Assembly building installation precision intelligent calibration system fusing point cloud data and BIM

Abstract

This invention relates to the field of prefabricated building installation accuracy control technology, specifically disclosing an intelligent calibration system for prefabricated building installation accuracy that integrates point cloud data and BIM. The system collects on-site point clouds and acquires design geometric information; after distance attenuation compensation for the point cloud reflection intensity, it generates on-site and design projection intensity maps along the normal direction of the component installation plane, and obtains the in-plane pre-translation amount through phase correlation calculation; using this pre-translation amount as the prior center, it constructs a spatial transformation objective function containing an exponentially growing dynamic penalty coefficient, and solves the final spatial transformation parameters through weighted centroid closed-loop inversion iteration; it calculates the signed vertical distance along the normal of the aligned point cloud and removes outliers to form a deviation field, outputting translation or rotation fine-tuning amounts according to the median deviation of the joint sub-region; this invention solves the in-plane sliding problem in the registration of flat components, improving the accuracy of installation deviation calculation and calibration reliability.

Description

Technical Field

[0001] This invention relates to the field of installation accuracy control technology for prefabricated buildings, specifically to an intelligent calibration system for installation accuracy of prefabricated buildings that integrates point cloud data and BIM. Background Technology

[0002] In prefabricated buildings, prefabricated components (such as wall panels, beams, columns, and stairs) are manufactured in factories and transported to the site for assembly. The installation accuracy between components directly affects structural safety and functionality. Currently, 3D laser scanning technology is widely used to acquire on-site point cloud data of installed components. By registering and comparing this data with the BIM model from the design phase, the actual installation deviation of the components is calculated, thus guiding calibration. Among these methods, the Iterative Closest Point (ICP) algorithm and its improved forms are the mainstream methods for registering point clouds with BIM models.

[0003] Existing technologies suffer from the problem that ICP registration of large-area flat precast wall panels is trapped in a local optimum solution due to the sparse geometric features. This invention first uses reflection intensity information to independently obtain the in-plane pre-translation amount as a priori center through phase correlation calculation. Then, it constructs a spatial transformation objective function with a dynamic penalty coefficient that grows exponentially with the number of iterations. Through progressive forced constraints, the translation vector gradually converges to the pre-translation amount, thereby completely eliminating the registration ambiguity in the in-plane translation degrees of freedom and ensuring the accuracy of subsequent installation deviation calculations. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent calibration system for the installation accuracy of prefabricated buildings that integrates point cloud data and BIM, so as to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: An intelligent calibration system for the installation accuracy of prefabricated buildings that integrates point cloud data and BIM includes: The data acquisition module is used to collect on-site 3D point cloud data of prefabricated building components and obtain the corresponding design geometric information of the components; The pre-translation calculation module generates projection intensity maps along the normal direction of the component installation plane based on the on-site 3D point cloud data and design geometric information. It then performs phase correlation calculations on the two projection intensity maps to obtain the in-plane pre-translation amount. The objective function construction module establishes a spatial transformation objective function with a dynamic penalty coefficient, using the in-plane pre-translation as the constraint center. The dynamic penalty coefficient increases with the number of iterations. The iterative solution module inputs the on-site 3D point cloud data and design geometric information into the spatial transformation objective function for iterative solution. After each iteration, the dynamic penalty coefficient is increased to force the in-plane translation to gradually converge to the pre-translation amount in the plane. When the iteration stops, the final spatial transformation parameters are obtained. The deviation calibration module uses the final spatial transformation parameters to align the on-site 3D point cloud data to the design geometry, calculates the deviation distribution between the aligned point cloud and the design geometry, and generates installation calibration values ​​based on the deviation distribution.

[0006] As a further aspect of the present invention: obtaining the in-plane pre-translation amount specifically includes: The original reflection intensity is attenuated and compensated by the slant distance from each point in the point cloud to the scanning device to obtain the corrected intensity value; Project the point cloud carrying the correction intensity value onto the component installation plane along the normal direction, divide it into equally spaced grids, take the median of the correction intensity value of all points in each grid as the projection intensity of the corresponding grid, and generate the on-site projection intensity map. The design geometric information is discretized into an ideal point set with the same grid resolution as the on-site point cloud. Each ideal point is assigned a theoretical intensity value based on the preset material reflectivity attribute in the design geometric information. The design projection intensity map is generated by taking the mode after dividing the points into the same grid. Normalized cross power spectrum calculation and inverse Fourier transform are performed on the on-site projected intensity map and the design projected intensity map to extract the peak coordinates as the in-plane pre-translation amount.

[0007] As a further aspect of the present invention: obtaining the correction intensity value specifically includes: Analyze the original reflection intensity value corresponding to each point and the spatial rectangular coordinates recorded by the scanning device, and calculate the Euclidean slant distance from the corresponding point to the origin of the scanning device from the spatial rectangular coordinates; Based on the distance-attenuation curve calibrated by the scanning equipment at the factory, look up the theoretical attenuation factor corresponding to the Euclidean slant distance; Divide the original reflection intensity value by the theoretical attenuation factor to obtain the corrected intensity value that eliminates the influence of distance.

[0008] As a further aspect of the present invention: the establishment of the spatial transformation objective function containing dynamic penalty coefficients specifically includes: Using the in-plane pre-translation amount as the prior center of the translation vector, a quadratic penalty term is constructed. The quadratic penalty term calculates the squared deviation between the currently estimated translation vector and the prior center. Set a baseline penalty weight and define an amplification factor that grows exponentially with each iteration round. Multiply the baseline penalty weight by the amplification factor to obtain the dynamic penalty coefficient for the current round. The quadratic penalty term is added to the sum of the squared distances from the point cloud to the design geometry to form the spatial transformation objective function, where the dynamic penalty coefficient serves as the multiplier of the quadratic penalty term.

[0009] As a further aspect of the present invention: the step of multiplying the baseline penalty weight by an amplification factor to obtain the dynamic penalty coefficient for the current round specifically includes: The baseline penalty weight in the first iteration is assigned a preset positive number; The exponential growth factor is calculated by using the natural constant as the base and the value after subtracting one from the current iteration round as the exponent. The baseline penalty weight is multiplied by the exponential growth factor, and the resulting product is used as the dynamic penalty coefficient for the current iteration round.

[0010] As a further aspect of the present invention: the iterative solution of inputting on-site 3D point cloud data and design geometric information into a spatial transformation objective function specifically includes: In the current iteration, establish the correspondence between each point in the on-site 3D point cloud and the nearest surface point in the design geometry information; Based on the correspondence, the squared distances from the joint point cloud to the design geometry and the squared translation deviation multiplied by the dynamic penalty coefficient are used to simultaneously solve for the rotation and translation parameters of the current cycle through least squares closed inversion. The rotation and translation parameters are applied to the 3D point cloud in the field to update the spatial position of the point cloud. The difference between the translation parameters before and after the update is compared. The iteration stops when the difference between the translation parameters is less than a preset threshold. The rotation and translation parameters of the current round are output as the final spatial transformation parameters.

[0011] As a further aspect of the present invention: the simultaneous solution of the rotation and translation parameters of the current cycle through least-squares closed-form inversion specifically includes: Calculate the weighted centroid of each point in the current corresponding point pair in the field 3D point cloud and the weighted centroid of each corresponding point in the design geometry information. The weighting coefficient is determined by the dynamic penalty coefficient and the distance between the point cloud and the design geometry. Subtract the weighted centroid from each point in the 3D point cloud on site, subtract the weighted centroid from each corresponding point in the design geometric information, construct the covariance matrix using dot product, and perform polar coordinate decomposition on the covariance matrix to obtain the rotation parameters. The rotation parameters are applied to the weighted centroid of the 3D point cloud at the site. The weighted centroid of the 3D point cloud at the site after rotation is subtracted from the weighted centroid of the design geometry information to obtain the translation parameters.

[0012] As a further aspect of the present invention: the step of generating installation calibration values ​​based on the deviation distribution specifically includes: Project each point in the aligned on-site 3D point cloud data along the normal direction of the design geometry information, and record the signed distance between the projected point and the corresponding design geometry surface point to form a deviation field. The deviation field is divided into several sub-regions according to the joint area of ​​the component, and the median of all signed distances in each sub-region is calculated as the representative deviation of the corresponding sub-region. When the representative deviation of a certain sub-region exceeds the preset allowable threshold, the installation calibration value containing translational and rotational fine-tuning amounts is output according to the positive and negative signs of the deviation of the corresponding sub-region and the direction of the deviation gradient between adjacent sub-regions.

[0013] As a further aspect of the present invention: the formation of the deviation field specifically includes: Taking each point in the aligned point cloud as the center, expand equidistantly on both sides along the normal direction of the design geometry information to form a line segment neighborhood. Retrieve the triangular facets in the design geometry information that intersect with the line segment neighborhood as candidate corresponding faces. Select the surface with the smallest angle between the normal direction and the normal direction of the current point in the point cloud from the candidate corresponding surfaces as the target surface, and calculate the signed perpendicular distance from the current point to the target surface. When the absolute value of the signed vertical distance is greater than three times the median of the local density of the point cloud, the corresponding point is identified as an outlier and removed. The signed vertical distances of the remaining points are then arranged according to their spatial coordinates to form an offset field.

[0014] The beneficial effects of this invention are: (1) This invention effectively solves the problem of in-plane slippage in registration of large-area flat precast wall panels caused by sparse geometric features through a joint strategy of phase-correlated pre-translation estimation and dynamic penalty constraint ICP. In the traditional ICP method, since the wall panel surface is approximately an ideal plane and lacks protruding corners or edges, the registration process is prone to getting stuck in a local optimum where the plane fits well but the wall is offset by several centimeters in the horizontal or vertical direction. This invention first uses the frequency domain global phase characteristics of reflection intensity information to obtain a stable in-plane pre-translation amount independently of local geometric features, and then forces the subsequent iterative translation vector to converge to the pre-translation amount through an exponentially increasing dynamic penalty coefficient, thereby eliminating the ambiguity of the in-plane translation degree of freedom, reducing the positioning error of the final registration result in two directions in the plane, and ensuring the accuracy of subsequent calculation of installation accuracy indicators such as wall panel joint width and verticality.

[0015] (2) This invention generates calibration values ​​based on a signed vertical deviation field and a representative deviation of the median in a sub-region, enabling the output of quantitative fine-tuning instructions for specific installation error patterns. The signed vertical distance is calculated along the design geometric normal direction from each point in the aligned point cloud, and outliers are removed to form a deviation field. Sub-regions are then divided according to the actual joint area of ​​the component, and the median is taken as the representative deviation, avoiding interference from local noise or individual abnormal measuring points in deviation judgment. When the deviation exceeds the limit, translational and rotational fine-tuning amounts are output based on the sign of the deviation in the sub-region and the gradient direction of the deviation with adjacent sub-regions, directly guiding construction personnel to adjust the adjustment amounts of braces, shims, or jacks, reducing the time spent on repeated on-site calibration and improving installation adjustment efficiency. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 As shown, this invention is an intelligent calibration system for the installation accuracy of prefabricated buildings that integrates point cloud data and BIM, comprising: The data acquisition module is used to collect on-site 3D point cloud data of prefabricated building components and obtain the corresponding design geometric information of the components; The pre-translation calculation module generates projection intensity maps along the normal direction of the component installation plane based on the on-site 3D point cloud data and design geometric information. It then performs phase correlation calculations on the two projection intensity maps to obtain the in-plane pre-translation amount. The objective function construction module establishes a spatial transformation objective function with a dynamic penalty coefficient, using the in-plane pre-translation as the constraint center. The dynamic penalty coefficient increases with the number of iterations. The iterative solution module inputs the on-site 3D point cloud data and design geometric information into the spatial transformation objective function for iterative solution. After each iteration, the dynamic penalty coefficient is increased to force the in-plane translation to gradually converge to the pre-translation amount in the plane. When the iteration stops, the final spatial transformation parameters are obtained. The deviation calibration module uses the final spatial transformation parameters to align the on-site 3D point cloud data to the design geometry, calculates the deviation distribution between the aligned point cloud and the design geometry, and generates installation calibration values ​​based on the deviation distribution.

[0020] The data acquisition module is used to collect on-site 3D point cloud data of installed components of prefabricated buildings and obtain the corresponding design geometric information of the components, specifically including: When collecting 3D point cloud data of installed components in prefabricated buildings, 3 to 4 time-of-flight depth sensors are deployed around the perimeter of the component to be measured. Each sensor includes an infrared transmitter and receiver array, capable of simultaneously outputting depth and intensity images. All sensors are connected to the same control terminal via a synchronization trigger line. The terminal sends a pulse signal to cause all sensors to start collecting data at the same time, thus avoiding misalignment of point cloud stitching due to time asynchrony.

[0021] Based on the spatial relationships between the sensors, a rigid body transformation calibration is performed beforehand. During calibration, a calibration plate with a known geometry is placed near the component's installation location, and each sensor simultaneously acquires depth images of this calibration plate. By minimizing the reprojection error between feature points detected by each sensor on the calibration plate, the rotation matrix and translation vector between each pair of sensors are calculated and used as the pre-calibrated rigid body transformation parameters. During actual data acquisition, these parameters are used to transform each pixel in the depth images output by each sensor from its own coordinate system to a unified construction site coordinate system, thereby fusing all depth images into a complete 3D point cloud dataset of the site.

[0022] Meanwhile, during the component manufacturing stage, a passive RFID tag is pre-embedded on the surface of the component, storing a unique identification code for that component. Before scanning at the construction site, a handheld reader is brought close to the component surface to read the identification code within the tag. This code is then transmitted wirelessly to a cloud database, which pre-stores the design geometry information of all components, including the component's three-dimensional outline, the normal direction of the mounting plane, and the preset material reflectivity properties of each surface. Upon receiving the code, the cloud database returns the corresponding design geometry information to the on-site control terminal, thereby obtaining the component's design geometry information.

[0023] In the pre-translation calculation module, based on the on-site 3D point cloud data and design geometric information, projection intensity maps are generated along the normal direction of the component installation plane. Phase correlation calculations are performed on the two projection intensity maps to obtain the in-plane pre-translation amount, specifically including: After obtaining the on-site 3D point cloud data and design geometric information, attenuation compensation is performed on the original reflection intensity of each point in the point cloud. Each point recorded by each scanning device contains the original reflection intensity value and the spatial rectangular coordinates of that point in the scanning device's own coordinate system. Based on these spatial rectangular coordinates, the Euclidean slant distance between that point and the origin of the scanning device is calculated, which is calculated by taking the square root of the sum of the squares of the three coordinate components. Each scanning device is calibrated with a distance-attenuation curve at the factory, which records the theoretical attenuation factor corresponding to different Euclidean slant distances. Based on the calculated Euclidean slant distance, the corresponding theoretical attenuation factor is retrieved from this curve. Finally, the original reflection intensity value is divided by this theoretical attenuation factor, and the result is used as the corrected intensity value for that point. This corrected intensity value eliminates the influence of distance on the reflection intensity.

[0024] Using on-site 3D point cloud data carrying corrected intensity values, the normal direction of the installation plane of the component under test is determined. This normal direction is extracted from the design geometry information. Each point in the point cloud is projected perpendicularly along the normal direction onto an infinitely large plane containing the component's installation plane. After projection, each point obtains a two-dimensional coordinate within the plane. This plane is divided into fixed, equally spaced grids, with the grid spacing being 1.5 times the average point spacing of the point cloud. For each grid, the corrected intensity values ​​corresponding to all points falling within that grid are collected, and the median of these values ​​is taken as the projected intensity value of that grid. The projected intensity values ​​of all grids are arranged in rows and columns to form an on-site projected intensity map.

[0025] For the design geometry information, it is discretized into an ideal point set with the same raster resolution as the on-site projection intensity map. Specifically, on the 3D surface of the design geometry information, a dense point cloud is generated by sampling at equal intervals, with the sampling interval consistent with the average point spacing of the on-site point cloud. The material reflectivity attributes of each surface of the component are pre-stored in the design geometry information, such as a certain value for the reflectivity of the concrete surface. Based on the surface material to which each ideal point belongs, a theoretical strength value is assigned to that point. This theoretical strength value is the normalized value of the reflectivity of that material at a standard distance. Subsequently, these ideal points are projected onto the component installation plane along the normal direction. Using the same raster division method as the on-site point cloud, for each raster, the theoretical strength values ​​of all ideal points falling within it are collected, and the mode of these values ​​is taken as the projection intensity value of that raster, thereby constructing the design projection intensity map.

[0026] After obtaining the on-site projected intensity map and the design projected intensity map, phase correlation calculations are performed on both. Two-dimensional fast Fourier transforms are then performed on both maps to obtain their respective frequency domain representations. The normalized cross-power spectrum is calculated by multiplying the complex conjugates of the frequency domain representations of the on-site projected intensity map and the design projected intensity map point-by-point, and then dividing by the magnitude of the product. The result is the normalized cross-power spectrum. A two-dimensional inverse Fourier transform is then performed on this cross-power spectrum to obtain a two-dimensional correlation function image. A distinct peak position will appear in this image.

[0027] Find the global maximum point in the graph of the correlation function and record the row and column coordinate offsets of this maximum point. These two offsets correspond to the translation amounts in two orthogonal directions within the component installation plane, in units of grid cells. Based on the actual physical size of the grid cells, convert the row and column offsets into in-plane translation amounts in millimeters, denoted as the in-plane pre-translation. This pre-translation reflects the initial offset estimate of the site point cloud relative to the design geometry within the component installation plane.

[0028] The extracted in-plane pre-translation is used as the prior center for subsequent constraint registration. This pre-translation does not depend on local geometric features, but only utilizes the frequency-domain global phase characteristics of reflection intensity information, thus providing a stable estimation effect for large-area, flat, and feature-sparse precast wall panels. After calculating this pre-translation, its output is used to establish a spatial transformation objective function that includes dynamic penalty coefficients.

[0029] In the objective function construction module, a spatial transformation objective function with dynamic penalty coefficients is established, using the in-plane pre-translation as the constraint center. The dynamic penalty coefficients increase with the number of iterations, specifically including: The in-plane pre-translation is used as the prior center of the translation vector in the subsequent registration process. This prior center represents the in-plane translation position to which the field point cloud should be adjusted under ideal conditions. Based on this, a quadratic penalty term is constructed. The specific calculation method of the quadratic penalty term is as follows: take the translation vector estimated in the current iteration, square the difference between the translation vector and the prior center in two orthogonal directions in the plane, and then add the two squared values ​​to obtain the sum of squared deviations. This sum of squared deviations is the value of the quadratic penalty term.

[0030] To allow for a larger translation adjustment space in the early stages of iteration, and to force the translation vector to approach the prior center in the later stages of iteration, a dynamic penalty coefficient needs to be introduced. This dynamic penalty coefficient gradually increases with the number of iterations. The construction of the dynamic penalty coefficient consists of two steps: first, setting a baseline penalty weight, which is a preset positive number, and in this embodiment, the value is taken as one-thousandth of the total number of points in the point cloud; second, defining an amplification factor that increases exponentially with the number of iterations, and multiplying the baseline penalty weight by this amplification factor, the resulting product is the dynamic penalty coefficient used in the current iteration.

[0031] The quadratic penalty term is added to the sum of squared distances from the point cloud to the design geometry to form the spatial transformation objective function. The sum of squared distances from the point cloud to the design geometry is calculated by summing the squared Euclidean distances from each point in the field point cloud after the currently estimated rotation and translation transformation to the nearest point on the design geometric surface. The dynamic penalty coefficient acts as a multiplier for the quadratic penalty term, representing its weight in the objective function. The overall objective function value equals the sum of squared distances from the point cloud to the design geometry plus the dynamic penalty coefficient multiplied by the quadratic penalty term.

[0032] The amplification factor in the dynamic penalty coefficient adopts an exponential growth method. Specifically, the exponential growth factor is calculated using the natural constant (approximately 2.71828) as the base and the value after subtracting one from the current iteration number as the exponent. For example, in the first iteration, the current iteration minus one equals zero, and the exponential growth factor is 1; in the second iteration, the exponent is 1, and the exponential growth factor equals the natural constant; in the third iteration, the exponent is 2, and the exponential growth factor equals the square of the natural constant, and so on.

[0033] The baseline penalty weight is assigned a preset positive number during the first iteration. This preset positive number is determined by counting the total number of points in the point cloud, dividing the total number of points by one thousand, and taking the integer part as the baseline penalty weight. If the total number of points is ten thousand, the baseline penalty weight is ten. This value ensures that the contribution of the initial penalty to the objective function is on the same order of magnitude as the sum of squared distances. Before each iteration, based on the current iteration round, the exponential growth factor is calculated according to the method described in paragraph four. Then, the baseline penalty weight is multiplied by this exponential growth factor to obtain the dynamic penalty coefficient used in the current round.

[0034] The calculated dynamic penalty coefficient for the current round is substituted into the spatial transformation objective function. This objective function is then used for iterative optimization: in each iteration, the rotation and translation parameters are solved simultaneously by minimizing the objective function, and the dynamic penalty coefficient is updated after the solution. Since the dynamic penalty coefficient grows exponentially with each round, the weight of the quadratic penalty term in later rounds is much greater than the sum of squared distances, thus forcing the translation vector to gradually converge to the prior center (i.e., the pre-translation amount in the plane). When the iteration stops, the deviation between the final translation vector and the pre-translation amount in the plane is compressed to below a preset threshold, achieving precise constraints on the in-plane translation.

[0035] In the iterative solution module, the on-site 3D point cloud data and design geometric information are input into the spatial transformation objective function for iterative solution. After each iteration, the dynamic penalty coefficient is increased to force the in-plane translation to gradually converge to the pre-translation amount in the plane. When the iteration stops, the final spatial transformation parameters are obtained, which specifically include: The spatial transformation objective function, along with the on-site 3D point cloud data and design geometry information, is input into the iterative solution process. At the beginning of each iteration, the on-site 3D point cloud is first transformed according to the current spatial transformation parameters (initially set as unit rotation and zero vector for translation). Then, the correspondence between each point in the transformed point cloud and the nearest surface point in the design geometry information is established. Specifically, for each point in the transformed point cloud, the nearest point (perpendicular foot or vertex) on all triangular faces of the design geometry information is calculated, and the point with the shortest distance is taken as the corresponding point. The spatial coordinates of the corresponding point and the normal direction of its surface are recorded. If the nearest distance of a point is greater than a preset threshold (which is 5 times the average point spacing of the point cloud), the point is considered to have no corresponding point and is skipped.

[0036] After establishing the correspondence, it is necessary to combine the sum of squared distances from the point cloud to the design geometry with the squared translation deviation multiplied by the dynamic penalty coefficient, and simultaneously solve for the rotation and translation parameters of the current round through least-squares closed-form inversion. Assume that the current round corresponds to a point cloud containing... There are 1 valid corresponding points, denoted as _____. The nearest point on the corresponding design geometry is denoted as Meanwhile, let the dynamic penalty coefficient for the current round be... The pre-translation amount in the plane is denoted as The translation vector to be solved is denoted as The rotation matrix is ​​denoted as The goal is to minimize the total weighted error, which is equal to the sum of the errors at all corresponding points. The sum plus in This represents the sum of squares of the components of a three-dimensional vector.

[0037] To obtain a closed-form solution, the weighted centroids of each point in the field point cloud and the corresponding points in the design geometry information are calculated. The weighting coefficients are determined by the dynamic penalty coefficient and the distance between the point cloud and the design geometry. Specifically, for the ... There are corresponding points, and their distance weights are defined as follows: in and The transformation parameters obtained in the previous iteration (during the first iteration) It is the identity matrix. (The zero vector). Then construct the combined weights. The weighted centroid of the point cloud at the site. and corresponding point weighted centroid Calculate according to the following formulas: ; The numerator is the sum of the coordinates of each point multiplied by their corresponding weights, and the denominator is the sum of all weights.

[0038] After obtaining the weighted centroid, subtract the weighted centroid from each point in the on-site point cloud. To obtain a decentralized point Subtract the weighted centroid from each corresponding point in the design geometry. To obtain a decentralized point Construct the weighted covariance matrix. The calculation method is as follows: for the three coordinate directions (i.e., X, Y, and Z directions), calculate the sum of the products of each pair of directions in turn. Specifically, the covariance matrix is ​​a 3x3 square matrix, and its i-th... Line 1 The elements of the column equal all corresponding Multiply The Each coordinate component is then multiplied by The The sum of all coordinate components. This can be expressed mathematically as: ; in This represents the X coordinate of the point. Represents the Y-coordinate. Represents the Z-coordinate, for Similarly.

[0039] Obtain the covariance matrix Then, polar coordinate decomposition is performed to obtain the rotation parameters. The specific implementation of polar coordinate decomposition is as follows: First, for... Performing singular value decomposition yields the product of three matrices, i.e. ;in and It is an orthogonal matrix. It is a diagonal matrix (diagonal elements are singular values). Then calculate... If the calculated If the determinant is negative (indicating reflection), then... Invert the last column and recalculate. The result is... This refers to the rotation parameters for the current round. Next, the translation parameters are calculated: applying the rotation parameters to the weighted centroid of the point cloud, we obtain... Then use the weighted centroid of the corresponding points in the design geometry. Subtract this result, that is The translation parameter This is the translation vector obtained in the current round.

[0040] The rotation parameters obtained in the current round Translation parameters It applies to the original 3D point cloud data of the site, that is, to each point calculate The global spatial position of the point cloud is updated. Then, the difference in translation parameters before and after the update is compared: the Euclidean distance between the current and previous translation parameters is calculated (the square root of the sum of the squares of the three directional components). When this difference is less than a preset threshold (e.g., 0.5 mm), the iteration stops, and the rotation and translation parameters for the current iteration are output as the final spatial transformation parameters; otherwise, the next iteration begins, increasing the dynamic penalty coefficient (exponentially) according to the current iteration, re-establishing the correspondence, and repeating the above solution process. The final spatial transformation parameters ensure that the deviation between the in-plane translation and the pre-translation amount is minimal, thus achieving high-precision registration between the point cloud and the designed geometry.

[0041] In the deviation calibration module, the final spatial transformation parameters are used to align the on-site 3D point cloud data to the design geometry information. The deviation distribution between the aligned point cloud and the design geometry is calculated, and installation calibration values ​​are generated based on the deviation distribution. Specifically, this includes: After obtaining the final spatial transformation parameters, these rotation and translation parameters are applied to each point in the original 3D point cloud data, causing the point cloud to rotate and translate as a whole, thus precisely aligning it with the design geometry in space. The aligned point cloud coordinate system is completely consistent with the coordinate system of the design geometry, and each point in the point cloud corresponds to an actual measured position near the surface of the design geometry.

[0042] To calculate the deviation distribution between the aligned point cloud and the design geometry, a deviation field is constructed. Centered on each point in the aligned point cloud, an equal distance is extended to both sides of the normal direction of the surface of the corresponding design geometry information. This extension distance is twice the average spacing of the local point cloud region at that point, thus forming a line segment neighborhood centered on that point. Within this line segment neighborhood, all triangular faces in the design geometry information are retrieved, and those triangular faces that intersect with the line segment neighborhood (i.e., the line segment passes through the plane of the triangular facet) are identified and used as candidate corresponding faces.

[0043] From the candidate faces, calculate the normal direction of each candidate face and the angle between this normal direction and the normal direction of the current point cloud point. The normal direction of the current point cloud point is obtained by performing principal component analysis on its surrounding neighboring points: take the 10 nearest points, calculate the covariance matrix of these points, and the eigenvector corresponding to the smallest eigenvalue is the normal direction. From all candidate faces, select the triangular face with the smallest normal angle as the target face. Then calculate the signed perpendicular distance from the current point cloud point to the target face. The sign of the signed perpendicular distance is determined by whether the point is located on the side pointed to by the normal of the target face or the opposite side, and the absolute value of the perpendicular distance is the perpendicular distance from the point to the plane containing the target face.

[0044] After obtaining the signed vertical distance for each point, outliers need to be removed. First, the local point cloud density around each point in the point cloud is calculated: within a sphere centered at that point and with a radius of 5 times the average point spacing, the number of neighboring points is counted, and this number is divided by the volume of the sphere to obtain the local density. Then, the median of the local density for all points is calculated. For each point's signed vertical distance, if its absolute value is greater than three times the median of the local density, that point is considered an outlier and removed from the bias calculation. The signed vertical distances of the remaining points are arranged according to their spatial coordinates (X, Y, Z directions) projected onto the component's mounting plane, forming a two-dimensional gridded bias field. Each grid cell stores the signed vertical distance value corresponding to that position.

[0045] After obtaining the complete deviation field, according to the joint area division rules of prefabricated components, the deviation field is divided into several sub-regions based on the actual joint positions of the components during installation (e.g., vertical and horizontal joints between wall panels). For each sub-region, the signed vertical distance values ​​of all grid cells within that region are collected, these values ​​are sorted from smallest to largest, and the value at the middle position (i.e., the median) is taken as the representative deviation of that sub-region. The median is calculated as follows: if there are N signed vertical distance values ​​in the sub-region, the value corresponding to the integer part of the (N+1)th division by 2 after sorting is the median.

[0046] The representative deviation of each sub-region is compared with a preset allowable threshold. The preset allowable threshold is determined according to the prefabricated building quality acceptance code, for example, the allowable deviation of wall panel verticality is 5 mm. When the representative deviation of a sub-region exceeds this allowable threshold, the sign of the deviation of the sub-region is first read (positive value indicates that the point cloud protrudes from the design surface, negative value indicates that it is concave). At the same time, the deviation gradient direction between the sub-region and the adjacent sub-region is calculated, that is, the difference in representative deviations of adjacent sub-regions divided by the distance between the center points of the two sub-regions. According to the sign of the deviation and the gradient direction, the installation calibration value is output: if the deviation is positive and the gradient points towards the joint, the translation fine adjustment amount that moves the entire sub-region into the design surface is output; if the deviation gradient shows a linear trend, the rotation fine adjustment amount around the joint axis is output. The translation fine adjustment amount is half of the negative number of the representative deviation of the sub-region, and the rotation fine adjustment amount is the product of the deviation gradient and the joint length divided by 2. The output calibration value is directly used to guide on-site construction adjustments.

[0047] The working principle of this invention is as follows: Multiple time-of-flight depth sensors are deployed around the perimeter of the component under test. Simultaneous triggering is used to acquire 3D point cloud data from the site, and the corresponding design geometry information is obtained from a cloud database by reading the RFID tags embedded on the component surface. Distance attenuation compensation is applied to the original reflection intensity of each point in the point cloud to obtain a corrected intensity value. A site projection intensity map and a design projection intensity map are generated along the normal direction of the component's installation plane. Phase correlation calculations are performed on the two projection intensity maps to obtain the in-plane pre-translation amount. Using this in-plane pre-translation amount as the prior center, a sum of squared distances from the point cloud to the design geometry is constructed, multiplied by a dynamic penalty coefficient, and then summed with the squared translation deviations before and after. The spatial transformation objective function is generated, where the dynamic penalty coefficient increases exponentially with each iteration. The point cloud and design geometry are input into this objective function for iterative solution. In each iteration, rotation and translation parameters are solved simultaneously through weighted centroid closed-loop inversion, and the point cloud position is updated until the translation change is less than a preset threshold. The final spatial transformation parameters are then output. These parameters are used to align the point cloud to the design geometry. The signed perpendicular distance is calculated along the normal direction, and outliers are removed to form a deviation field. Sub-regions are divided according to the component joint area, and the median is taken as the representative deviation. When the deviation exceeds the limit, the installation calibration value containing translation and rotation fine-tuning is output according to the sign of the deviation and the gradient direction.

[0048] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent calibration system for the installation accuracy of prefabricated buildings that integrates point cloud data and BIM, characterized in that: include: The data acquisition module is used to collect on-site 3D point cloud data of prefabricated building components and obtain the corresponding design geometric information of the components; The pre-translation calculation module generates projection intensity maps along the normal direction of the component installation plane based on the on-site 3D point cloud data and design geometric information. It then performs phase correlation calculations on the two projection intensity maps to obtain the in-plane pre-translation amount. The objective function construction module establishes a spatial transformation objective function with a dynamic penalty coefficient, using the in-plane pre-translation as the constraint center. The dynamic penalty coefficient increases with the number of iterations. The iterative solution module inputs the on-site 3D point cloud data and design geometric information into the spatial transformation objective function for iterative solution. After each iteration, the dynamic penalty coefficient is increased to force the in-plane translation to gradually converge to the pre-translation amount in the plane. When the iteration stops, the final spatial transformation parameters are obtained. The deviation calibration module uses the final spatial transformation parameters to align the on-site 3D point cloud data to the design geometry, calculates the deviation distribution between the aligned point cloud and the design geometry, and generates installation calibration values ​​based on the deviation distribution.

2. The intelligent calibration system for prefabricated building installation accuracy integrating point cloud data and BIM as described in claim 1, characterized in that, The obtained in-plane pre-translation amount specifically includes: The original reflection intensity is attenuated and compensated by the slant distance from each point in the point cloud to the scanning device to obtain the corrected intensity value; Project the point cloud carrying the correction intensity value onto the component installation plane along the normal direction, divide it into equally spaced grids, take the median of the correction intensity value of all points in each grid as the projection intensity of the corresponding grid, and generate the on-site projection intensity map. The design geometric information is discretized into an ideal point set with the same grid resolution as the on-site point cloud. Each ideal point is assigned a theoretical intensity value based on the preset material reflectivity attribute in the design geometric information. The design projection intensity map is generated by taking the mode after dividing the points into the same grid. Normalized cross power spectrum calculation and inverse Fourier transform are performed on the on-site projected intensity map and the design projected intensity map to extract the peak coordinates as the in-plane pre-translation amount.

3. The intelligent calibration system for prefabricated building installation accuracy integrating point cloud data and BIM as described in claim 2, characterized in that, The process of obtaining the corrected intensity value specifically includes: Analyze the original reflection intensity value corresponding to each point and the spatial rectangular coordinates recorded by the scanning device, and calculate the Euclidean slant distance from the corresponding point to the origin of the scanning device from the spatial rectangular coordinates; Based on the distance-attenuation curve calibrated by the scanning equipment at the factory, look up the theoretical attenuation factor corresponding to the Euclidean slant distance; Divide the original reflection intensity value by the theoretical attenuation factor to obtain the corrected intensity value that eliminates the influence of distance.

4. The intelligent calibration system for prefabricated building installation accuracy integrating point cloud data and BIM as described in claim 1, characterized in that, The establishment of the spatial transformation objective function including dynamic penalty coefficients specifically includes: Using the in-plane pre-translation amount as the prior center of the translation vector, a quadratic penalty term is constructed. The quadratic penalty term calculates the squared deviation between the currently estimated translation vector and the prior center. Set a baseline penalty weight and define an amplification factor that grows exponentially with each iteration round. Multiply the baseline penalty weight by the amplification factor to obtain the dynamic penalty coefficient for the current round. The quadratic penalty term is added to the sum of the squared distances from the point cloud to the design geometry to form the spatial transformation objective function, where the dynamic penalty coefficient serves as the multiplier of the quadratic penalty term.

5. The intelligent calibration system for prefabricated building installation accuracy integrating point cloud data and BIM as described in claim 4, characterized in that, The step of multiplying the baseline penalty weight by an amplification factor to obtain the dynamic penalty coefficient for the current round specifically includes: The baseline penalty weight in the first iteration is assigned a preset positive number; The exponential growth factor is calculated by using the natural constant as the base and the value after subtracting one from the current iteration round as the exponent. The baseline penalty weight is multiplied by the exponential growth factor, and the resulting product is used as the dynamic penalty coefficient for the current iteration round.

6. The intelligent calibration system for prefabricated building installation accuracy integrating point cloud data and BIM as described in claim 1, characterized in that, The step of iteratively solving the objective function by inputting on-site 3D point cloud data and design geometric information into the spatial transformation specifically includes: In the current iteration, establish the correspondence between each point in the on-site 3D point cloud and the nearest surface point in the design geometry information; Based on the correspondence, the squared distances from the joint point cloud to the design geometry and the squared translation deviation multiplied by the dynamic penalty coefficient are used to simultaneously solve for the rotation and translation parameters of the current cycle through least squares closed inversion. The rotation and translation parameters are applied to the 3D point cloud in the field to update the spatial position of the point cloud. The difference between the translation parameters before and after the update is compared. The iteration stops when the difference between the translation parameters is less than a preset threshold. The rotation and translation parameters of the current round are output as the final spatial transformation parameters.

7. The intelligent calibration system for prefabricated building installation accuracy integrating point cloud data and BIM as described in claim 6, characterized in that, The method of simultaneously solving for the rotation and translation parameters of the current cycle through least-squares closed-form inversion specifically includes: Calculate the weighted centroid of each point in the current corresponding point pair in the field 3D point cloud and the weighted centroid of each corresponding point in the design geometry information. The weighting coefficient is determined by the dynamic penalty coefficient and the distance between the point cloud and the design geometry. Subtract the weighted centroid from each point in the 3D point cloud on site, subtract the weighted centroid from each corresponding point in the design geometric information, construct the covariance matrix using dot product, and perform polar coordinate decomposition on the covariance matrix to obtain the rotation parameters. The rotation parameters are applied to the weighted centroid of the 3D point cloud at the site. The weighted centroid of the 3D point cloud at the site after rotation is subtracted from the weighted centroid of the design geometry information to obtain the translation parameters.

8. The intelligent calibration system for prefabricated building installation accuracy integrating point cloud data and BIM as described in claim 1, characterized in that, The process of generating installation calibration values ​​based on the deviation distribution specifically includes: Project each point in the aligned on-site 3D point cloud data along the normal direction of the design geometry information, and record the signed distance between the projected point and the corresponding design geometry surface point to form a deviation field. The deviation field is divided into several sub-regions according to the joint area of ​​the component, and the median of all signed distances in each sub-region is calculated as the representative deviation of the corresponding sub-region. When the representative deviation of a certain sub-region exceeds the preset allowable threshold, the installation calibration value containing translational and rotational fine-tuning amounts is output according to the positive and negative signs of the deviation of the corresponding sub-region and the direction of the deviation gradient between adjacent sub-regions.

9. The intelligent calibration system for prefabricated building installation accuracy integrating point cloud data and BIM as described in claim 8, characterized in that, The aforementioned constituting deviation field specifically includes: Taking each point in the aligned point cloud as the center, expand equidistantly on both sides along the normal direction of the design geometry information to form a line segment neighborhood. Retrieve the triangular facets in the design geometry information that intersect with the line segment neighborhood as candidate corresponding faces. Select the surface with the smallest angle between the normal direction and the normal direction of the current point in the point cloud from the candidate corresponding surfaces as the target surface, and calculate the signed perpendicular distance from the current point to the target surface. When the absolute value of the signed vertical distance is greater than three times the median of the local density of the point cloud, the corresponding point is identified as an outlier and removed. The signed vertical distances of the remaining points are then arranged according to their spatial coordinates to form an offset field.

Images