A graph frequency adaptive BIM and point cloud registration method

Abstract

The application discloses a kind of BIM and point cloud registration methods of graph frequency self-adaption, it is related to intelligent building technical field.The application at least includes following steps: S1: BIM component is triangulated, and the vertex after triangulation is reserved and is recorded as BIM point cloud;Scan the real component corresponding to BIM, and the point cloud after scanning is recorded as component point cloud;S2: the corresponding method is used to calculate the atlas cumulative distribution of component point cloud and BIM point cloud respectively;S3: the corresponding method is used to calculate feature point of component point cloud and BIM point cloud respectively;S4: according to the reserved feature point, using ICP algorithm registration BIM point cloud and component point cloud.The application does not need artificial setting feature point selection parameter, atlas cut-off frequency is automatically calculated according to point cloud, adaptively generates for matching feature point, improves the robustness when point cloud and BIM match.

Description

Technical Field

[0001] The present invention relates to the field of intelligent building technology, and in particular to a frequency-adaptive BIM and point cloud registration method. Background Art

[0002] Currently, the existing BIM and point cloud registration often directly selects all points or randomly selects some points, and then uses the iterative nearest neighbor algorithm for registration;

[0003] However, in actual scanning, the amount of point cloud data is often very large, and selecting all points for registration requires a lot of computing resources;

[0004] At the same time, since point cloud data is usually affected by scanning accuracy, environmental factors, scanning angle, etc., there may be noise, missing points or accuracy deviations. Randomly selecting some points cannot guarantee the alignment effect.

[0005] To this end, it is necessary to comprehensively consider the characteristics of the component itself and adaptively select the points used for registration to improve the registration accuracy.

[0006] Therefore, a new solution to the above problems needs to be proposed. Summary of the Invention

[0007] The purpose of the present invention is to provide a BIM and point cloud registration method with adaptive graph frequency, which does not require manually set feature point selection parameters, automatically calculates the graph cutoff frequency according to the point cloud, and adaptively generates matching feature points, thereby improving the robustness of point cloud and BIM matching, so as to solve the technical problems raised in the background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a frequency-adaptive BIM and point cloud registration method, comprising at least the following steps:

[0009] S1: Triangulate the BIM component and retain the triangulated vertices as the BIM point cloud; scan the real component corresponding to the BIM and record the scanned point cloud as the component point cloud;

[0010] S2: Calculate the cumulative distribution of the component point cloud and BIM point cloud respectively using the corresponding method;

[0011] S3: Calculate the feature points of component point cloud and BIM point cloud respectively using corresponding methods;

[0012] S4: Based on the retained feature points, the ICP algorithm is used to align the BIM point cloud and the component point cloud.

[0013] Furthermore, the step S2 includes at least the following steps:

[0014] S2.1: Calculate point cloud The corresponding undirected weighted graph

[0015] Where: V is the vertex set of point cloud X, ε is the graph The edge set of is constructed by the nearest neighbor algorithm, is the weight matrix, point x i With neighbors x j The weight W i，j equal:

[0016]

[0017] where δ i is x i The indexes of the nearest M neighbors in X;

[0018] S2.2: Calculate the graph Laplacian matrix A = D -1 W;

[0019] Where: D is a diagonal matrix, and the diagonal elements satisfy:

[0020] Use the power iteration method to calculate the largest eigenvalue λ of A N and the minimum eigenvalue λ1;

[0021] Take the linear partition α1=λ1=0,…,α p ,…,α P =λ N , use the window function to estimate the interval [0, α p ], refer to the following formula (1):

[0022]

[0023] in:

[0024]

[0025] z j ∈ is a standard normal distribution vector, Tr(·) represents the trace operation;

[0026] The last approximation in the above formula means that the trace of A can be estimated using the quadratic expectation of the multivariate standard normal distribution vector z:

[0027]

[0028] represents the expectation operation;

[0029] The window function in formula (1) First, the eigenvalue λ∈[0,αp] is Linearly scale to [-1, 1], and then use K-order Chebyshev polynomials to approximate the window function:

[0030]

[0031] in is the Chebyshev polynomial,

[0032] Coefficient c k Given by:

[0033]

[0034] Compensation factor g k Formula (2):

[0035]

[0036] S2.3: Record the frequency separation points and the number of normalized feature points as a coordinate series p∈[0, P], perform cubic interpolation on the coordinate series to obtain the cumulative distribution of the point cloud.

[0037] Furthermore, the step S3 at least includes the following steps:

[0038] S3.1: Perform differential operation on the cumulative frequency distribution of the point cloud to calculate the probability density of the spectrum P(λ);

[0039] S3.2: In the frequency range of the spectrum The cutoff feature λ is found according to the following heuristic algorithm c , where λ N is the maximum eigenvalue of A; λ1 is the minimum eigenvalue λ1 of A;

[0040] Assumptions The corresponding graph probability densities are and P(λ N ),

[0041] when If there is a trough in the , the eigenvalue corresponding to the trough is taken as λ c ;

[0042] when There is no trough in is the cutoff frequency λ c ;

[0043] Design the following high-pass filter:

[0044]

[0045] The interval [λ c ,λ N]After linear transformation Transformed to [-1, 1], the point cloud X passes through the high-pass filter h H (λ) is transformed into:

[0046]

[0047] Where A is the Laplace matrix, is a Chebyshev polynomial:

[0048]

[0049] c k are the coefficients of the polynomial approximation:

[0050]

[0051] Compensation factor g k For the value, refer to formula (2);

[0052] Solve the following optimization problem to obtain the feature points, refer to formula (3):

[0053]

[0054] in, It is a diagonal matrix, and the diagonal elements represent the sampling probability of each point;

[0055] Solve equation (3) and retain a predetermined proportion of points as feature points.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] The present invention does not require manually set feature point selection parameters, automatically calculates the atlas cutoff frequency according to the point cloud, and adaptively generates feature points for matching, thereby improving the robustness of point cloud and BIM matching. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0059] Figure 1 This is a schematic diagram of obtaining the point cloud of a component of the present invention;

[0060] Figure 2 Schematic diagram of the cumulative distribution of the point cloud of the present invention;

[0061] Figure 3 This is a schematic diagram of the probability density of the point cloud map of the present invention;

[0062] Figure 4 This is a schematic diagram of the point cloud cutoff frequency of the present invention;

[0063] Figure 5 This is a schematic diagram of the final registration result of the present invention. DETAILED DESCRIPTION

[0064] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0065] See Figure 1-Figure 5 A frequency-adaptive BIM and point cloud registration method includes at least the following steps:

[0066] S1: Triangulate the BIM component and retain the triangulated vertices as the BIM point cloud; scan the real component corresponding to the BIM and record the scanned point cloud as the component point cloud;

[0067] S2: Use corresponding methods to calculate the cumulative atlas density distribution of component point cloud and BIM point cloud respectively;

[0068] S2.1: Calculate point cloud The corresponding undirected weighted graph

[0069] Where: V is the vertex set of point cloud X, ε is the graph The edge set of is constructed by the nearest neighbor algorithm, is the weight matrix, point x i With neighbors x j The weight W i，j equal:

[0070]

[0071] where δ i For X i The indexes of the nearest M neighbors in X;

[0072] S2.2: Calculate the graph Laplacian matrix A = D -1 W;

[0073] Where: D is a diagonal matrix, and the diagonal elements satisfy:

[0074] Use the power iteration method to calculate the largest eigenvalue λ of A N and the minimum eigenvalue λ1;

[0075] Take the linear partition α1=λ1=0,…,α p,…,α P =λ N , use the window function to estimate the interval [0, α p ], refer to the following formula (1):

[0076]

[0077] in:

[0078]

[0079] z j ∈ is a standard normal distribution vector, Tr(·) represents the trace operation;

[0080] The last approximation in the above formula means that the trace of A can be estimated using the quadratic expectation of the multivariate standard normal distribution vector z:

[0081]

[0082] represents the expectation operation;

[0083] The window function in formula (1) First, the eigenvalue λ∈[0,αp] is Linearly scale to [-1, 1], and then use K-order Chebyshev polynomials to approximate the window function:

[0084]

[0085] in is the Chebyshev polynomial,

[0086] Coefficient c k Given by:

[0087]

[0088] Compensation factor g k Formula (2):

[0089]

[0090] S2.3: Record the frequency separation points and the number of normalized feature points as a coordinate series p∈[0, P], perform cubic interpolation on the coordinate series to obtain the cumulative distribution of the point cloud.

[0091] S3: Calculate the feature points of component point cloud and BIM point cloud respectively using corresponding methods;

[0092] S3.1: Perform differential operation on the cumulative distribution of the point cloud to calculate the probability density of the point cloud P(λ).

[0093] S3.2: In the frequency range of the spectrum The cutoff feature λ is found according to the following heuristic algorithm c , where λ N is the maximum eigenvalue of A and λ1 is the minimum eigenvalue of A;

[0094] Assumptions The corresponding graph probability densities are and P(λ N ),

[0095] when If there is a trough in the , the eigenvalue corresponding to the trough is taken as λ c ;

[0096] when There is no trough in is the cutoff frequency λ c ;

[0097] Design the following high-pass filter:

[0098]

[0099] The interval [λ c ,λ N ]After linear transformation Transformed to [-1, 1], the point cloud X passes through the high-pass filter h H (λ) is transformed into:

[0100]

[0101] Where A is the Laplace matrix, is a Chebyshev polynomial:

[0102]

[0103] c k are the coefficients of the polynomial approximation:

[0104]

[0105] Compensation factor g k For the value, refer to formula (2);

[0106] Solve the following optimization problem to obtain the feature points, refer to formula (3):

[0107]

[0108] in, It is a diagonal matrix, and the diagonal elements represent the sampling probability of each point;

[0109] Solve equation (3) and retain a predetermined proportion of points as feature points.

[0110] S4: Based on the retained feature points, the ICP algorithm is used to align the BIM point cloud and the component point cloud.

[0111] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A frequency-adaptive BIM and point cloud registration method, characterized by: At least the following steps are included: S1: Triangulate the BIM component and retain the triangulated vertices as the BIM point cloud; scan the real component corresponding to the BIM and record the scanned point cloud as the component point cloud; S2: Calculate the cumulative distribution of the component point cloud and BIM point cloud respectively using the corresponding method; Said S2 at least comprises the following steps: S2.1: Calculate point cloud The corresponding undirected weighted graph in: is the vertex set of point cloud X, ε is the graph The edge set of is constructed by the nearest neighbor algorithm, is the weight matrix, point x i With neighbors x j The weight W i,j equal: where δ i is x i The indexes of the nearest M neighbors in X; S2.2: Computational graph translation operator A=D -1 W; Where: D is a diagonal matrix, and the diagonal elements satisfy: Use the eigenvalue power iteration algorithm to calculate the maximum eigenvalue of A into λ N and the minimum eigenvalue into λ1; Take the linear partition α1=λ1=0,…,α p ,…,α P =λ N , use the window function to estimate the interval [0,α p ], refer to the following formula (1): in: z j ∈ is a standard normal distribution vector, Tr(·) represents the trace operation; The last approximation in the above formula means that the trace of A can be estimated using the quadratic expectation of the multivariate standard normal distribution vector z: represents the expectation operation; The window function in formula (1) First, the eigenvalue λ∈[0,α p ]according to Linearly scale to [-1,1], and then use K-order Chebyshev polynomials to approximate the window function: in is the Chebyshev polynomial, Coefficient c k Given by: Compensation factor g k Formula (2): S2.3: Record the frequency separation points and the number of normalized feature points as a coordinate series Perform cubic interpolation on the coordinate series to obtain the cumulative distribution of the point cloud; S3: Calculate the feature points of component point cloud and BIM point cloud respectively using corresponding methods; S4: Based on the retained feature points, the ICP algorithm is used to align the BIM point cloud and the component point cloud.

2. The image frequency adaptive BIM and point cloud registration method according to claim 1, characterized in that: The S3 at least includes the following steps: S3.1: Perform differential operation on the cumulative frequency distribution of the point cloud to calculate the probability density of the spectrum P(λ); S3.2: In the frequency range of the spectrum The cutoff feature λ is found according to the following heuristic algorithm c , where: N is the maximum eigenvalue of A; λ1 is the minimum eigenvalue λ1 of A; Assumptions The corresponding graph probability densities are and P(λ N ), when If there is a trough in the , the eigenvalue corresponding to the trough is taken as λ c ; when There is no trough in is the cutoff frequency λ c ; Design the following high-pass filter: The interval [λ c ,λ N ]After linear transformation Transformed to [-1,1], the point cloud X passes through the high-pass filter h H (λ) is transformed into: Where A is the Laplace matrix, is a Chebyshev polynomial: c k are the coefficients of the polynomial approximation: Compensation factor g k For the value, refer to formula (2); Solve the following optimization problem to obtain the feature points, refer to formula (3): in, It is a diagonal matrix, and the diagonal elements represent the sampling probability of each point; Solve equation (3) and retain a predetermined proportion of points as feature points.

Images