algorithm, and the results depend on random sampling, which cannot guarantee a unique solution. Moreover, its design goal is to find "interior points" and eliminate "exterior points". However, in the flatness evaluation of wall putty layer, the most concave or most convex "exterior points" are precisely the key evaluation data. Eliminating them will result in the inability to reflect the true macroscopic shape of the wall putty layer.

[0006] Therefore, there is an urgent need for a method to calculate the flatness of wall putty layers that can effectively reduce noise while preserving geometric features, providing stable and reliable fitting results, and establishing a unified quantitative benchmark, in order to overcome the shortcomings of existing technologies. Summary of the Invention

[0007] In order to overcome the shortcomings of the prior art, the technical problem to be solved by the present invention is to propose a flatness calculation method that effectively removes noise while preserving geometric features, provides stable fitting results, and unifies the quantization benchmark.

[0008] To achieve this objective, the present invention adopts the following technical solution:

[0009] This invention provides a method for identifying the smoothness of wall putty layers using guided filtering and orthogonal least squares SVD, comprising the following steps:

[0010] S1. 3D point cloud data acquisition of wall putty layer: Complete point cloud data of the wall putty layer to be inspected is acquired using a 3D scanning device to obtain the raw point cloud dataset. ,in Let be the three-dimensional coordinates of the i-th point, and n be the total number of points in the cloud;

[0011] S2. Point cloud data denoising: Perform structure-preserving denoising processing based on guided filtering on the acquired raw point cloud data to obtain the denoised point cloud dataset. Preserve the edges and local features of the wall putty layer;

[0012] S3, SVD Orthogonal Least Squares Plane Fitting: Based on the orthogonal least squares method of singular value decomposition, global plane fitting is performed on the denoised point cloud data to obtain the average plane representing the overall attitude of the wall.

[0013] S4. Global flatness calculation: The average plane is shifted to the bottom of the point cloud to construct a zero-point reference plane. The vertical distance from each point in the denoised point cloud data to the zero-point reference plane is calculated. The vertical distance is the global flatness quantification value of the corresponding point, realizing the digital assessment of the global flatness of the wall putty layer.

[0014] A preferred embodiment of the present invention is that, in step S2, the noise reduction process includes the following steps:

[0015] S21. Constructing a neighborhood point set: Using a fixed radius search strategy, for any point to be processed in the point cloud data... A neighborhood set is formed by combining all points within a predetermined radius r in the three-dimensional space. The shape and density of this neighborhood directly reflect the points'... Surrounding surface features; for any point in the point cloud data P Its neighborhood point set Defined as: (1)

[0016] in: : Represents the three-dimensional coordinate vector of any i points in a point cloud dataset. ; : Represents any other point in the point cloud that may belong to the neighborhood;

[0017] r: represents a predefined neighborhood search radius; : indicates a point Let r be the set of all points in a spherical space with center r and radius r. : Represents the Euclidean distance between two points;

[0018] S22, Geometric Analysis of Neighborhood Point Sets: Calculating Neighborhood Point Sets center of mass Covariance Matrix The centroid represents the geometric center of the neighborhood, while the covariance matrix... This describes the dispersion and correlation of the point set along various directions, and its eigenvectors and eigenvalues ​​contain information about the main direction and shape of the neighborhood; the centroid is: (2)

[0019] The covariance matrix is: (3)

[0020] in: k: the number of point clouds in the point cloud domain;

[0021] S23. Construct a local linear transformation model for the neighborhood of a point cloud: based on the neighborhood centroid. Covariance Matrix Construct a local linear transformation model for the neighborhood. Its local linear transformation model is defined as: (4)

[0022] Transformation matrix and bias vector The expression is: (5) (6)

[0023] in: A 3×3 transformation matrix responsible for scaling, rotating, or shearing points; b: A 3×1 translation vector to prevent the point cloud from shifting as a whole during the filtering process; Smoothing parameter: This parameter ensures the stability of the calculation; The size of the filter also controls the smoothness of the filter. The larger the value, the stronger the smoothing effect, but the more severe the blurring of details; I: 3×3 identity matrix;

[0024] S24. Update point coordinates: Update the coordinates of each point in S23. The neighboring area Calculated linear model Application to point Based on the original coordinates, the filtered new coordinates are calculated. After traversing all point cloud data and completing the noise reduction process, a brand new point cloud is obtained.

[0025] A preferred embodiment of the present invention is that the regularization parameter is: Positive numbers are used to ensure the matrix It is reversible, especially when neighboring points are collinear or coplanar (leading to...). In the case of degeneracy (singular or near-singular), ensure computational stability. The size of this value controls the smoothness of the filter. The larger the value, the stronger the smoothing effect, but the more severe the blurring of details.

[0026] A preferred embodiment of the present invention is that the radius r in the fixed radius search strategy is preset according to the density characteristics of the point cloud data, ensuring that the neighborhood point set can effectively reflect the local geometric features of the point to be processed.

[0027] A preferred embodiment of the present invention is that, in step S3, the average plane fitting based on the orthogonal least squares method of singular value decomposition includes the following steps:

[0028] In step S3, the mean plane fitting based on the orthogonal least squares method of singular value decomposition includes the following steps:

[0029] S31. Calculate the global point cloud centroid: According to the principle of orthogonal least squares, the best-fit plane will pass through the centroid of the point cloud. The centroid can be obtained by calculating the arithmetic mean of the three-dimensional coordinates of all points in the denoised point cloud data. It can be represented as: (7)

[0030] in:

[0031] N: Represents the total number of data points in the point cloud;

[0032] S32. Centroid Data Decentralization: The entire denoised point cloud dataset is translated so that its centroid coincides with the origin of the coordinate system (0, 0, 0), achieving decentralization. The decentralization formula is: (8)

[0033] in: : Represents the coordinate vector of the i-th point after centering. Together they form a new N×3 data matrix X;

[0034] S33. Singular Value Decomposition for Normal Vector: Considering the uneven distribution of discrete points in the point cloud on the surface of the putty layer, the singular value decomposition method is introduced. By constructing and decomposing the covariance matrix of the centered point cloud, the singular value decomposition is performed on the decentralized matrix X, decomposing X into the product of three specific matrices. Its standard form is as follows: (9)

[0035] in: X: N×3 centralized data matrix; U: An N×3 column orthogonal matrix; D: A 3×3 diagonal matrix whose diagonal elements are singular values ​​arranged in descending order. These quantify the distribution scale (i.e., the variance) of the data along the three principal axes. A 3×3 orthogonal matrix whose row vectors represent the three principal directions of the point cloud distribution;

[0036] diagonal elements , , Geometrically, it measures the extent of the extension of the data point cloud along its principal direction. Furthermore, due to the D matrix and... The matrices are closely related. row vectors , , The three orthogonal principal directions of the data distribution are defined, and the singular values ​​in D... , , These correspond to quantifying the importance or scale of the data in the three directions mentioned above, while the two largest singular values... , and its corresponding right singular matrix , The vectors describe the two directions with the largest data variance. These two vectors form the two orthogonal directions of the length and width of the fitting plane S, while the smallest singular value... This corresponds to the direction with the minimum data variance, and this direction with the minimum variance is perpendicular to the plane of the wall putty layer. Therefore, Corresponding right singular vector The unit normal vector determined as the best-fit plane ;

[0037] S34. Constructing the average plane equation: based on the normal vector n and the centroid Constructing the average plane equation using the plane point normal form equation, i.e., any point on the plane With the center of mass The vector formed Since it must be perpendicular to the normal vector n, their dot product is always zero, which leads to the following: (10)

[0038] in: : The coordinate vector of any point on the plane;

[0039] The general form of the standard plane equation is: The constant term d can be derived from the point normal form, therefore the equation of the average plane is: (11)

[0040] A preferred embodiment of the present invention is that, in step S4, the global flatness calculation includes the following steps:

[0041] S41. Calculate the directed perpendicular distance from all points to the mean plane: For any point in the denoised point cloud data... Calculate its directed vertical distance to the mean plane. : (12)

[0042] in: : Represents the signed perpendicular distance from any point in the i-th point cloud to the average plane;

[0043] S42. Translate to construct the zero-point reference plane: Extract The minimum value (usually negative) in the mean plane corresponds to the point of greatest concavity or furthest distance from the plane. The mean plane is translated along its normal vector by this minimum distance until it just touches this furthest point. The new translated plane is the zero-point reference plane, and the equation constants of the zero-point reference plane are obtained. : (13)

[0044] in: : Represents the signed perpendicular distance from any point in the point cloud to the mean plane. : Zero-point reference surface constant after translation; The minimum value among all i distance values ​​calculated by the first formula represents the maximum indentation depth of the point cloud relative to the "mean plane".

[0045] S43. Global Smoothness Calculation Based on Reference Surface: Global smoothness is the vertical distance from each denoised point cloud to the zero-point reference surface, representing the bulge height of each point relative to the lowest point of the entire surface. The formula for calculating global smoothness is: (14)

[0046] in: : The global flatness value of the i-th point, and ≥0.

[0047] The beneficial effects of this invention are as follows:

[0048] 1. Excellent noise reduction effect and complete feature preservation: Guided filtering constructs a local linear model through neighborhood geometric features, without the need for complex normal vector calculation and Gaussian weighting. While effectively suppressing high-frequency noise, it accurately preserves key geometric features such as wall edges and corners, overcoming the fuzziness defects of traditional filtering. Moreover, the algorithm complexity is much lower than that of bilateral filtering, and the computational efficiency is improved by an order of magnitude.

[0049] 2. Stable and reliable plane fitting results: The SVD orthogonal least squares method is adopted. It is a deterministic algorithm that always outputs a unique best fitting plane for the same point cloud data, avoiding the random sampling bias of the RANSAC algorithm. At the same time, this method includes all points (including key "outside points") in the calculation, minimizes the sum of squared global orthogonal distances, and the fitting results can truly reflect the macroscopic shape of the wall, solving the problem of RANSAC removing key data.

[0050] 3. Unified and absolute flatness measurement benchmark: By using the translation strategy from the average plane to the zero-point reference plane, an absolute flatness measurement benchmark is constructed. The flatness value of all points is non-negative and has a clear physical meaning (the height of the bulge relative to the lowest point). This achieves high-density and repeatable digital evaluation of the entire wall putty layer, making up for the global consistency defects of the traditional straightedge method and the existing point cloud method.

[0051] 4. High engineering practicality: Combining 3D scanning technology to achieve non-contact inspection, with high scanning efficiency and comprehensive data coverage, the detection results have an error of less than 1mm compared to the manual ruler method, with high consistency, and can provide efficient, accurate and comprehensive technical support for engineering quality assessment. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the flatness calculation process provided in a specific embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of the scanner testing wall provided in a specific embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram of point cloud data of wall putty layers before and after noise reduction provided in a specific embodiment of the present invention;

[0055] Figure 4 This is a schematic diagram of the point cloud flatness of the wall putty layer before and after noise reduction in a specific embodiment of the present invention.

[0056] Figure 5 This is a schematic diagram of the smoothness result of the wall putty layer after noise reduction provided in a specific embodiment of the present invention;

[0057] Figure 6 This is a schematic diagram of the area for extracting the flatness of the wall putty layer after noise reduction, provided in a specific embodiment of the present invention. Detailed Implementation

[0058] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0059] See attached document Figure 1-6 The guided filtering and SVD orthogonal least squares wall putty layer smoothness identification method includes the following steps:

[0060] S1. Experiment Preparation

[0061] S11. Test Object: Select a wall 2.4m long and 1.2m high. The wall surface has been treated with a putty layer, and the appearance of the putty layer is as follows: Figure 2 As shown;

[0062] S12. Scanning object: A 3D laser scanner is used. The equipment parameters are as follows: scanning rate 640,000 points / s, relative error ±2mm, scanning distance 1.7m (vertical distance from the center of the wall).

[0063] S13. Data Acquisition: The scanner is fixed at a distance of 1.7m from the wall for standing fixed acquisition. The scanning time is 90s to obtain complete original point cloud data of the wall putty layer. A total of [number] point cloud data were collected. The original point cloud dataset consists of [number] data points. ,in .

[0064] S2, Point Cloud Data Denoising Processing

[0065] S21. Constructing a neighborhood point set: The shape and density of the neighborhood point set need to directly reflect the surface features around the point to be processed. Therefore, the determination of r needs to be based on the point cloud density calculation to ensure that the neighborhood contains 30-50 points to ensure the reliability of geometric feature analysis.

[0066] Step 1: Calculate the point cloud surface density: scanned area of ​​the wall putty layer. Total number of scans If there are 1, then the point cloud surface density ;

[0067] The second step is to deduce the neighborhood radius: the neighborhood is spherical, so its projection onto the wall putty layer is circular. Let the number of target points within the neighborhood be N (30-50). Then the number of points within the neighborhood satisfies: According to the formula for the area of ​​a circle , can be obtained ;

[0068] Substitute the values ​​into the calculation: when N=30, ; When N=50 ;

[0069] Considering the spatial discreteness of the 3D point cloud and the slight divergence caused by the 1.7m scanning distance, the theoretical value needs to be amplified and corrected. Finally, a fixed search radius of r=0.02m was preset, and the actual verification showed that the number of points in the neighborhood was 35-45, which met the requirements of local geometric feature analysis.

[0070] S22, Geometric analysis of neighborhood point sets:

[0071] Centroid calculation: For each point to be processed neighborhood point set According to formula (2) Calculate the centroid, if a certain neighborhood Contains 40 points, that is Their coordinates are respectively ... Then the centroid of that neighborhood That is, by summing the x, y, and z coordinates of all points in the neighborhood and taking the average, the three-dimensional centroid coordinates are obtained.

[0072] Covariance matrix calculation: According to formula (3) To calculate the 3×3 covariance matrix, taking this neighborhood as an example, first calculate the covariance matrix for each point. With the center of mass deviation vector Then, perform matrix multiplication on the deviation vector and its transpose to obtain a 3×3 matrix. Summate this 3×3 matrix over 40 points and divide by the number of neighborhood points (40) to obtain the covariance matrix of the neighborhood. .

[0073] S23. Construct a local linear transformation model for the neighborhood of the point cloud:

[0074] Preset regularization parameters This value ensures the matrix reversible;

[0075] Transformation matrix Calculation: According to formula (5) First calculate The covariance matrix Add to the diagonal elements respectively Then find the inverse matrix of the matrix, and finally combine it with the covariance matrix. Perform matrix multiplication to obtain a 3×3 transformation matrix. ;

[0076] bias vector Calculation: According to formula (6) , the centroid vector Substitute the values ​​into the calculation and perform matrix multiplication. The 3×1 bias vector can be obtained by subtracting the result from the centroid vector.

[0077] S24. Update point coordinates: According to formula (4) Points to be processed Substitute the original coordinates (3×1) into the matrix and calculate using matrix multiplication. Then with the bias vector Perform vector addition to obtain the filtered points. Given the coordinates, traverse all points to complete the denoising process, and obtain the denoised point cloud dataset q.

[0078] Comparison of the before and after effects of guided filtering point cloud denoising technology on wall putty layer point clouds. Figure 3 As shown, Figure 3 (a) is the result before denoising. Figure 3 (b) shows the result after denoising. Compared with the point cloud before filtering, there are fewer noise points and the overall point cloud density is reduced.

[0079] Figure 4 ( ), Figure 4 ( The images show 3D point cloud graphics of the putty layer before and after noise reduction. It is clear that there are fewer noise points and the overall point cloud density is significantly reduced. The 3D model shows that the point cloud of the putty layer is less discrete and has better smoothness and continuity.

[0080] S3, SVD orthogonal least squares plane fitting

[0081] S31: Calculate the global point cloud centroid: according to formula (7) The total number of point clouds after denoising is N=356860. For all denoised point clouds... Summing the x, y, and z coordinates respectively, we get Then divide by 356860 to obtain the centroid. .

[0082] S32, Centroid data processing: Calculate according to formula (8) After obtaining the centered coordinates ,all The data matrix X is composed of 356860×3.

[0083] S33: Solving for the normal vector using singular value decomposition: According to formula (9) Perform singular value decomposition on the data matrix X to obtain an N×3 column orthogonal matrix U and a 3×3 diagonal matrix. and 3×3 orthogonal matrix .in Singular values ​​are sorted in descending order, with the smallest singular value being... Corresponding The last row of vectors represents the normal vector of the best-fit plane. .

[0084] S34. Construct the average plane equation: According to formula (10) Formula (11) is derived. , normal vector and center of mass Substitute and calculate (Engineering accuracy is preserved), therefore the average plane equation is 0.001x + 0.002y + 0.999z = 0.002; this average plane completely fits the overall posture of the wall and contains the macroscopic distribution information of all points in the wall putty layer.

[0085] S4, Global Smoothness Calculation

[0086] S41. Calculate the directed perpendicular distance from all points to the average plane: according to formula (12) For each denoised point First, calculate the dot product between the normal vector and the point. Subtracting d = 0.002m from this gives the directed vertical distance. .

[0087] In this experiment, all points The value range is [-0.0032m, 0.0098m]. A positive value indicates that the point is above the mean plane, and a negative value indicates that the point is below the mean plane.

[0088] S42. Translate to construct the zero-point reference plane: First extract all... The minimum value in min ( = -0.0032m, according to formula (13) Substituting d=0.002m and min( = -0.0032m, calculated as follows The equation of the zero-point reference plane after translation is 0.001x + 0.002y + 0.999z = -0.0012.

[0089] S43. Global flatness calculation based on the reference plane: According to formula (14) , each point (Step S41 has been calculated) Subtract =-0.0012m, obtaining the global flatness quantization value. For example, at a certain point =0.0005m, then =0.0005−(-0.0012)=0.0017m=1.7mm. All points in this experiment... The value range is [0m, 0.013m] (i.e., 0-13mm), realizing the digital quantification of the entire wall putty layer. The calculated result is as follows: Figure 5 As shown.

[0090] Result verification:

[0091] Select wall putty layer In the area, the straightedge is placed at a -1.04° angle. The flatness values ​​at three points (0.67m, 1.13m, and 1.39m) on the left end of the straightedge, from left to right, are extracted and verified against the flatness measured manually at these three points. The straightedge placement area and the area extracted by the algorithm are shown below. Figure 6 As shown, the accuracy of the calculation is verified according to Table 1:

[0092] Table 1 Comparison of Flatness Algorithm Detection Error and Manual Straightedge Detection Error

[0093] The results show that the proposed flatness calculation method has a high degree of agreement with the manual comparison, with errors all less than 1 mm, which meets the accuracy requirements of engineering quality testing.

[0094] In terms of noise reduction, the guided filtering method of this invention is implemented through a local linear model. Its algorithm complexity is much lower than that of the bilateral filtering method, which relies on Gaussian weighting and normal vector calculation. While achieving better geometric feature preservation, the computational efficiency is an order of magnitude faster.

[0095] Regarding flatness calculation, firstly, this invention employs the SVD orthogonal least squares method, a deterministic algorithm. For the same point cloud data, SVD decomposition always produces a unique and repeatable best-fit plane solution, overcoming the inherent defect of RANSAC, which may lead to inconsistent results due to random sampling. Secondly, this method mathematically seeks a globally optimal solution, taking all input point clouds into consideration and minimizing the sum of squared orthogonal distances from all points to the plane, thus obtaining an average plane that most accurately represents the overall macroscopic shape of the wall. This differs from RANSAC's logic of finding the largest subset of interior points and potentially ignoring "outlier" points that are crucial for flatness assessment, and also avoids the additional computational complexity and potential accuracy loss that may result from multiple fittings and averaging in that literature. Furthermore, this invention, through explicit "two-step" reference plane construction, first determines the best average plane and then translates it to the lowest point, ensuring the absoluteness of the zero-point reference plane and the non-negativity and physical meaning of all subsequent height calculations, providing a more rigorous and robust global flatness measurement basis.

[0096] In summary, this invention achieves non-contact, full-coverage wall putty layer flatness detection through data acquisition, guided filtering denoising, SVD plane fitting, and global flatness calculation. It overcomes the limitations of traditional ruler-based sampling inspection, provides high-density global data, and utilizes the neighborhood covariance matrix to construct a local linear model. This effectively removes point cloud noise while resolving the ambiguity of key geometric features such as wall edges and corners caused by traditional filtering. It integrates structure-preserving denoising and deterministic plane fitting techniques, solving three core problems in existing methods: the contradiction between denoising and feature preservation, unstable fitting results, and inconsistent quantization benchmarks. This forms a complete solution from data preprocessing to result output. Furthermore, it boasts high detection efficiency and strong repeatability; the scanning acquisition time is only 90-180 seconds depending on the wall putty layer size, and subsequent algorithm processing is automated, avoiding human subjective error and providing a standardized technical means for engineering quality assessment.

[0097] This invention has been described through preferred embodiments. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. This invention is not limited to the specific embodiments disclosed herein; other embodiments falling within the scope of the claims are also within the protection scope of this invention.

Claims

1. A method for identifying the smoothness of wall putty layer using guided filtering and orthogonal least squares SVD, characterized in that: Includes the following steps: S1. 3D point cloud data acquisition of wall putty layer: Complete point cloud data of the wall putty layer to be inspected is acquired using a 3D scanning device to obtain the raw point cloud dataset. ,in Let be the three-dimensional coordinates of the i-th point, and n be the total number of points in the cloud; S2. Point cloud data denoising: Perform structure-preserving denoising processing based on guided filtering on the acquired raw point cloud data to obtain the denoised point cloud dataset. Preserve the edges and local features of the wall putty layer; S3, SVD Orthogonal Least Squares Plane Fitting: Based on the orthogonal least squares method of singular value decomposition, global plane fitting is performed on the denoised point cloud data to obtain the average plane representing the wall putty layer. S4. Global flatness calculation: The average plane is shifted to the bottom of the point cloud to construct a zero-point reference plane. The vertical distance from each point in the denoised point cloud data to the zero-point reference plane is calculated. The vertical distance is the global flatness quantification value of the corresponding point, realizing the digital assessment of the global flatness of the wall putty layer.

2. The method for identifying the smoothness of wall putty layer using guided filtering and SVD orthogonal least squares as described in claim 1, characterized in that: In step S2, the noise reduction process includes the following steps: S21. Constructing a neighborhood point set: Using a fixed radius search strategy, for any point to be processed in the point cloud data... A neighborhood set is formed by combining all points within a predetermined radius r in the three-dimensional space. The shape and density of this neighborhood directly reflect the points'... Surrounding surface features; for any point in the point cloud data P Its neighborhood point set Defined as: （1） in: : Represents the three-dimensional coordinate vector of any i points in a point cloud dataset. ; : Represents any other point in the point cloud that may belong to the neighborhood; r: represents a predefined neighborhood search radius; : indicates a point Let r be the set of all points in a spherical space with center r and radius r. : Represents the Euclidean distance between two points; S22, Geometric Analysis of Neighborhood Point Sets: Calculating Neighborhood Point Sets center of mass Covariance Matrix The centroid represents the geometric center of the neighborhood, while the covariance matrix... This describes the dispersion and correlation of the point set along various directions, and its eigenvectors and eigenvalues ​​contain information about the main direction and shape of the neighborhood; the centroid is: （2） The covariance matrix is: （3） in: k: the number of point clouds in the point cloud domain; S23. Construct a local linear transformation model for the neighborhood of a point cloud: based on the neighborhood centroid. Covariance Matrix Construct a local linear transformation model for the neighborhood. Its local linear transformation model is defined as: （4） Transformation matrix and bias vector The expression is: （5） （6） in: A 3×3 transformation matrix responsible for scaling, rotating, or shearing points; b: A 3×1 translation vector to prevent the point cloud from shifting as a whole during the filtering process; Smoothing parameter: This parameter ensures the stability of the calculation; The size of the filter also controls the smoothness of the filter. The larger the value, the stronger the smoothing effect, but the more severe the blurring of details; I: 3×3 identity matrix; S24. Update point coordinates: Update the coordinates of each point in S23. The neighboring area Calculated linear model Application to point Based on the original coordinates, the filtered new coordinates are calculated. After traversing all point cloud data and completing the noise reduction process, a brand new point cloud is obtained.

3. The method for identifying the smoothness of wall putty layer using guided filtering and SVD orthogonal least squares as described in claim 2, characterized in that: The regularization parameter is: Positive numbers are used to ensure the matrix It is reversible, especially when neighboring points are collinear or coplanar (leading to...). In the case of degeneracy (singular or near-singular), ensure computational stability. The size of this value controls the smoothness of the filter. The larger the value, the stronger the smoothing effect, but the more severe the blurring of details.

4. The method for identifying the smoothness of wall putty layer using guided filtering and SVD orthogonal least squares as described in claim 2, characterized in that: The radius r in the fixed radius search strategy is preset based on the density characteristics of the point cloud data to ensure that the neighborhood point set can effectively reflect the local geometric features of the point to be processed.

5. The method for identifying the smoothness of wall putty layer using guided filtering and SVD orthogonal least squares as described in claim 1, characterized in that: In step S3, the mean plane fitting based on the orthogonal least squares method of singular value decomposition includes the following steps: S31. Calculate the global point cloud centroid: According to the principle of orthogonal least squares, the best-fit plane will pass through the centroid of the point cloud. The centroid can be obtained by calculating the arithmetic mean of the three-dimensional coordinates of all points in the denoised point cloud data. It can be represented as: （7） in: N: Represents the total number of data points in the point cloud; S32. Centroid Data Decentralization: The entire denoised point cloud dataset is translated so that its centroid coincides with the origin of the coordinate system (0, 0, 0), achieving decentralization. The decentralization formula is: （8） in: : Represents the coordinate vector of the i-th point after centering. Together they form a new N×3 data matrix X; S33. Singular Value Decomposition for Normal Vector: Considering the uneven distribution of discrete points in the point cloud on the surface of the putty layer, the singular value decomposition method is introduced. By constructing and decomposing the covariance matrix of the centered point cloud, the singular value decomposition is performed on the decentralized matrix X, decomposing X into the product of three specific matrices. Its standard form is as follows: （9） in: X: N×3 centralized data matrix; U: An N×3 column orthogonal matrix; D: A 3×3 diagonal matrix whose diagonal elements are singular values ​​arranged in descending order. These quantify the distribution scale (i.e., the variance) of the data along the three principal axes. A 3×3 orthogonal matrix whose row vectors represent the three principal directions of the point cloud distribution; diagonal elements , , Geometrically, it measures the extent of the extension of the data point cloud along its principal direction. Furthermore, due to the D matrix and... The matrices are closely related. row vectors , , Three orthogonal principal directions of the data distribution were defined, and Singular values ​​in , , These correspond to quantifying the importance or scale of the data in the three directions mentioned above, while the two largest singular values... , and its corresponding right singular matrix , The vectors describe the two directions with the largest data variance. These two vectors form the two orthogonal directions of the length and width of the fitting plane S, while the smallest singular value... This corresponds to the direction with the minimum data variance, and this direction with the minimum variance is perpendicular to the plane of the wall putty layer. Therefore, Corresponding right singular vector The unit normal vector determined as the best-fit plane ; S34. Constructing the average plane equation: based on the normal vector n and the centroid Constructing the average plane equation using the plane point normal form equation, i.e., any point on the plane With center of mass The vector formed Since it must be perpendicular to the normal vector n, their dot product is always zero, which leads to the following: （10） in: : The coordinate vector of any point on the plane; The general form of the standard plane equation is: The constant term d can be derived from the point normal form, therefore the equation of the average plane is: （11）。 6. The method for identifying the smoothness of wall putty layer using guided filtering and SVD orthogonal least squares as described in claim 1, characterized in that: In step S4, the global flatness calculation includes the following steps: S41. Calculate the directed perpendicular distance from all points to the mean plane: For any point in the denoised point cloud data... Calculate its directed vertical distance to the mean plane. : （12） in: : Represents the signed perpendicular distance from any point in the i-th point cloud to the average plane; S42. Translate to construct the zero-point reference plane: Extract The minimum value (usually negative) in the mean plane corresponds to the point of greatest concavity or furthest distance from the plane. The mean plane is translated along its normal vector by this minimum distance until it just touches this furthest point. The new translated plane is the zero-point reference plane, and the equation constants of the zero-point reference plane are obtained. : （13） in: : Represents the signed perpendicular distance from any point in the point cloud to the mean plane. : Zero-point reference surface constant after translation; The minimum value among all i distance values ​​calculated by the first formula represents the maximum indentation depth of the point cloud relative to the "mean plane". S43. Global Smoothness Calculation Based on Reference Surface: Global smoothness is the vertical distance from each denoised point cloud to the zero-point reference surface, representing the bulge height of each point relative to the lowest point of the entire surface. The formula for calculating global smoothness is: （14） in: : The global flatness value of the i-th point, and ≥0.

Images