Point cloud-based method for detecting flatness of prefabricated wall with holes

Abstract

The invention discloses a method for detecting the flatness of a prefabricated wall with holes based on point cloud, and the method comprises the steps: firstly collecting the point cloud data of the surface of the prefabricated wall, and constructing a point cloud data set of the surface of the prefabricated wall; carrying out downsampling processing on the point cloud data in the point cloud data set by adopting a curvature feature and adjacent point normal vector angle dual-threshold screening mechanism; aligning the point cloud data after downsampling processing with a BIM model of the prefabricated wall, marking a hole area, generating a hole area mask, and eliminating interference of invalid data of the hole area; and finally, fitting a global reference plane based on a curvature compensation RANSAC algorithm, calculating a deviation value between the central point of the local curved surface and the global reference plane through a dynamic window, and evaluating the surface flatness of the prefabricated wall body. According to the method, the detection precision and efficiency are improved, noise interference of invalid data in a hole area is avoided, construction defect information is reserved, and the robustness of prefabricated wall surface flatness detection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of construction engineering quality detection, and in particular to a point cloud-based flatness detection method for prefabricated walls with holes. Background Art

[0002] With the rapid development of prefabricated building technology, precast concrete walls are widely used due to their standardized design, efficient construction, and environmentally friendly design. However, the surface flatness of precast walls directly impacts building structural safety, construction progress, and the quality of subsequent renovations. Traditional testing methods face significant challenges, especially in areas with holes such as pipe holes and bolt connection holes. Traditional flatness testing relies primarily on contact distance meters or manual visual inspection, which suffers from low efficiency, insufficient accuracy, and difficulty adapting to complex hole structures.

[0003] While 3D laser scanning and BIM-integrated inspection technologies are being explored in current engineering practice, the following core issues remain: During the point cloud screening and downsampling stages, traditional methods rely on fixed parameters (such as voxel grid filtering or simple random sampling), making it difficult to adaptively process holes and concave-convex wall features. For example, fixed thresholds can lead to loss of critical geometric information or residual noise in areas with sudden changes in curvature (such as hole edges), while excessive downsampling can create data redundancy and reduce subsequent processing efficiency. While some studies have incorporated curvature features for screening, they lack a mechanism for jointly distinguishing the normal vector angles of neighboring points, making it impossible to effectively distinguish sharp features at hole boundaries from random noise points, ultimately affecting the precise alignment of the BIM model with the point cloud. During the flatness calculation stage, existing methods often use global plane fitting algorithms (such as RANSAC), which fail to consider the impact of local curvature changes on the inspection results. Furthermore, hole areas are susceptible to noise interference, resulting in inaccurate assessment results. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a point cloud-based flatness detection method for prefabricated walls with holes, which solves the problems of low accuracy, poor adaptability and insufficient efficiency of traditional wall flatness detection technology in hole areas.

[0005] The technical solution of the present invention is: A method for detecting the flatness of a prefabricated wall with holes based on a point cloud comprises the following steps: (1) Collect point cloud data of the prefabricated wall surface and construct a point cloud dataset of the prefabricated wall surface; (2) Using a dual-threshold screening mechanism of curvature features and the normal vector angle of the neighboring points, the point cloud data in the point cloud dataset is downsampled while retaining the concave and convex areas of the wall and the edge features of the holes; (3) Align the downsampled point cloud data with the BIM model of the prefabricated wall, mark the hole area, and generate a hole area mask to eliminate the interference of invalid data in the hole area; (4) Based on the curvature compensation RANSAC algorithm, the global reference plane is fitted, and the deviation value between the local surface center point and the global reference plane is calculated through the dynamic window to evaluate the surface flatness of the prefabricated wall.

[0006] The point cloud data of the prefabricated wall surface is collected by using a 3D line laser camera.

[0007] The specific steps of downsampling the point cloud data in the point cloud dataset using the dual threshold screening mechanism of curvature features and neighboring point normal vector angles while retaining the concave and convex areas of the wall and the edge features of the holes are as follows: S21. For each point in the point cloud dataset , use KD tree to search for its nearest multiple neighbors, build a neighborhood point set, and calculate the covariance matrix of the neighborhood point set, the covariance matrix The calculation formula is shown in the following formula (1): (1); In formula (1), the neighborhood centroid coordinates , Neighboring points in the neighborhood point set the number of Represents the transpose operation of the vector, which transforms the row vector Convert to column vector , so that the outer product operation result is a 3×3 covariance matrix; Then the covariance matrix Perform eigenvalue decomposition to obtain eigenvalues , the maximum eigenvalue The direction in which the data expands fastest is the direction with the smallest curvature and the smallest eigenvalue. The direction in which the corresponding data is most concentrated is the direction with the greatest curvature; Among them, the minimum eigenvalue The corresponding eigenvector is the point Normal vector ,point The curvature value Calculated by the following formula (2): (2); In formula (2), the curvature value It is a local curvature metric based on a set of neighborhood points. It quantifies the curvature of the local surface. High-curvature wall areas will be significantly different from flat wall areas. S22, all points The curvature values ​​of the point cloud dataset are summed and then divided by the total number of points in the point cloud dataset to obtain the mean curvature of the point cloud dataset. , see the following formula (3) for details: (3); In formula (3), Represents the total number of points in the point cloud dataset, is the curvature of the i-th point; Then calculate the curvature threshold , the calculation process is shown in the following formula (4): (4); In formula (4), represents the standard deviation of curvature; S23, in order to capture the tiny concave and convex defects on the surface of the prefabricated wall, calculate the points All its neighbors The normal vector angle , and take the normal vector angle The maximum normal angle in , normal vector angle The calculation of is shown in the following formula (5): (5); In formula (5), Representative Points The normal vector of Representative neighbor The normal vector of S24, retain the point cloud in the point cloud dataset that satisfies the following formula (6) , thereby retaining points with high curvature or large normal vector angles, that is, preferentially capturing hole edges and surface uneven areas, and realizing downsampling processing of point cloud data; (6); In formula (6), Representative Points The curvature value of Representative Points Its neighbors The maximum normal vector angle, represents the curvature threshold, represents the angle threshold, is the set value.

[0008] The specific steps of aligning the downsampled point cloud data with the BIM model of the prefabricated wall, marking the hole area, and generating the hole area mask are as follows: S31. Analyze the BIM model of the prefabricated wall to extract the boundary vertex set of the holes on the prefabricated wall, and determine the boundary point set of the holes based on the high curvature point cloud data retained after downsampling. Then, construct a bidirectional deviation function and match the boundary vertices in the BIM model with the hole boundary points in the point cloud dataset using a bidirectional distance deviation iterative optimization algorithm. S32. First, the positioning deviation of the point cloud is compensated according to the distance deviation calculated by the constructed bidirectional deviation function. Then, the dynamic expansion radius is calculated according to the density of the point cloud, and the normal vector direction of the point cloud is obtained. For each hole boundary point after positioning deviation compensation, the corresponding dynamic expansion radius value is moved along the positive direction of its normal vector to generate an expansion point, determine the expansion area, and obtain the mask of the hole area. Finally, the point cloud in the mask is removed to obtain the point cloud data of the prefabricated wall surface with the hole area removed.

[0009] The formula of the bidirectional deviation function is shown in the following formula (7): (7); In formula (7), Represents the distance deviation, represents the Euclidean distance, Represents the coordinates of the boundary vertices in the boundary vertex set, Represents the number of boundary vertices in the boundary vertex set; is the coordinate of the hole boundary point in the boundary point set, Represents the number of hole boundary points in the boundary point set; Represents the boundary vertex The nearest match of Represents the hole boundary point The nearest match of ; The specific steps of matching through the bidirectional distance deviation iterative optimization algorithm are: (a) The boundary vertices Project to the boundary point set and calculate the distance to the nearest point; (b) The hole boundary points Project to the boundary vertex set and calculate the nearest point distance; (c) Minimize the bidirectional deviation through singular value decomposition (SVD) and update the translation and rotation matrix of the BIM model; (d) When the distance deviation is less than The optimization is terminated when the set threshold or the maximum number of iterations is reached.

[0010] The dynamic expansion radius is calculated according to the density of the point cloud, and the calculation formula is shown in the following formula (8): (8); In formula (8), represents the dynamic expansion radius, represents the average spacing of the local point cloud, is the coordinate of the hole boundary point in the boundary point set, Represents the hole boundary point The coordinates of the nearest neighbor, Hole boundary point The nearest neighbor of The number of

[0011] The specific steps of fitting the global reference plane based on the curvature compensation RANSAC algorithm and calculating the deviation value between the local surface center point and the global reference plane through the dynamic window are as follows: S41. Fitting the global reference plane based on the curvature-compensated RANSAC algorithm: S411, first randomly select three non-collinear points from the point cloud data of the prefabricated wall surface with the hole area removed 、 and , the three non-collinear points are connected to form a candidate plane, and the plane equation is as follows (9); (9); In formula (9), The coordinates are , The coordinates are , The coordinates are ; Then use the difference multiplication method to calculate the parameters in formula (9) 、 、 Calculate, see the following formula (10): (10); In formula (10), Represented by three non-collinear points The normal vector of the determined plane, that is, the vector perpendicular to the plane; 、 、 are the unit basis vectors in the 3D Cartesian coordinate system: The unit vector corresponding to the x-axis direction, The unit vector corresponding to the y-axis direction, The unit vector corresponding to the z-axis direction; After expanding formula (10), we can get: 、 、 、 ; 、 、 is the normal vector Components in the three coordinate axes: is the component of the normal vector in the x-axis direction, is the component of the normal vector in the y-axis direction, is the component of the normal vector in the z-axis direction; S412, for each point in the point cloud data of the prefabricated wall surface after removing the hole area , calculate its weighted distance to the plane , see the following formula (11) for details: (11); In formula (11), the radius of curvature is Yes The curvature value The reciprocal of ; S413, set dynamic threshold: for each point , according to its curvature radius Calculating dynamic thresholds , see the following formula (12) for details: (12); In formula (12), BaseThreshold is the global baseline threshold; S414, determine whether the point meets the conditions: for each point ,when , then it is an inlier, and the number of inliers that meet the conditions is counted. When the number of inliers Ninlier of the current plane is greater than the number of inliers of all previous planes, the optimal plane is updated, and the updated optimal plane is regarded as the global reference plane of the current point cloud data. S42. For the point cloud data of the prefabricated wall surface with the hole area removed, adaptively adjust the size of the dynamic window based on the point cloud density, and calculate the deviation value from the center point of the local surface corresponding to each dynamic window to the global reference plane; S421. Calculate the dynamic window size, as shown in the following formula (13): (13); In formula (13), Represents the minimum size of the window; local elevation standard deviation , is the elevation value of the i-th point in the point cloud, i.e., the z coordinate, is the average elevation value of the point cloud in the window, , is the total number of point clouds in the window; point cloud density , represents the reference radius; Formula (13) expresses Calculate the dynamic window size from the calculated dynamic window size and the minimum window size Select the maximum value as the dynamic window size ; S422, point set in dynamic window , use the moving least squares method MLS to fit the quadratic surface, as shown in the following formula (14), and calculate the center point of the local surface , see the following formula (15) for details: (14); (15); In formula (14) and formula (15), is the elevation coordinate perpendicular to the wall, 、 is the position coordinate in the wall plane, and through formula (14), Expressed as and Function used to fit the local surface; 、 are the principal curvature coefficients of the surface, representing the opening direction and curvature along the x and y directions respectively; is the coupled curvature coefficient of the surface, representing The degree of saddle distortion of the plane; 、 is the linear slope coefficient of the surface, representing the inclination along the x and y directions respectively; is a constant term, representing the base elevation offset of the surface; represents the point set in the dynamic window, and the dynamic window size is adaptively adjusted by formula (13); 、 、 is the point set in the dynamic window interior point The x, y, and z axis coordinate values ​​in three-dimensional space; Represents the point set in the dynamic window the number of inliers; S43: Calculate the center point of the local surface Perpendicular distance to the global reference plane , see the following formula (16) for details: (16); In formula (16), 、 、 Represent the center points of the local surfaces The x, y, and z axis coordinate values, 、 、 、 The parameters of the plane equation of the global reference plane fitted in step S41; Then calculate the average deviation of the local surface center point to the global reference plane The average deviation reflects the overall flatness level of the wall surface as shown in the following formula (17); calculate the standard deviation from the center point of the local surface to the global reference plane , see the following formula (18); calculate the maximum deviation from the center point of the local surface to the global reference plane , see the following formula (19); (17); (18); (19); In formula (17)-formula (19), Represents the number of local surfaces; Finally, the surface flatness of the prefabricated wall is evaluated by the average deviation, standard deviation and maximum deviation: average deviation Reflects the overall flatness level of the prefabricated wall surface; standard deviation Measures the degree of dispersion of the deviation distribution; maximum deviation Locate critical defective areas requiring repair on precast wall surfaces.

[0012] Advantages of the present invention: (1) The present invention adopts a dual-threshold screening mechanism of curvature features and normal vector angles of neighboring points to downsample the point cloud data, which retains the concave and convex areas of the wall and the edge features of the holes, while eliminating random noise points, thus balancing the detection accuracy and detection efficiency; (2) The present invention realizes the precise alignment between the BIM model of the prefabricated wall and the point cloud based on the extraction of high-curvature point cloud data and the iterative optimization algorithm of bidirectional distance deviation, and removes the surface point cloud data of the prefabricated wall in the hole area through the dynamic expansion radius mask, thereby avoiding the noise interference of invalid data in the hole area and retaining the construction defect information; (3) The present invention combines the curvature compensation RANSAC algorithm to fit the global reference plane, and then dynamically adjusts the window size based on the curvature change. The deviation value between the local surface center point and the global reference plane is calculated through the dynamic window instead of a fixed window size, thereby improving the robustness of the prefabricated wall surface flatness detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flow chart of the present invention.

[0014] Figure 2 It is a flow chart of the point cloud data downsampling processing of the present invention.

[0015] Figure 3 This is a flow chart of the present invention for completing flatness detection based on a global reference plane and a local curved surface. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] See Figure 1 A method for detecting the flatness of a prefabricated wall with holes based on point cloud comprises the following steps: (1) A 3D line laser camera (accuracy 0.02mm) was used to collect point cloud data of the surface of a prefabricated wall (width 1.2m × height 2.95m). 1,257,342 points of data were obtained to construct a point cloud dataset of the prefabricated wall surface. The 3D line laser camera used binocular parallax stereo imaging technology and line laser assisted positioning. The three-dimensional spatial coordinates of the wall were obtained through a binocular stereo vision system, and a high-precision three-dimensional model was reconstructed to obtain a three-dimensional point cloud data model. (2) See Figure 2 , using a dual-threshold screening mechanism of curvature features and neighboring point normal vector angles, while retaining the concave and convex areas of the wall and the edge features of the holes, the point cloud data in the point cloud dataset was downsampled, retaining 386,522 points of data. The processing took 4.3 seconds. The specific steps are as follows: S21. For each point in the point cloud dataset , use KD tree to search its nearest thirty neighboring points, construct a neighborhood point set, and calculate the covariance matrix of the neighborhood point set. The covariance matrix The calculation formula is shown in the following formula (1): (1); In formula (1), the neighborhood centroid coordinates , Neighboring points in the neighborhood point set the number of (i.e. 30), Represents the transpose operation of the vector, which transforms the row vector Convert to column vector , so that the outer product operation result is a 3×3 covariance matrix; Then the covariance matrix Perform eigenvalue decomposition to obtain eigenvalues , the maximum eigenvalue The direction in which the data expands fastest is the direction with the smallest curvature and the smallest eigenvalue. The direction in which the corresponding data is most concentrated is the direction with the greatest curvature; Among them, the minimum eigenvalue The corresponding eigenvector is the point Normal vector ,point The curvature value Calculated by the following formula (2): (2); In formula (2), the curvature value It is a local curvature metric based on a set of neighborhood points. It quantifies the curvature of the local surface. High-curvature wall areas will be significantly different from flat wall areas. S22, all points The curvature values ​​of the point cloud dataset are summed and then divided by the total number of points in the point cloud dataset to obtain the mean curvature of the point cloud dataset. , see the following formula (3) for details: (3); In formula (3), Represents the total number of points in the point cloud dataset, is the curvature of the i-th point; Then calculate the curvature threshold , the calculation process is shown in the following formula (4): (4); In formula (4), Represents the standard deviation of curvature, covering the first 15% of feature points in the curvature distribution; S23, in order to capture the tiny concave and convex defects on the surface of the prefabricated wall, calculate the points All its neighbors The normal vector angle , and take the normal vector angle The maximum normal angle in , normal vector angle The calculation of is shown in the following formula (5): (5); In formula (5), Representative Points The normal vector of Representative neighbor The normal vector of S24, retain the point cloud in the point cloud dataset that satisfies the following formula (6) , thereby retaining points with high curvature or large normal vector angles, that is, preferentially capturing hole edges and surface uneven areas, and realizing downsampling processing of point cloud data; (6); In formula (6), Representative Points The curvature value of Representative Points Its neighbors The maximum normal vector angle, represents the curvature threshold, represents the angle threshold, ; (3) Align the downsampled point cloud data with the BIM model of the prefabricated wall, mark the hole area, and generate a hole area mask to eliminate the interference of invalid data in the hole area. The specific steps are as follows: S31. Analyze the BIM model of the prefabricated wall to extract the boundary vertex set of the holes on the prefabricated wall , ; The α-shape algorithm is used to extract the hole boundary points (high curvature point cloud data) from the point data retained after downsampling processing, and determine the boundary point set of the hole , ; Construct a bidirectional deviation function, as shown in the following formula (7), and match the boundary vertices in the BIM model with the hole boundary points in the point cloud dataset through a bidirectional distance deviation iterative optimization algorithm; (7); In formula (7), Represents the distance deviation, represents the Euclidean distance, Represents a set of boundary vertices The coordinates of the mid-boundary vertices, Represents a set of boundary vertices The number of boundary vertices; is a set of boundary points The coordinates of the hole boundary points, Represents a set of boundary points The number of boundary points of the hole; Represents the boundary vertex The nearest match of Represents the hole boundary point The nearest match of ; The specific steps of matching through the bidirectional distance deviation iterative optimization algorithm are: (a) The boundary vertices Project to the boundary point set and calculate the distance to the nearest point; (b) The hole boundary points Project to the boundary vertex set and calculate the nearest point distance; (c) Minimize the bidirectional deviation through singular value decomposition (SVD) and update the translation and rotation matrix of the BIM model; (d) When the distance deviation is less than The threshold value is set to 0.5 mm 2 Or the maximum number of iterations is reached and the optimization is terminated; S32, firstly, the positioning deviation of the point cloud is compensated according to the distance deviation calculated by the constructed bidirectional deviation function, and the distance deviation , and then calculate the dynamic expansion radius according to the density of the point cloud. The calculation formula is shown in the following formula (8): (8); In formula (8), represents the dynamic expansion radius, represents the average spacing of the local point cloud, is the coordinate of the hole boundary point in the boundary point set, Represents the hole boundary point The coordinates of the nearest neighbor, Hole boundary point The nearest neighbor of the number of Then, based on the direction of the normal vector of the acquired point cloud, the corresponding dynamic expansion radius value is moved along the positive direction of the normal vector of each hole boundary point after positioning deviation compensation to generate an expansion point and determine the expansion area. The masks of the twelve hole areas are obtained, and the hole radius is 28-52mm. Finally, the spatial clipping method is used to remove the point cloud within the mask to obtain the point cloud data of the prefabricated wall surface without the hole area. (4) See Figure 3 , based on the curvature compensation RANSAC algorithm to fit the global reference plane, the deviation value between the local surface center point and the global reference plane is calculated through the dynamic window to evaluate the surface flatness of the prefabricated wall. The specific steps are as follows: S41. Fitting the global reference plane based on the curvature-compensated RANSAC algorithm: S411, first randomly select three non-collinear points from the point cloud data of the prefabricated wall surface with the hole area removed 、 and , the three non-collinear points are connected to form a candidate plane, and the plane equation is as follows (9); (9); In formula (9), The coordinates are , The coordinates are , The coordinates are ; Then use the difference multiplication method to calculate the parameters in formula (9) 、 、 Calculate, see the following formula (10): (10); In formula (10), Represented by three non-collinear points The normal vector of the determined plane, that is, the vector perpendicular to the plane; 、 、 are the unit basis vectors in the 3D Cartesian coordinate system: The unit vector corresponding to the x-axis direction, The unit vector corresponding to the y-axis direction, The unit vector corresponding to the z-axis direction; After expanding formula (10), we can get: 、 、 、 ; 、 、 is the normal vector Components in the three coordinate axes: is the component of the normal vector in the x-axis direction, is the component of the normal vector in the y-axis direction, is the component of the normal vector in the z-axis direction; S412, for each point in the point cloud data of the prefabricated wall surface after removing the hole area , calculate its weighted distance to the plane , see the following formula (11) for details: (11); In formula (11), the radius of curvature is Yes The curvature value The reciprocal of ; S413, set dynamic threshold: for each point , according to its curvature radius Calculating dynamic thresholds , see the following formula (12) for details: (12); In formula (12), BaseThreshold is the global reference threshold; for the curvature radius : curvature value The bigger, The smaller the dynamic threshold The stricter; S414, determine whether the point meets the conditions: for each point ,when , then it is an inlier, and the number of inliers that meet the conditions is counted. When the number of inliers Ninlier of the current plane is greater than the number of inliers of all previous planes, the optimal plane is updated, and the updated optimal plane is regarded as the global reference plane of the current point cloud data. S42. For the point cloud data of the prefabricated wall surface with the hole area removed, adaptively adjust the size of the dynamic window based on the point cloud density, and calculate the deviation value from the center point of the local surface corresponding to each dynamic window to the global reference plane; S421. Calculate the dynamic window size, as shown in the following formula (13): (13); In formula (13), Represents the minimum size of the window; local elevation standard deviation , is the elevation value of the i-th point in the point cloud, i.e., the z coordinate, is the average elevation value of the point cloud in the window, , is the total number of point clouds in the window; point cloud density , represents the reference radius, ; Formula (13) expresses that Calculate the dynamic window size from the calculated dynamic window size and the minimum window size Select the maximum value as the dynamic window size , that is, the dynamic window size is not less than the minimum window size ; S422, point set in dynamic window , use the moving least squares method MLS to fit the quadratic surface, as shown in the following formula (14), and calculate the center point of the local surface , see the following formula (15) for details: (14); (15); In formula (14) and formula (15), is the elevation coordinate perpendicular to the wall, 、 is the position coordinate in the wall plane, and through formula (14), Expressed as and Function used to fit the local surface; 、 is the principal curvature coefficient of the surface, which represents the opening direction and curvature along the x and y directions respectively. represents a convex surface opening upward along the x direction, represents a concave surface opening downward, represents a convex surface opening upward along the y direction, Indicates a concave surface with an opening downward. When the wall surface is partially convex, and Significantly greater than zero, and the concave area corresponds to a negative value less than zero; is the coupled curvature coefficient of the surface, representing The degree of saddle distortion of the plane, When the surface is The plane presents a saddle-shaped structure. The larger the value, the more significant the saddle distortion is. Non-zero values ​​are often detected at wall joints or edges. value; 、 is the linear slope coefficient of the surface, representing the inclination along the x and y directions respectively, Indicates downhill in the positive x direction, Indicates downhill in the negative x direction; Indicates downhill in the positive y direction, Indicates downhill in the negative y direction, for example: a prefabricated wall installed at an angle or It shows significant non-zero values; is a constant term that represents the base elevation offset of the surface. It is used to adjust the base height of the surface to align it with the global coordinate system. It reflects the overall elevation difference between the local window and the global plane. represents the point set in the dynamic window, and the dynamic window size is adaptively adjusted by formula (13); 、 、 is the point set in the dynamic window interior point The x, y, and z axis coordinate values ​​in three-dimensional space; Represents the point set in the dynamic window the number of inliers; S43: Calculate the center point of the local surface Perpendicular distance to the global reference plane , see the following formula (16) for details: (16); In formula (16), 、 、 Represent the center points of the local surfaces The x, y, and z axis coordinate values, 、 、 、 The parameters of the plane equation of the global reference plane fitted in step S41 are the same as those in equations (9) and (11). 、 、 、 has the same meaning; Then calculate the average deviation of the local surface center point to the global reference plane The average deviation reflects the overall flatness level of the wall surface as shown in the following formula (17); calculate the standard deviation from the center point of the local surface to the global reference plane , see the following formula (18); calculate the maximum deviation from the center point of the local surface to the global reference plane , see the following formula (19); (17); (18); (19); In formula (17)-formula (19), Represents the number of local surfaces.

[0018] Finally, the surface flatness of the prefabricated wall is evaluated by the average deviation, standard deviation and maximum deviation: average deviation Reflects the overall flatness level of the prefabricated wall surface; standard deviation Measures the degree of dispersion of the deviation distribution; maximum deviation Locate the critical defect areas that need to be repaired on the prefabricated wall surface, and the detection result is the average deviation , the maximum deviation , and the surface inspection results of the prefabricated wall are qualified.

[0019] Performance Analysis: (1) Analysis of sample test results: The prefabricated wall samples numbered W-001 and W-002 were tested for flatness in multiple areas. The flatness test results are shown in Table 1 below: W-001 sample in the hole edge area (average deviation , standard deviation ) and leveling the central area ( , ) meet the flatness requirements (maximum deviation ), it was judged as qualified; W-002 sample was judged as qualified due to the deviation of the hole edge area exceeding the limit ( ), was judged as unqualified, but the lower right concave-convex area still met the standards ( ), indicating that the flatness detection method of this embodiment can accurately locate local defects on the surface of prefabricated walls.

[0020] Table 1

[0021] (2) Algorithm Performance Comparison: The curvature-compensated RANSAC in the embodiment of the present invention was compared with the traditional RANSAC. The experimental results are shown in Table 2 below. The plane fitting deviation of the embodiment of the present invention was reduced from 1.1 mm to 0.3 mm, an improvement of 72.7% compared with the traditional RANSAC; the hole edge misjudgment rate was reduced from 22.7% to 4.2%, a reduction of 81.5%; and the local defect detection rate was increased from 76.3% to 98.5%, effectively solving the problem of complex surface interference. The advantages of the embodiment of the present invention are derived from the dynamic correction of the non-uniform curvature value of the hole edge and the ability of the adaptive weight to suppress noise points.

[0022] Table 2

[0023] (3) Regional deviation feature analysis: The W-001 sample was tested in different regions. The test results are shown in Table 3 below: When the dynamic window size D = 0.15 is used in the hole edge area, the detection sensitivity and stability can be balanced (standard deviation ); A smaller window (D = 0.10 m) is used in the leveling center area to ensure high accuracy (mean deviation The transition zone, i.e., the outer edge of the hole, is smoothed by using a window size gradient strategy (D = 0.12 m) to avoid sudden deviation (maximum deviation). ). The adaptability of the dynamic window optimization strategy to different regional characteristics was verified.

[0024] Table 3

[0025] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting the flatness of prefabricated walls with holes based on point clouds, characterized by: The specific steps include: (1) Collect point cloud data of the prefabricated wall surface and construct a point cloud dataset of the prefabricated wall surface; (2) Using a dual-threshold screening mechanism of curvature features and the normal vector angle of the neighboring points, the point cloud data in the point cloud dataset is downsampled while retaining the concave and convex areas of the wall and the edge features of the holes; (3) Align the downsampled point cloud data with the BIM model of the prefabricated wall, mark the hole area, and generate a hole area mask to eliminate the interference of invalid data in the hole area; (4) Based on the curvature compensation RANSAC algorithm, the global reference plane is fitted, and the deviation value between the local surface center point and the global reference plane is calculated through the dynamic window to evaluate the surface flatness of the prefabricated wall.

2. The method for detecting the flatness of a prefabricated wall with holes based on a point cloud according to claim 1, characterized in that: The point cloud data of the prefabricated wall surface is collected by using a 3D line laser camera.

3. The method for detecting the flatness of a prefabricated wall with holes based on point cloud according to claim 1, characterized in that: The specific steps of downsampling the point cloud data in the point cloud dataset using the dual threshold screening mechanism of curvature features and neighboring point normal vector angles while retaining the concave and convex areas of the wall and the edge features of the holes are as follows: S21. For each point in the point cloud dataset , use KD tree to search for its nearest multiple neighbors, build a neighborhood point set, and calculate the covariance matrix of the neighborhood point set. The covariance matrix The calculation formula is shown in the following formula (1): （1）； In formula (1), the neighborhood centroid coordinates , Neighboring points in the neighborhood point set the number of Represents the transpose operation of the vector, which transforms the row vector Convert to column vector , so that the outer product operation result is a 3×3 covariance matrix; Then the covariance matrix Perform eigenvalue decomposition to obtain eigenvalues , the maximum eigenvalue The direction in which the data expands fastest is the direction with the smallest curvature and the smallest eigenvalue. The direction in which the corresponding data is most concentrated is the direction with the greatest curvature; Among them, the minimum eigenvalue The corresponding eigenvector is the point Normal vector ,point The curvature value Calculated by the following formula (2): （2）； In formula (2), the curvature value It is a local curvature metric based on a set of neighborhood points. It quantifies the curvature of the local surface. High-curvature wall areas will be significantly different from flat wall areas. S22, all points The curvature values ​​of the point cloud dataset are summed and then divided by the total number of points in the point cloud dataset to obtain the mean curvature of the point cloud dataset. , see the following formula (3) for details: （3）； In formula (3), Represents the total number of points in the point cloud dataset, is the curvature of the i-th point; Then calculate the curvature threshold , the calculation process is shown in the following formula (4): （4）； In formula (4), represents the standard deviation of curvature; S23, in order to capture the tiny concave and convex defects on the surface of the prefabricated wall, calculate the points All its neighbors The normal vector angle , and take the normal vector angle The maximum normal angle in , normal vector angle The calculation of is shown in the following formula (5): （5）； In formula (5), Representative Points The normal vector of Representative neighbor The normal vector of S24, retain the point cloud in the point cloud dataset that satisfies the following formula (6) , thereby retaining points with high curvature or large normal vector angles, that is, preferentially capturing hole edges and surface uneven areas, and realizing downsampling processing of point cloud data; （6）； In formula (6), Representative Points The curvature value of Representative Points Its neighbors The maximum normal vector angle, represents the curvature threshold, represents the angle threshold, is the set value.

4. The method for detecting flatness of prefabricated walls with holes based on point cloud according to claim 3, characterized in that: The specific steps of aligning the downsampled point cloud data with the BIM model of the prefabricated wall, marking the hole area, and generating the hole area mask are as follows: S31. Analyze the BIM model of the prefabricated wall to extract the boundary vertex set of the holes on the prefabricated wall, and determine the boundary point set of the holes based on the high curvature point cloud data retained after downsampling. Then, construct a bidirectional deviation function and match the boundary vertices in the BIM model with the hole boundary points in the point cloud dataset using a bidirectional distance deviation iterative optimization algorithm. S32. First, the positioning deviation of the point cloud is compensated according to the distance deviation calculated by the constructed bidirectional deviation function. Then, the dynamic expansion radius is calculated according to the density of the point cloud, and the normal vector direction of the point cloud is obtained. For each hole boundary point after positioning deviation compensation, the corresponding dynamic expansion radius value is moved along the positive direction of its normal vector to generate an expansion point, determine the expansion area, and obtain the mask of the hole area. Finally, the point cloud in the mask is removed to obtain the point cloud data of the prefabricated wall surface with the hole area removed.

5. The method for detecting flatness of prefabricated walls with holes based on point cloud according to claim 4, characterized in that: The formula of the bidirectional deviation function is shown in the following formula (7): （7）； In formula (7), Represents the distance deviation, represents the Euclidean distance, Represents the coordinates of the boundary vertices in the boundary vertex set, Represents the number of boundary vertices in the boundary vertex set; is the coordinate of the hole boundary point in the boundary point set, Represents the number of hole boundary points in the boundary point set; Represents the boundary vertex The nearest match of Represents the hole boundary point The nearest match of ; The specific steps of matching through the bidirectional distance deviation iterative optimization algorithm are: (a) The boundary vertices Project to the boundary point set and calculate the distance to the nearest point; (b) The hole boundary points Project to the boundary vertex set and calculate the nearest point distance; (c) Minimize the bidirectional deviation through singular value decomposition (SVD) and update the translation and rotation matrix of the BIM model; (d) When the distance deviation is less than The optimization is terminated when the set threshold or the maximum number of iterations is reached.

6. The method for detecting flatness of prefabricated walls with holes based on point cloud according to claim 4, characterized in that: The dynamic expansion radius is calculated according to the density of the point cloud, and the calculation formula is shown in the following formula (8): （8）； In formula (8), represents the dynamic expansion radius, represents the average spacing of the local point cloud, is the coordinate of the hole boundary point in the boundary point set, Represents the hole boundary point The coordinates of the nearest neighbor, Hole boundary point The nearest neighbor of The number of 7. The method for detecting flatness of prefabricated walls with holes based on point cloud according to claim 4, characterized in that: The specific steps of fitting the global reference plane based on the curvature compensation RANSAC algorithm and calculating the deviation value between the local surface center point and the global reference plane through the dynamic window are as follows: S41. Fitting the global reference plane based on the curvature-compensated RANSAC algorithm: S411, first randomly select three non-collinear points from the point cloud data of the prefabricated wall surface with the hole area removed 、 and , the three non-collinear points are connected to form a candidate plane, and the plane equation is as follows (9); （9）； In formula (9), The coordinates are , The coordinates are , The coordinates are ; Then use the difference multiplication method to calculate the parameters in formula (9) 、 、 Calculate, see the following formula (10): （10）； In formula (10), Represented by three non-collinear points The normal vector of the determined plane, that is, the vector perpendicular to the plane; 、 、 are the unit basis vectors in the 3D Cartesian coordinate system: The unit vector corresponding to the x-axis direction, The unit vector corresponding to the y-axis direction, The unit vector corresponding to the z-axis direction; After expanding formula (10), we can get: 、 、 、 ; 、 、 is the normal vector Components in the three coordinate axes: is the component of the normal vector in the x-axis direction, is the component of the normal vector in the y-axis direction, is the component of the normal vector in the z-axis direction; S412, for each point in the point cloud data of the prefabricated wall surface after removing the hole area , calculate its weighted distance to the plane , see the following formula (11) for details: （11）； In formula (11), the radius of curvature is Yes The curvature value The reciprocal of ; S413, set dynamic threshold: for each point , according to its curvature radius Calculating dynamic thresholds , see the following formula (12) for details: （12）； In formula (12), BaseThreshold is the global baseline threshold; S414, determine whether the point meets the conditions: for each point ,when , then it is an inlier, and the number of inliers that meet the conditions is counted. When the number of inliers Ninlier of the current plane is greater than the number of inliers of all previous planes, the optimal plane is updated, and the updated optimal plane is regarded as the global reference plane of the current point cloud data. S42. For the point cloud data of the prefabricated wall surface with the hole area removed, adaptively adjust the size of the dynamic window based on the point cloud density, and calculate the deviation value from the center point of the local surface corresponding to each dynamic window to the global reference plane; S421. Calculate the dynamic window size, as shown in the following formula (13): （13）； In formula (13), Represents the minimum size of the window; local elevation standard deviation , is the elevation value of the i-th point in the point cloud, i.e., the z coordinate, is the average elevation value of the point cloud in the window, , is the total number of point clouds in the window; point cloud density , represents the reference radius; Formula (13) expresses Calculate the dynamic window size from the calculated dynamic window size and the minimum window size Select the maximum value as the dynamic window size ; S422, point set in dynamic window , use the moving least squares method MLS to fit the quadratic surface, as shown in the following formula (14), and calculate the center point of the local surface , see the following formula (15) for details: （14）； （15）； In formula (14) and formula (15), is the elevation coordinate perpendicular to the wall, 、 is the position coordinate in the wall plane, and through formula (14), Expressed as and Function used to fit the local surface; 、 are the principal curvature coefficients of the surface, representing the opening direction and curvature along the x and y directions respectively; is the coupled curvature coefficient of the surface, representing The degree of saddle distortion of the plane; 、 is the linear slope coefficient of the surface, representing the inclination along the x and y directions respectively; is a constant term, representing the base elevation offset of the surface; represents the point set in the dynamic window, and the size of the dynamic window is adaptively adjusted by formula (13); 、 、 is the point set in the dynamic window interior point The x, y, and z axis coordinate values ​​in three-dimensional space; Represents the point set in the dynamic window the number of inliers; S43: Calculate the center point of the local surface Perpendicular distance to the global reference plane , see the following formula (16) for details: （16）； In formula (16), 、 、 Represent the center points of the local surfaces The x, y, and z axis coordinate values, 、 、 、 The parameters of the plane equation of the global reference plane fitted in step S41; Then calculate the average deviation of the local surface center point to the global reference plane The average deviation reflects the overall flatness level of the wall surface as shown in the following formula (17); calculate the standard deviation from the center point of the local surface to the global reference plane , see the following formula (18); calculate the maximum deviation from the center point of the local surface to the global reference plane , see the following formula (19); （17）； （18）； （19）； In formula (17)-formula (19), Represents the number of local surfaces; Finally, the surface flatness of the prefabricated wall is evaluated by the average deviation, standard deviation and maximum deviation: average deviation Reflects the overall flatness level of the prefabricated wall surface; standard deviation Measures the degree of dispersion of the deviation distribution; maximum deviation Locate critical defective areas requiring repair on precast wall surfaces.

Images