Construction component quality detection method and device based on three-dimensional point cloud and BIM

Abstract

The present application provides a method and device for quality inspection of construction components based on three-dimensional point cloud and BIM. After obtaining the three-dimensional point cloud data set of the construction scene and the BIM model of the construction target, the information in the BIM model is converted to obtain the BIM three-dimensional point cloud data set; then, the three-dimensional point cloud data set of the construction scene and the BIM three-dimensional point cloud data set are registered to obtain the registration result; the target three-dimensional point cloud data set is input into the first semantic segmentation model, and the three-dimensional point cloud model set of the construction scene is output; finally, the quality inspection result of the construction component is generated according to the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the construction component corresponding to the BIM three-dimensional point cloud data set. The quality inspection of the construction components is realized, and the objectivity and accuracy of the quality inspection of the engineering construction are improved.

Description

Technical Field

[0001] The present application relates to the technical field of quality inspection, and in particular to a method and device for inspecting the quality of construction components based on three-dimensional point cloud and BIM. Background Art

[0002] In order to ensure the quality of construction during construction, it is necessary to strictly control the quality of each component on the construction site, especially the hidden components. However, there are a large number of components being constructed simultaneously on the construction site of a large project, and it is difficult for on-site quality inspectors to accurately detect the composition and geometric parameters of each component in real time, which can easily cause hidden dangers in project quality.

[0003] At present, accurate detection of the composition and geometric parameters of each component (such as the composition, size and relative spatial position of the component) mainly relies on manual inspection. Manual inspection is the most primitive traditional construction quality inspection method. Its main operation mode is that the quality inspector or engineer directly observes or uses tools such as rulers, total stations, laser rangefinders, etc. to conduct manual measurement and observation, and relies on the construction site engineering design drawings for judgment. Although the manual inspection method has certain operational flexibility when implemented, it still has many limitations and shortcomings. First, manual inspection is only applicable to simple construction sites, and it is difficult to use manual inspection in environments such as towering buildings or complex construction sites; secondly, for large-scale engineering construction sites with a large number of simultaneous construction components, it is difficult for on-site quality inspectors to accurately detect the composition and geometric parameters of each component at the same time, which has defects such as poor real-time performance, low efficiency, and strong subjectivity, which is easy to affect the construction period; for large-scale construction quality inspection, when the workload of the staff is too large, it is easy to cause fatigue, and the safety of the inspectors cannot be guaranteed. At the same time, the poor traceability of manual inspections can easily cause engineering quality and safety hazards. Summary of the invention

[0004] In view of this, the present application provides a construction component quality inspection method and device based on three-dimensional point cloud and BIM, so as to realize the quality inspection of construction components and improve the objectivity and accuracy of engineering construction quality inspection.

[0005] The first aspect of the present application provides a construction component quality detection method based on three-dimensional point cloud and BIM, comprising:

[0006] Obtain the 3D point cloud dataset of the construction scene and the BIM model of the construction target;

[0007] Converting the information in the BIM model to obtain a BIM three-dimensional point cloud dataset; wherein the number of points in the BIM three-dimensional point cloud dataset is in the same order of magnitude as the number of points in the three-dimensional point cloud dataset of the construction scene;

[0008] The three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM are registered to obtain a registration result; wherein the registration result includes a target three-dimensional point cloud dataset; the target three-dimensional point cloud dataset is a three-dimensional point cloud dataset obtained after the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM are registered; the target three-dimensional point cloud dataset is paired with points in the BIM three-dimensional point cloud dataset one by one;

[0009] The target three-dimensional point cloud data set is input into the first semantic segmentation model, and a three-dimensional point cloud model set of the construction scene is output; wherein the three-dimensional point cloud model set of the construction scene includes a three-dimensional point cloud model of each construction component in the construction scene and semantic information corresponding to the construction component; the first semantic segmentation model is obtained by training a deep learning network model with a first training sample set; the first training sample set includes at least one training sample; the training sample includes training sample three-dimensional point cloud data and a three-dimensional point cloud model of a construction component corresponding to the training sample three-dimensional point cloud data and real semantic information corresponding to the construction component;

[0010] A quality inspection result of the construction component is generated according to the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the corresponding construction component in the BIM three-dimensional point cloud data set.

[0011] Optionally, before registering the three-dimensional point cloud dataset of the construction scene and the BIM three-dimensional point cloud dataset to obtain a registration result, the method further includes:

[0012] For each point in the three-dimensional point cloud data set of the construction scene, calculate the average distance between the point and all nearby points of the point;

[0013] If the average distance is greater than the standard range threshold, the point is confirmed to be an outlier and is removed from the three-dimensional point cloud data set of the construction scene.

[0014] Optionally, registering the three-dimensional point cloud dataset of the construction scene and the BIM three-dimensional point cloud dataset to obtain a registration result includes:

[0015] The three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM are roughly aligned to obtain a pending three-dimensional point cloud dataset; wherein the pending three-dimensional point cloud dataset is a three-dimensional point cloud dataset obtained after the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM are roughly aligned;

[0016] The pending three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset are precisely aligned to obtain an alignment result.

[0017] Optionally, the coarsely registering the three-dimensional point cloud dataset of the construction scene with the BIM three-dimensional point cloud dataset to obtain the pending three-dimensional point cloud dataset includes:

[0018] In the BIM three-dimensional point cloud data set, N random points are randomly selected; wherein N is greater than or equal to 3;

[0019] In the three-dimensional point cloud data set of the construction scene, a first assumed point is randomly selected; wherein the first assumed point is used as a corresponding point of the first random point;

[0020] Calculate the Euclidean distance between the first random point and the second random point to obtain the first target distance;

[0021] In the three-dimensional point cloud data set of the construction scene, find a point whose distance from the first assumed point is the first target distance, and randomly select one as a second assumed point; wherein the second assumed point serves as a corresponding point of the second random point;

[0022] Calculate the Euclidean distance between the Nth random point and the N-2th random point to obtain the 2N-4th target distance;

[0023] Calculate the Euclidean distance between the Nth random point and the N-1th random point to obtain the 2N-3th target distance;

[0024] In the three-dimensional point cloud data set of the construction scene, find a point whose distance from the N-2th assumed point is the 2N-4th target distance and whose distance from the N-1th assumed point is the 2N-3th target distance, and randomly select one of them as the Nth assumed point; wherein the Nth assumed point serves as the corresponding point of the Nth random point;

[0025] Generate a first rigid body transformation matrix according to the assumed point set and the random point set; wherein the first rigid body transformation matrix includes a first rotation matrix and a first displacement vector;

[0026] Using the first rigid body transformation matrix, a rigid body transformation is performed on the three-dimensional point cloud dataset of the construction scene to obtain a transformed three-dimensional point cloud dataset of the construction scene;

[0027] For each remaining point in the transformed three-dimensional point cloud data set of the construction scene, calculate the Euclidean distance between the remaining point and the corresponding point in the BIM three-dimensional point cloud data set; wherein the remaining point is the point remaining after removing all the assumed points;

[0028] If the Euclidean distance between the remaining point and the corresponding point in the BIM three-dimensional point cloud dataset is less than the error threshold, the remaining point and the corresponding point in the BIM three-dimensional point cloud dataset are taken as sample inliers, and the process returns to the step of randomly selecting N random points in the BIM three-dimensional point cloud dataset;

[0029] When the preset number of iterations is reached, the 3D point cloud dataset of the construction scene with the largest number of points in the sample is selected as the pending 3D point cloud dataset.

[0030] Optionally, the accurately registering the undetermined three-dimensional point cloud dataset with the BIM three-dimensional point cloud dataset to obtain a registration result includes:

[0031] Performing preliminary matching on the pending three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset to obtain a preliminary matched pending three-dimensional point cloud dataset and a preliminary matched BIM three-dimensional point cloud dataset;

[0032] Generate a second rigid body transformation matrix according to the initially matched pending three-dimensional point cloud dataset and the initially matched BIM three-dimensional point cloud dataset; wherein the second rigid body transformation matrix includes a second rotation matrix and a second displacement vector;

[0033] Using the second rigid body transformation matrix, performing a rigid body transformation on the pending three-dimensional point cloud dataset to obtain a transformed pending three-dimensional point cloud dataset;

[0034] Calculate the average Euclidean distance between the transformed undetermined three-dimensional point cloud dataset and all points in the BIM three-dimensional point cloud dataset;

[0035] If the average Euclidean distance between the transformed pending 3D point cloud dataset and all points in the BIM 3D point cloud dataset is greater than the average Euclidean distance threshold, the transformed pending 3D point cloud dataset is used as a new pending 3D point cloud dataset, and the step of performing preliminary matching between the pending 3D point cloud dataset and the BIM 3D point cloud dataset is performed to obtain the preliminary matched pending 3D point cloud dataset and the preliminary matched BIM 3D point cloud dataset.

[0036] When the average Euclidean distance between the transformed pending three-dimensional point cloud dataset and all points in the BIM three-dimensional point cloud dataset reaches a preset convergence condition, the transformed pending three-dimensional point cloud dataset is used as the target three-dimensional point cloud dataset.

[0037] Optionally, generating a first rigid body transformation matrix according to the assumed point set and the random point set includes:

[0038] The centroid of the assumed point set and the centroid of the random point set are calculated respectively;

[0039] According to the centroid of the assumed point set and the centroid of the random point set, the assumed point set and the random point set are de-centroided to obtain a de-centroided assumed point set and a de-centroided random point set;

[0040] A first covariance matrix is ​​calculated based on the de-centred assumed point set, the de-centred random point set and the first weight set;

[0041] The first covariance matrix is ​​solved by a singular value decomposition method to obtain an optimal rotation matrix and an optimal translation matrix between the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM;

[0042] A first rigid body transformation matrix is ​​determined according to an optimal rotation matrix and an optimal translation matrix between the three-dimensional point cloud dataset of the construction scene and the BIM three-dimensional point cloud dataset.

[0043] Optionally, generating a second rigid body transformation matrix according to the initially matched pending three-dimensional point cloud dataset and the initially matched BIM three-dimensional point cloud dataset includes:

[0044] The centroid of the undetermined 3D point cloud dataset after the initial matching and the centroid of the BIM 3D point cloud dataset after the initial matching are calculated respectively;

[0045] According to the centroid of the initially matched pending 3D point cloud dataset and the centroid of the initially matched BIM 3D point cloud dataset, de-centroiding the initially matched pending 3D point cloud dataset and the initially matched BIM 3D point cloud dataset to obtain a de-centroided initially matched pending 3D point cloud dataset and a de-centroided initially matched BIM 3D point cloud dataset;

[0046] A second covariance matrix is ​​obtained by calculating the undetermined three-dimensional point cloud dataset after the initial matching without centroid, the BIM three-dimensional point cloud dataset after the initial matching without centroid, and the second weight set;

[0047] The second covariance matrix is ​​solved by a singular value decomposition method to obtain an optimal rotation matrix and an optimal translation matrix between the undetermined three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset;

[0048] A second rigid body transformation matrix is ​​determined according to an optimal rotation matrix and an optimal translation matrix between the pending three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset.

[0049] Optionally, after inputting the target three-dimensional point cloud data set into the first semantic segmentation model to output a three-dimensional point cloud model set of the construction scene, the method further includes:

[0050] Acquire construction scene images;

[0051] The construction scene image is input into a second semantic segmentation model, and the semantic segmentation result of the construction scene image is obtained as an output; wherein the second semantic segmentation model is obtained by training a deep learning neural network with a training sample set; the training sample set includes at least one training sample; the training sample includes: a training sample image and a real semantic segmentation result corresponding to the training sample image;

[0052] The three-dimensional point cloud model set of the construction scene is verified using the semantic segmentation result of the construction scene image to obtain a verification result.

[0053] Optionally, generating the quality inspection result of the construction component according to the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the construction component corresponding to the three-dimensional point cloud model in the BIM three-dimensional point cloud data set includes:

[0054] The root mean square deviation of the construction component is calculated based on the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the corresponding construction component in the BIM three-dimensional point cloud data set;

[0055] According to the semantic information corresponding to the construction component, the type of the construction component is determined, and the target threshold is determined according to the type of the construction component; wherein, if the root mean square deviation of the construction component is greater than the target threshold, the construction component is determined to be unqualified; if the root mean square deviation of the construction component is not greater than the target threshold, the construction component is determined to be qualified.

[0056] The second aspect of the present application provides a construction component quality detection device based on three-dimensional point cloud and BIM, comprising:

[0057] A first acquisition unit is used to acquire a three-dimensional point cloud data set of a construction scene and a BIM model of a construction target;

[0058] A first conversion unit is used to convert the information in the BIM model to obtain a BIM three-dimensional point cloud data set; wherein the number of points in the BIM three-dimensional point cloud data set is in the same order of magnitude as the number of points in the three-dimensional point cloud data set of the construction scene;

[0059] A registration unit is used to register the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM to obtain a registration result; wherein the registration result includes a target three-dimensional point cloud dataset; the target three-dimensional point cloud dataset is a three-dimensional point cloud dataset obtained after registering the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM; the target three-dimensional point cloud dataset is paired with points in the BIM three-dimensional point cloud dataset one by one;

[0060] A first input unit is used to input the target three-dimensional point cloud data set into a first semantic segmentation model, and output a three-dimensional point cloud model set of the construction scene; wherein the three-dimensional point cloud model set of the construction scene includes a three-dimensional point cloud model of each construction component in the construction scene and semantic information corresponding to the construction component; the first semantic segmentation model is obtained by training a deep learning network model with a first training sample set; the first training sample set includes at least one training sample; the training sample includes training sample three-dimensional point cloud data and a three-dimensional point cloud model of a construction component corresponding to the training sample three-dimensional point cloud data, and real semantic information corresponding to the construction component;

[0061] The quality inspection unit is used to generate quality inspection results of the construction components according to the three-dimensional point cloud model of the construction components in the construction scene and the three-dimensional point cloud model of the corresponding construction components in the BIM three-dimensional point cloud data set.

[0062] A third aspect of the present application provides an electronic device, including:

[0063] one or more processors;

[0064] a storage device having one or more programs stored thereon;

[0065] When the one or more programs are executed by the one or more processors, the one or more processors implement the construction component quality detection method based on three-dimensional point cloud and BIM as described in any one of the first aspects.

[0066] A fourth aspect of the present application provides a computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the construction component quality inspection method based on three-dimensional point cloud and BIM as described in any one of the first aspects is implemented.

[0067] As can be seen from the above scheme, the present application provides a method and device for quality inspection of construction components based on three-dimensional point cloud and BIM. After obtaining the three-dimensional point cloud data set of the construction scene and the BIM model of the construction target, the information in the BIM model is converted to obtain the BIM three-dimensional point cloud data set; wherein the number of points in the BIM three-dimensional point cloud data set is in the same order of magnitude as the number of points in the three-dimensional point cloud data set of the construction scene; then, the three-dimensional point cloud data set of the construction scene and the BIM three-dimensional point cloud data set are aligned to obtain the alignment result; the target three-dimensional point cloud data set is input into the first semantic segmentation model, and the three-dimensional point cloud model set of the construction scene is output; finally, the quality inspection result of the construction component is generated according to the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the corresponding construction component in the BIM three-dimensional point cloud data set. The quality inspection of the construction components is realized, and the objectivity and accuracy of the quality inspection of the engineering construction are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0069] Figure 1 A specific flow chart of a construction component quality detection method based on three-dimensional point cloud and BIM provided in an embodiment of the present application;

[0070] Figure 2 A flowchart for registering a three-dimensional point cloud model and a BIM point cloud model of a construction component is provided in another embodiment of the present application;

[0071] Figure 3 A flowchart for roughly aligning a BIM point cloud model with a three-dimensional point cloud model of a construction component is provided in another embodiment of the present application;

[0072] Figure 4 A flowchart of a method for generating a first rigid body transformation matrix provided in another embodiment of the present application;

[0073] Figure 5 A flowchart for accurately aligning a three-dimensional point cloud model of a target construction component with a BIM point cloud model provided in another embodiment of the present application;

[0074] Figure 6 A flowchart of a method for generating a second rigid body transformation matrix provided in another embodiment of the present application;

[0075] Figure 7 A flowchart of a method for generating quality inspection results of construction components provided in another embodiment of the present application;

[0076] Figure 8 A schematic diagram of a construction component quality inspection device based on three-dimensional point cloud and BIM provided in another embodiment of the present application. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0078] The term "including" and its variations used herein are open inclusions, i.e., "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0079] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0080] It should be noted that the concepts such as "first" and "second" mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0081] It should be noted that the modifications of "one" and "plurality" mentioned in the present application are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0082] The present application embodiment provides a construction component quality detection method based on three-dimensional point cloud and BIM, such as Figure 1 As shown, the specific steps include:

[0083] S101, obtaining a three-dimensional point cloud data set of a construction scene and a BIM model of a construction target.

[0084] Among them, BIM refers to Building Information Modeling.

[0085] In the specific implementation process of the present application, the construction scene information can be collected by instruments such as but not limited to three-dimensional scanners to obtain three-dimensional point cloud data, which is not limited here.

[0086] S102: Convert the information in the BIM model to obtain a BIM three-dimensional point cloud data set.

[0087] Among them, the number of points in the BIM 3D point cloud dataset is at the same order of magnitude as the number of points in the 3D point cloud dataset of the construction scene.

[0088] It should be noted that when matching the real physical space with the virtual model space in complex scenarios, the BIM model of the construction target and the three-dimensional point cloud data of the construction site collected by the three-dimensional scanner are two data objects with different sources and formats. Therefore, it is necessary to convert the spatial and semantic information of different construction components in the BIM model into an object format that can be operated and saved.

[0089] In the specific implementation process of this application, the coordinates of the construction components in the BIM model can be extracted and converted into corresponding point cloud images using but not limited to Revit software and Python, while ensuring that the number of points in the extracted BIM point cloud model is at the same order of magnitude as the number of points in the three-dimensional point cloud model of the construction components. At the same time, the semantic information corresponding to the construction components in the BIM model is extracted to prepare for the subsequent comparison of the BIM model with the three-dimensional point cloud model.

[0090] Specifically, first use software such as Revit to export the completed building BIM model, which contains the types, materials, and geometric information of building components. Use Python database to extract the geometric semantic information of the corresponding building construction components from the BIM model, mainly extract key nodes, component surfaces, geometric contours and other parts (such as: the size and type information of columns, beams, and plates), and then create a point cloud coordinate system for the BIM model, and select a suitable three-dimensional coordinate origin in the point cloud three-dimensional space coordinate system. In the created three-dimensional coordinate system, the extracted BIM data is converted into points with corresponding three-dimensional coordinates (converted into XYZ coordinates in this coordinate system), and finally use relevant point cloud processing tools (such as Open3D, etc.) to open the BIM three-dimensional point cloud data set and perform subsequent operations.

[0091] During the specific implementation of this application, due to the complexity of the construction site, which may be affected by interference factors such as obstruction by debris and changes in perspective, the identification and semantic segmentation of construction components in three-dimensional point cloud data may result in information loss, blurred boundaries and other problems, resulting in deviations between the identification features of the composition and geometric parameters of the construction components and the corresponding design values.

[0092] Therefore, in another embodiment of the present application, before executing step S102, an implementation of the construction component quality detection method based on three-dimensional point cloud and BIM further includes:

[0093] For each point in the 3D point cloud dataset of the construction scene, the average distance between the point and all nearby points is calculated; if the average distance is greater than the standard range threshold, the point is confirmed as an outlier and removed from the 3D point cloud dataset of the construction scene. The standard range threshold is set according to the type of construction components, in accordance with relevant specifications, and combined with human experience, and is not limited here.

[0094] This removes noise or useless points other than workers, debris and other non-under-construction components at the construction site.

[0095] S103, registering the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM to obtain a registration result.

[0096] Among them, the registration result includes a target 3D point cloud dataset; the target 3D point cloud dataset is a 3D point cloud dataset obtained after the 3D point cloud dataset of the construction scene and the BIM 3D point cloud dataset are registered; the target 3D point cloud dataset is paired one by one with the points in the BIM 3D point cloud dataset.

[0097] Optionally, in another embodiment of the present application, an implementation of step S103 is as follows: Figure 2 As shown, including:

[0098] S201 . Roughly align the 3D point cloud dataset of the construction scene and the BIM 3D point cloud dataset to obtain a pending 3D point cloud dataset.

[0099] The undetermined three-dimensional point cloud dataset is a three-dimensional point cloud dataset obtained after the three-dimensional point cloud dataset of the construction scene and the BIM three-dimensional point cloud dataset are roughly aligned.

[0100] It should be noted that the rough registration of the 3D point cloud dataset of the construction scene and the 3D point cloud dataset of the BIM means that when the initial state of the point cloud dataset is unknown, a relatively simple and fast transformation is used to make the two as aligned and overlapped as possible in space, in preparation for the subsequent fine registration.

[0101] Optionally, in another embodiment of the present application, an implementation of step S201 is as follows: Figure 3 As shown, including:

[0102] S301. Randomly select N random points in a BIM three-dimensional point cloud data set.

[0103] Wherein, N is greater than or equal to 3.

[0104] It is understandable that when performing rough alignment, at least 3 pairs of points must be selected from at least two point clouds (the BIM 3D point cloud dataset and the 3D point cloud dataset of the construction scene in this application) to solve the stiffness transformation matrix. Usually, in order to reduce the impact of noise and increase robustness, the number of point pairs selected is often more than 3, which is generally determined by the number of point cloud models.

[0105] In the specific implementation process of the present application, the BIM point cloud model can be set as the original point cloud, that is, the point cloud to be matched, and the three-dimensional point cloud model of the construction component can be set as the target point cloud.

[0106] First, in the BIM point cloud model, randomly select the first random point , the second random point , the third random point ...Nth random point .

[0107] In the specific implementation process of this application, restrictions can also be added to these N random points. For example, these three points cannot be collinear, and there are corresponding requirements for the distance between these N points. When the distance is too far or too close, they need to be discarded. The selected points should be distributed in different areas as much as possible to better represent the overall point cloud. There is no limitation here.

[0108] S302: randomly select a first hypothetical point in the three-dimensional point cloud data set of the construction scene.

[0109] The first assumed point is used as the corresponding point of the first random point.

[0110] Continuing with the above example, in the 3D point cloud model of the construction component, the first hypothetical point is randomly selected As The corresponding point of .

[0111] S303: Calculate the Euclidean distance between the first random point and the second random point to obtain a first target distance.

[0112] Continuing with the above example, calculate and The Euclidean distance between , that is, the first target distance.

[0113] S304. In the three-dimensional point cloud data set of the construction scene, find a point whose distance from the first assumed point is the first target distance, and randomly select one of them as the second assumed point.

[0114] The second assumed point serves as the corresponding point of the second random point.

[0115] Continuing with the above example, find the corresponding The Euclidean distance is There may be multiple points, randomly select one as The corresponding point , which is the second assumption point.

[0116] S305 , calculating the Euclidean distance between the Nth random point and the N-2th random point to obtain the 2N-4th target distance.

[0117] Continuing with the above example, calculate and The Euclidean distance between them is the 2N-4th target distance.

[0118] Take N=3 as an example, calculate and The Euclidean distance between , which is the second target distance.

[0119] S306, calculating the Euclidean distance between the Nth random point and the N-1th random point to obtain the 2N-3th target distance.

[0120] Continuing with the above example, calculate and The Euclidean distance between them is the 2N-3th target distance.

[0121] Take N=3 as an example, calculate and The Euclidean distance between , that is, the third target distance.

[0122] S307. In the three-dimensional point cloud data set of the construction scene, find a point whose distance to the N-2th assumed point is the 2N-4th target distance and whose distance to the N-1th assumed point is the 2N-3th target distance, and randomly select one of them as the Nth assumed point.

[0123] The third assumed point serves as the corresponding point of the third random point.

[0124] Continuing with the above example (taking N=3 as an example), find the points in the 3D point cloud model of the construction component that match Distance is ,and Distance is There may be multiple points, randomly select one as the third hypothetical point .

[0125] In the specific implementation process of the present application, restrictions may also be imposed on two triangles consisting of three pairs of points selected from the source point cloud and the target point cloud. For example, the length difference of the corresponding sides of the triangles must be within a certain threshold.

[0126] S308. Generate a first rigid body transformation matrix according to the assumed point set and the random point set.

[0127] The first rigid body transformation matrix includes a first rotation matrix and a first displacement vector.

[0128] Since the process of registering the 3D point cloud dataset of the construction scene to the 3D point cloud dataset of the BIM is a rigid body transformation, that is, the 3D point cloud dataset of the construction scene only undergoes a certain rotation and translation, and its size does not change. In the rigid body transformation process, a 3×3 matrix is ​​used to represent the rotation transformation of the point cloud, and a 3×1 translation matrix is ​​used. Represents translation transformation. For rigid body transformation, point cloud registration is actually to solve the optimal rotation matrix With the optimal translation matrix , the two together form a 4×4 transformation matrix :

[0129] .

[0130] Optionally, in another embodiment of the present application, an implementation of step S308 is as follows: Figure 4 As shown, including:

[0131] S401. Calculate the centroid of the assumed point set and the centroid of the random point set respectively.

[0132] Continuing with the above example, assume that the set of random points is , assuming that the point set is .

[0133] Then, the centroid of a set of random points is expressed as:

[0134] ;in, For the corresponding The weight of a point.

[0135] Then, assume that the centroid of the point set is expressed as:

[0136] ;in, For the corresponding The weight of a point.

[0137] S402. De-centroiding the assumed point set and the random point set according to the centroid of the assumed point set and the centroid of the random point set to obtain a de-centroided assumed point set and a de-centroided random point set.

[0138] Specifically, the assumed point set and the random point set are de-centroided to obtain:

[0139] , , ;in, is a set of random points that have been de-centred. is the set of assumed points to be de-centred.

[0140] S403 . Calculate a first covariance matrix according to the de-centred assumed point set, the de-centred random point set and the first weight set.

[0141] Continuing with the above example, the first covariance matrix S can be calculated using the following formula:

[0142] ;

[0143] in, is the first weight matrix; and Respectively and A 3×n matrix of columns.

[0144] S404: Using a singular value decomposition method, the first covariance matrix is ​​solved to obtain an optimal rotation matrix and an optimal translation matrix between the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM.

[0145] Among them, the optimal rotation matrix is: ; The optimal translation matrix is .

[0146] S405. Determine a first rigid body transformation matrix according to an optimal rotation matrix and an optimal translation matrix between the three-dimensional point cloud dataset of the construction scene and the BIM three-dimensional point cloud dataset.

[0147] S309 , using the first rigid body transformation matrix, perform a rigid body transformation on the three-dimensional point cloud dataset of the construction scene to obtain a transformed three-dimensional point cloud dataset of the construction scene.

[0148] S310 . For each remaining point in the transformed three-dimensional point cloud data set of the construction scene, calculate the Euclidean distance between the remaining point and the corresponding point in the BIM three-dimensional point cloud data set.

[0149] Among them, the remaining points are the points remaining after eliminating all assumed points.

[0150] S311. If the Euclidean distance between the remaining points and the corresponding points in the BIM three-dimensional point cloud data set is less than the error threshold, the remaining points and the corresponding points in the BIM three-dimensional point cloud data set are taken as sample inliers.

[0151] The error threshold is preset and changed by technicians, experts, etc. and is not limited here.

[0152] S312. When the preset number of iterations is reached, the three-dimensional point cloud dataset of the construction scene with the largest number of points in the sample is selected as the pending three-dimensional point cloud dataset.

[0153] The preset number of iterations is pre-set and changed by technicians, experts, etc. and is not limited here.

[0154] S202, accurately aligning the predetermined 3D point cloud dataset with the BIM 3D point cloud dataset to obtain a alignment result.

[0155] After the rough registration of the 3D point cloud dataset of the construction scene and the BIM 3D point cloud dataset, the pending 3D point cloud dataset and the BIM 3D point cloud dataset have a good initial position. The pending 3D point cloud dataset and the BIM 3D point cloud dataset are finely registered to obtain the registration result, so that the overlap area of ​​the target 3D point cloud dataset and the BIM 3D point cloud dataset is the largest. The registration objective function between the two point cloud datasets is selected as:

[0156] ;

[0157] in, is the weight of each point, It refers to the rotation matrix With displacement vector Find the target distance function between two point clouds.

[0158] The purpose of the fine registration operation is to find the most suitable rotation matrix With displacement vector The objective function is minimized and the overlapping area between the target 3D point cloud dataset and the BIM 3D point cloud dataset is maximized.

[0159] Optionally, in another embodiment of the present application, an implementation of step S202 is as follows: Figure 5 As shown, including:

[0160] S501 , preliminarily matching the pending 3D point cloud dataset with the BIM 3D point cloud dataset to obtain the preliminarily matched pending 3D point cloud dataset and the preliminarily matched BIM 3D point cloud dataset.

[0161] In the specific implementation process of this application, the BIM 3D point cloud dataset can be found and pending 3D point cloud datasets The nearest point of each point in is taken as the first matching point pair. When performing the initial matching, the Kd-Tree method can be used but not limited to realize the nearest neighbor search of two point cloud models, which is not limited here.

[0162] S502: Generate a second rigid body transformation matrix according to the initially matched pending three-dimensional point cloud dataset and the initially matched BIM three-dimensional point cloud dataset.

[0163] The second rigid body transformation matrix includes a second rotation matrix and a second displacement vector.

[0164] Optionally, in another embodiment of the present application, an implementation of step S502 is as follows: Figure 6 As shown, including:

[0165] S601, respectively calculating the centroid of the pending 3D point cloud dataset after initial matching and the centroid of the BIM 3D point cloud dataset after initial matching.

[0166] S602. De-centroiding the initially matched pending three-dimensional point cloud dataset and the initially matched BIM three-dimensional point cloud dataset according to the centroid of the initially matched pending three-dimensional point cloud dataset and the centroid of the initially matched BIM three-dimensional point cloud dataset to obtain a de-centroided initially matched pending three-dimensional point cloud dataset and a de-centroided initially matched BIM three-dimensional point cloud dataset.

[0167] S603, calculating a second covariance matrix according to the undetermined three-dimensional point cloud dataset after the initial matching without centroid, the BIM three-dimensional point cloud dataset after the initial matching without centroid, and the second weight set.

[0168] S604: Using a singular value decomposition method, the second covariance matrix is ​​solved to obtain an optimal rotation matrix and an optimal translation matrix between the undetermined three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset.

[0169] S605 . Determine a second rigid body transformation matrix according to an optimal rotation matrix and an optimal translation matrix between the pending three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset.

[0170] It should be noted that, in the specific implementation process of the present application, the specific implementation method of steps S601 to S605 can refer to the corresponding steps of steps S401 to S405, which will not be repeated here.

[0171] S503 , using the second rigid body transformation matrix, perform a rigid body transformation on the pending three-dimensional point cloud dataset to obtain a transformed pending three-dimensional point cloud dataset.

[0172] S504: Calculate the average Euclidean distance between the transformed pending 3D point cloud dataset and all points in the BIM 3D point cloud dataset.

[0173] Specifically, for each point in the transformed pending three-dimensional point cloud dataset, the Euclidean distance between the point in the transformed pending three-dimensional point cloud dataset and the corresponding point in the BIM three-dimensional point cloud dataset is calculated. After calculating the Euclidean distances of all points in the transformed pending three-dimensional point cloud dataset and the corresponding points in the BIM three-dimensional point cloud dataset, the average value is obtained to obtain the average Euclidean distance of all points in the transformed pending three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset.

[0174] S505: If the average Euclidean distance between the transformed pending 3D point cloud dataset and all points in the BIM 3D point cloud dataset is greater than the average Euclidean distance threshold, the transformed pending 3D point cloud dataset is used as a new pending 3D point cloud dataset.

[0175] S506: When the average Euclidean distance between the transformed pending 3D point cloud dataset and all points in the BIM 3D point cloud dataset reaches a preset convergence condition, the transformed pending 3D point cloud dataset is used as the target 3D point cloud dataset.

[0176] Among them, the preset convergence conditions are pre-set and changed by technicians, experts, etc. and are not limited here.

[0177] S104: Input the target three-dimensional point cloud data set into the first semantic segmentation model, and output a three-dimensional point cloud model set of the construction scene.

[0178] Among them, the three-dimensional point cloud model set of the construction scene includes the three-dimensional point cloud model of each construction component in the construction scene and the semantic information corresponding to the construction component; the first semantic segmentation model is obtained by training the deep learning network model with the first training sample set; the first training sample set includes at least one training sample; the training sample includes the training sample three-dimensional point cloud data and the three-dimensional point cloud model of the construction component corresponding to the training sample three-dimensional point cloud data and the real semantic information corresponding to the construction component.

[0179] In the specific implementation process of this application, the PointNet++ deep learning network can be used but not limited to extract point cloud features. The main operation steps include: first annotate a large amount of known construction scene point cloud data to establish a training data set suitable for the construction site; build a PointNet++ model suitable for the construction site scene, and at the same time complete the parameter adjustment and setting of the modules such as the convolution layer and the pooling layer according to the construction scene, and select a suitable loss function for the model according to the actual situation of the construction site; use the built PointNet++ model to semantically segment the target point cloud of the construction scene, and in this process, the point cloud model continuously uses the modules such as the convolution layer and the pooling layer to extract features of the overall point cloud model and finally fuse them. After a certain number of iterations, the global features of the three-dimensional point cloud model of the construction scene are finally obtained and feature recognition is completed. After completing the construction of the PointNet++ point cloud semantic segmentation model, the PointNet++ model is continuously trained using the labeled data set; at the same time, a verification data set is made to classify and predict the model, and the model parameters are continuously adjusted to make its accuracy meet the actual requirements of the construction scene, and finally the first semantic segmentation model is obtained.

[0180] Among them, the construction scene mainly includes the inspection of common components on the construction site, mainly including beams, slabs, walls, columns, etc., which are not limited here.

[0181] It should be noted that when choosing the loss function, you can choose according to the actual situation of the construction site. For example, when classifying columns, choose the cross entropy loss function. During training, the prediction result is measured by the probability distribution difference between the true value and the predicted value. The smaller the cross entropy loss function, the more accurate the prediction result. There is no limitation here.

[0182] Optionally, in another embodiment of the present application, after the target 3D point cloud dataset is input into the first semantic segmentation model in step S104 and a 3D point cloud model set of the construction scene is output, an implementation of the construction component quality detection method based on 3D point cloud and BIM further includes:

[0183] Post-processing operations are performed on the point cloud images in the three-dimensional point cloud model set of the construction scene, and algorithms such as mean filtering are used to further smooth the point cloud images to make them better conform to the actual engineering situation.

[0184] Since only a single selection of a set of 3D point cloud models of a construction scene may result in a small number of inaccurate predictions, in another embodiment of the present application, an implementation of a method for quality inspection of construction components based on 3D point cloud and BIM further includes:

[0185] Acquire a construction scene image, input the construction scene image into a second semantic segmentation model, and output a semantic segmentation result of the construction scene image; then, use the semantic segmentation result of the construction scene image to verify the three-dimensional point cloud model set of the construction scene to obtain a verification result.

[0186] Among them, the second semantic segmentation model is obtained by training a deep learning neural network with a training sample set; the training sample set includes at least one training sample; the training sample includes: a training sample image and a real semantic segmentation result corresponding to the training sample image.

[0187] In the specific implementation process of this application, semantic segmentation of construction components in construction scene images is actually a pixel-by-pixel screening of image data. Each pixel in the image of the under-construction components is labeled and classified through a deep learning neural network architecture.

[0188] Therefore, in the specific implementation process of the present application, the YOLOv8 deep learning neural network can be used but is not limited to perform semantic segmentation on the target construction scene image, which is not limited here.

[0189] The training process of the second semantic segmentation model may be, but is not limited to, collecting a large number of training sample images and the actual semantic segmentation results corresponding to the training sample images, inputting the training sample images into the YOLOv8 model, and outputting the predicted semantic segmentation results; comparing the predicted results with the actual semantic segmentation results, and adjusting the model parameters if there is a deviation until the model training is completed. When there is a deviation, a large number of data samples can be collected based on on-site construction and manually labeled, which is not limited here.

[0190] Afterwards, a corresponding verification dataset was created to verify the image semantic segmentation deep learning model, and the model parameters were continuously adjusted to make the YOLOv8 network fully adaptable to the image segmentation task of construction scenes.

[0191] In the specific implementation process of the present application, the construction scene images in the training samples can also be subjected to preprocessing operations such as cropping and scaling, thereby supplementing and expanding the data set before training, making it more robust when making predictions.

[0192] S105 , generating quality inspection results of the construction components according to the three-dimensional point cloud model of the construction components in the construction scene and the three-dimensional point cloud model of the corresponding construction components in the BIM three-dimensional point cloud data set.

[0193] Optionally, in another embodiment of the present application, before generating the quality inspection results of the construction components based on the three-dimensional point cloud model of the construction components in the construction scene and the three-dimensional point cloud model of the corresponding construction components in the BIM model, an intrinsic shape signature (ISS) extraction method can also be used to extract key points, that is, the intrinsic shape signature extraction method is used to extract key points from the three-dimensional point cloud model of the construction components in the construction scene, and the intrinsic shape signature extraction method is used to extract key points from the BIM point cloud model.

[0194] In the specific implementation process of the present application, since the points in the BIM point cloud model and the target three-dimensional point cloud dataset are already one-to-one corresponding during alignment, it is sufficient to extract only the key points of the BIM point cloud model and find corresponding points in another model, or to extract only the key points of the target three-dimensional point cloud dataset and find corresponding points in another model. No limitation is made here.

[0195] Specifically, the ISS (Intrinsic Shape Signature) key point extraction algorithm is used to extract feature points. On the basis of ensuring that a sufficient number of points are extracted, the points with obvious characteristics and representativeness in the construction component model are detected. By calculating the ISS value (ISS eigenvalue ratio) of each point and comparing it with the threshold, the required key points, that is, the points with ISS values ​​greater than the threshold, are screened out.

[0196] Optionally, in another embodiment of the present application, an implementation of step S105 is as follows: Figure 7 As shown, including:

[0197] S701. Calculate the root mean square deviation of the construction component according to the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the corresponding construction component in the BIM three-dimensional point cloud data set.

[0198] A 3D point cloud model of construction components in the construction scene , the three-dimensional coordinates of the corresponding points in the same coordinate system are , and the 3D point cloud model of the construction component corresponding to the BIM 3D point cloud dataset , the three-dimensional coordinates of the corresponding points in the same coordinate system are For example.

[0199] The root mean square deviation of the construction components is calculated based on the 3D point cloud model of the construction components in the construction scene and the 3D point cloud model of the corresponding construction components in the BIM model. The method is:

[0200] ;

[0201] S702: Determine the type of the construction component according to the semantic information corresponding to the construction component, and determine the target threshold according to the type of the construction component.

[0202] It should be noted that different target thresholds are pre-set for different types of construction components. To determine whether the composition and geometric parameters of the components meet the design requirements of the drawings.

[0203] S703: Determine the quality inspection result of the construction component according to the root mean square deviation of the construction component and the target threshold.

[0204] Among them, if the root mean square deviation of the construction component is greater than the target threshold, the construction component is determined to be unqualified; if the root mean square deviation of the construction component is not greater than the target threshold, the construction component is determined to be qualified.

[0205] Since there is still a problem of poor traceability at the current construction site, in another embodiment of the present application, before obtaining the three-dimensional point cloud data set of the construction scene and the BIM model of the construction target, an implementation method of the construction component quality detection method based on the three-dimensional point cloud and BIM also includes:

[0206] Acquire a construction scene image, and determine whether the current construction scene information has changed based on the construction scene image; if the current construction scene information has changed, start quality inspection of the construction components in the construction scene, that is, execute step S101 and subsequent steps.

[0207] In the specific implementation process of the present application, the corresponding feature points can be extracted from the image using, but not limited to, the scale-invariant feature transform (SIFT) image processing technology. The SIFT features extracted using the scale-invariant feature transform are invariant to rotation and scale, which means that the descriptors of the features of the same object are fixed. Moreover, the SIFT algorithm has a good detection effect even if the image shooting angle or image brightness is changed when the construction site environment is changeable, and has a certain applicability to the complex environment on the construction site.

[0208] The SIFT algorithm is used to extract the image feature points. Each feature point is described by a 128-dimensional vector to obtain a feature vector (floating point number). SIFT feature matching is then performed between images. The distance between feature vectors is calculated using Euclidean distance, that is, when the calculated distance is less than the set threshold, it means that the feature matching is successful and has the same feature points. By extracting and matching features of images at different times, it can be determined whether the construction scene information has changed.

[0209] Feature description operator: ;

[0210] ;

[0211] Euclidean distance: ;

[0212] Among them, des 1 With des 2 They represent the feature description operators in different images respectively, and the image features are matched by calculating the distance between different feature description operators.

[0213] It can be seen from the above scheme that the present application provides a method for quality inspection of construction components based on three-dimensional point cloud and BIM. After obtaining the three-dimensional point cloud data set of the construction scene and the BIM model of the construction target, the information in the BIM model is converted to obtain the BIM three-dimensional point cloud data set; wherein the number of points in the BIM three-dimensional point cloud data set is in the same order of magnitude as the number of points in the three-dimensional point cloud data set of the construction scene; then, the three-dimensional point cloud data set of the construction scene and the BIM three-dimensional point cloud data set are registered to obtain the registration result; the target three-dimensional point cloud data set is input into the first semantic segmentation model, and the three-dimensional point cloud model set of the construction scene is output to obtain; finally, the quality inspection result of the construction component is generated according to the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the corresponding construction component in the BIM model. The quality inspection of the construction components is realized, and the objectivity and accuracy of the quality inspection of the engineering construction are improved.

[0214] The present application embodiment provides a construction component quality detection device based on three-dimensional point cloud and BIM, such as Figure 8 As shown, specifically including:

[0215] The first acquisition unit 801 is used to acquire a three-dimensional point cloud dataset of a construction scene and a BIM model of a construction target.

[0216] The first conversion unit 802 is used to convert the information in the BIM model to obtain BIM three-dimensional point cloud data.

[0217] Among them, the number of points in the BIM 3D point cloud dataset is at the same order of magnitude as the number of points in the 3D point cloud dataset of the construction scene.

[0218] The registration unit 803 is used to register the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM to obtain a registration result.

[0219] Among them, the registration result includes a target 3D point cloud dataset; the target 3D point cloud dataset is a 3D point cloud dataset obtained after the 3D point cloud dataset of the construction scene and the BIM 3D point cloud dataset are registered; the target 3D point cloud dataset is paired one by one with the points in the BIM 3D point cloud dataset.

[0220] The first input unit 804 is used to input the target 3D point cloud data set into the first semantic segmentation model, and output a 3D point cloud model set of the construction scene.

[0221] Among them, the three-dimensional point cloud model set of the construction scene includes the three-dimensional point cloud model of each construction component in the construction scene and the semantic information corresponding to the construction component; the first semantic segmentation model is obtained by training the deep learning network model with the first training sample set; the first training sample set includes at least one training sample; the training sample includes the training sample three-dimensional point cloud data and the three-dimensional point cloud model of the construction component corresponding to the training sample three-dimensional point cloud data and the real semantic information corresponding to the construction component.

[0222] The quality inspection unit 805 is used to generate a quality inspection result of the construction component according to the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the corresponding construction component in the BIM three-dimensional point cloud data set.

[0223] For the specific working process of the units disclosed in the above embodiments of the present application, please refer to the corresponding method embodiments, such as Figure 1 As shown, no further description is given here.

[0224] As can be seen from the above scheme, the present application provides a construction component quality detection device based on three-dimensional point cloud and BIM. After the first acquisition unit 801 acquires the three-dimensional point cloud data set of the construction scene and the BIM model of the construction target, the first conversion unit 802 converts the information in the BIM model to obtain the BIM three-dimensional point cloud data set; wherein the number of points in the BIM three-dimensional point cloud data set is in the same order of magnitude as the number of points in the three-dimensional point cloud data set of the construction scene; then, the registration unit 803 registers the three-dimensional point cloud data set of the construction scene and the BIM three-dimensional point cloud data set to obtain the registration result; the first input unit 804 inputs the target three-dimensional point cloud data set into the first semantic segmentation model, and outputs a set of three-dimensional point cloud models of the construction scene; finally, the quality detection unit 805 generates the quality detection result of the construction component according to the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the construction component corresponding to the BIM three-dimensional point cloud data set. The quality detection of the construction component is realized, and the objectivity and accuracy of the engineering construction quality detection are improved.

[0225] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0226] Another embodiment of the present application provides an electronic device, including:

[0227] One or more processors.

[0228] A storage device having one or more programs stored thereon.

[0229] When the one or more programs are executed by the one or more processors, the one or more processors implement the construction component quality detection method based on three-dimensional point cloud and BIM as described in any one of the above embodiments.

[0230] Another embodiment of the present application provides a computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the construction component quality inspection method based on three-dimensional point cloud and BIM as described in any one of the above embodiments is implemented.

[0231] In the context of the present application, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0232] It should be noted that the computer-readable medium mentioned above in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0233] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0234] Another embodiment of the present application provides a computer program product, which, when executed, is used to execute any of the above-mentioned construction component quality inspection methods based on three-dimensional point cloud and BIM.

[0235] In particular, according to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processing device, the above-mentioned functions defined in the method of the embodiment of the present application are executed.

[0236] Although the subject matter has been described in language specific to structural features and / or method logic actions, it should be understood that the subject matter defined in this application is not necessarily limited to the specific features or actions described above. On the contrary, the specific features and actions described above are merely example forms of implementing this application.

[0237] Although several specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present application. Certain features described in the context of a separate embodiment can also be implemented in a single embodiment in combination. On the contrary, the various features described in the context of a single embodiment can also be implemented in multiple embodiments individually or in any suitable sub-combination.

[0238] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of application involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above application concept. For example, the above features are replaced with (but not limited to) technical features with similar functions applied in the present application.

Claims

1. A construction component quality inspection method based on three-dimensional point cloud and BIM, characterized in that: include: Obtain the 3D point cloud dataset of the construction scene and the BIM model of the construction target; Converting the information in the BIM model to obtain a BIM three-dimensional point cloud dataset; wherein the number of points in the BIM three-dimensional point cloud dataset is in the same order of magnitude as the number of points in the three-dimensional point cloud dataset of the construction scene; The three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM are registered to obtain a registration result; wherein the registration result includes a target three-dimensional point cloud dataset; the target three-dimensional point cloud dataset is a three-dimensional point cloud dataset obtained after registering the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM; the target three-dimensional point cloud dataset is paired with the points in the three-dimensional point cloud dataset of the BIM; wherein, the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM are registered to obtain a registration result, including: generating a rigid body transformation matrix according to the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM, generating a transformed three-dimensional point cloud dataset of the construction scene according to the rigid body transformation matrix, and when the Euclidean distance between the transformed three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM reaches a preset convergence condition, using the transformed three-dimensional point cloud dataset of the construction scene as the target three-dimensional point cloud dataset; The target three-dimensional point cloud data set is input into the first semantic segmentation model, and a three-dimensional point cloud model set of the construction scene is output; wherein the three-dimensional point cloud model set of the construction scene includes a three-dimensional point cloud model of each construction component in the construction scene and semantic information corresponding to the construction component; the first semantic segmentation model is obtained by training a deep learning network model with a first training sample set; the first training sample set includes at least one training sample; the training sample includes training sample three-dimensional point cloud data and a three-dimensional point cloud model of a construction component corresponding to the training sample three-dimensional point cloud data and real semantic information corresponding to the construction component; The quality inspection results of the construction components are generated according to the 3D point cloud models of the construction components in the construction scene and the corresponding 3D point cloud models of the construction components in the BIM 3D point cloud data set.

2. The construction component quality detection method based on three-dimensional point cloud and BIM according to claim 1 is characterized in that: Before registering the three-dimensional point cloud dataset of the construction scene and the BIM three-dimensional point cloud dataset to obtain the registration result, the method further includes: For each point in the three-dimensional point cloud data set of the construction scene, calculate the average distance between the point and all nearby points of the point; If the average distance is greater than the standard range threshold, the point is confirmed to be an outlier and is removed from the three-dimensional point cloud data set of the construction scene.

3. The construction component quality detection method based on three-dimensional point cloud and BIM according to claim 1 is characterized in that: The registering the three-dimensional point cloud dataset of the construction scene and the BIM three-dimensional point cloud dataset to obtain a registration result includes: The three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM are roughly aligned to obtain a pending three-dimensional point cloud dataset; wherein the pending three-dimensional point cloud dataset is a three-dimensional point cloud dataset obtained after the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM are roughly aligned; The pending three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset are precisely aligned to obtain an alignment result.

4. The construction component quality detection method based on three-dimensional point cloud and BIM according to claim 3 is characterized in that: The step of roughly registering the three-dimensional point cloud dataset of the construction scene with the BIM three-dimensional point cloud dataset to obtain a pending three-dimensional point cloud dataset includes: In the BIM three-dimensional point cloud data set, N random points are randomly selected; wherein N is greater than or equal to 3; In the three-dimensional point cloud data set of the construction scene, a first assumed point is randomly selected; wherein the first assumed point is used as a corresponding point of the first random point; Calculate the Euclidean distance between the first random point and the second random point to obtain the first target distance; In the three-dimensional point cloud data set of the construction scene, find a point whose distance from the first assumed point is the first target distance, and randomly select one as a second assumed point; wherein the second assumed point serves as a corresponding point of the second random point; Calculate the Euclidean distance between the Nth random point and the N-2th random point to obtain the 2N-4th target distance; Calculate the Euclidean distance between the Nth random point and the N-1th random point to obtain the 2N-3th target distance; In the three-dimensional point cloud data set of the construction scene, find a point whose distance from the N-2th assumed point is the 2N-4th target distance and whose distance from the N-1th assumed point is the 2N-3th target distance, and randomly select one of them as the Nth assumed point; wherein the Nth assumed point serves as the corresponding point of the Nth random point; Generate a first rigid body transformation matrix according to the assumed point set and the random point set; wherein the first rigid body transformation matrix includes a first rotation matrix and a first displacement vector; Using the first rigid body transformation matrix, a rigid body transformation is performed on the three-dimensional point cloud dataset of the construction scene to obtain a transformed three-dimensional point cloud dataset of the construction scene; For each remaining point in the transformed three-dimensional point cloud data set of the construction scene, calculate the Euclidean distance between the remaining point and the corresponding point in the BIM three-dimensional point cloud data set; wherein the remaining point is the point remaining after removing all the assumed points; If the Euclidean distance between the remaining point and the corresponding point in the BIM three-dimensional point cloud dataset is less than the error threshold, the remaining point and the corresponding point in the BIM three-dimensional point cloud dataset are taken as sample inliers, and the process returns to the step of randomly selecting N random points in the BIM three-dimensional point cloud dataset; When the preset number of iterations is reached, the 3D point cloud dataset of the construction scene with the largest number of points in the sample is selected as the pending 3D point cloud dataset.

5. The construction component quality detection method based on three-dimensional point cloud and BIM according to claim 3 is characterized in that: The step of precisely registering the pending three-dimensional point cloud dataset with the BIM three-dimensional point cloud dataset to obtain a registration result includes: Performing preliminary matching on the pending three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset to obtain a preliminary matched pending three-dimensional point cloud dataset and a preliminary matched BIM three-dimensional point cloud dataset; Generate a second rigid body transformation matrix according to the initially matched pending three-dimensional point cloud dataset and the initially matched BIM three-dimensional point cloud dataset; wherein the second rigid body transformation matrix includes a second rotation matrix and a second displacement vector; Using the second rigid body transformation matrix, performing a rigid body transformation on the pending three-dimensional point cloud dataset to obtain a transformed pending three-dimensional point cloud dataset; Calculate the average Euclidean distance between the transformed undetermined three-dimensional point cloud dataset and all points in the BIM three-dimensional point cloud dataset; If the average Euclidean distance between the transformed pending 3D point cloud dataset and all points in the BIM 3D point cloud dataset is greater than the average Euclidean distance threshold, the transformed pending 3D point cloud dataset is used as a new pending 3D point cloud dataset, and the step of performing preliminary matching between the pending 3D point cloud dataset and the BIM 3D point cloud dataset is performed to obtain the preliminary matched pending 3D point cloud dataset and the preliminary matched BIM 3D point cloud dataset. When the average Euclidean distance between the transformed pending three-dimensional point cloud dataset and all points in the BIM three-dimensional point cloud dataset reaches a preset convergence condition, the transformed pending three-dimensional point cloud dataset is used as the target three-dimensional point cloud dataset.

6. The construction component quality detection method based on three-dimensional point cloud and BIM according to claim 4 is characterized in that: The step of generating a first rigid body transformation matrix according to the assumed point set and the random point set includes: Calculate the centroid of the assumed point set and the centroid of the random point set respectively; According to the centroid of the assumed point set and the centroid of the random point set, the assumed point set and the random point set are de-centroided to obtain a de-centroided assumed point set and a de-centroided random point set; A first covariance matrix is ​​calculated based on the de-centred assumed point set, the de-centred random point set and the first weight set; The first covariance matrix is ​​solved by a singular value decomposition method to obtain an optimal rotation matrix and an optimal translation matrix between the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM; A first rigid body transformation matrix is ​​determined according to an optimal rotation matrix and an optimal translation matrix between the three-dimensional point cloud dataset of the construction scene and the BIM three-dimensional point cloud dataset.

7. The construction component quality detection method based on three-dimensional point cloud and BIM according to claim 5 is characterized in that: The generating a second rigid body transformation matrix according to the initially matched pending three-dimensional point cloud dataset and the initially matched BIM three-dimensional point cloud dataset comprises: The centroid of the undetermined 3D point cloud dataset after the initial matching and the centroid of the BIM 3D point cloud dataset after the initial matching are calculated respectively; According to the centroid of the initially matched pending 3D point cloud dataset and the centroid of the initially matched BIM 3D point cloud dataset, de-centroiding the initially matched pending 3D point cloud dataset and the initially matched BIM 3D point cloud dataset to obtain a de-centroided initially matched pending 3D point cloud dataset and a de-centroided initially matched BIM 3D point cloud dataset; A second covariance matrix is ​​obtained by calculating the undetermined three-dimensional point cloud dataset after the initial matching without centroid, the BIM three-dimensional point cloud dataset after the initial matching without centroid, and the second weight set; The second covariance matrix is ​​solved by a singular value decomposition method to obtain an optimal rotation matrix and an optimal translation matrix between the undetermined three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset; A second rigid body transformation matrix is ​​determined according to an optimal rotation matrix and an optimal translation matrix between the pending three-dimensional point cloud dataset and the BIM three-dimensional point cloud dataset.

8. The construction component quality detection method based on three-dimensional point cloud and BIM according to claim 1 is characterized in that: After inputting the target 3D point cloud data set into the first semantic segmentation model and outputting a 3D point cloud model set of the construction scene, the method further includes: Acquire construction scene images; The construction scene image is input into a second semantic segmentation model, and the semantic segmentation result of the construction scene image is obtained as an output; wherein the second semantic segmentation model is obtained by training a deep learning neural network with a training sample set; the training sample set includes at least one training sample; the training sample includes: a training sample image and a real semantic segmentation result corresponding to the training sample image; The three-dimensional point cloud model set of the construction scene is verified using the semantic segmentation result of the construction scene image to obtain a verification result.

9. The construction component quality detection method based on three-dimensional point cloud and BIM according to claim 1 is characterized in that: Generating the quality inspection result of the construction component according to the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the construction component corresponding to the three-dimensional point cloud data set of the BIM includes: The root mean square deviation of the construction component is calculated based on the three-dimensional point cloud model of the construction component in the construction scene and the three-dimensional point cloud model of the corresponding construction component in the BIM three-dimensional point cloud data set; According to the semantic information corresponding to the construction component, the type of the construction component is determined, and the target threshold is determined according to the type of the construction component; wherein, if the root mean square deviation of the construction component is greater than the target threshold, the construction component is determined to be unqualified; if the root mean square deviation of the construction component is not greater than the target threshold, the construction component is determined to be qualified.

10. A construction component quality inspection device based on three-dimensional point cloud and BIM, characterized in that: include: A first acquisition unit is used to acquire a three-dimensional point cloud data set of a construction scene and a BIM model of a construction target; A first conversion unit is used to convert the information in the BIM model to obtain a BIM three-dimensional point cloud data set; wherein the number of points in the BIM three-dimensional point cloud data set is in the same order of magnitude as the number of points in the three-dimensional point cloud data set of the construction scene; A registration unit is used to register the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM to obtain a registration result; wherein the registration result includes a target three-dimensional point cloud dataset; the target three-dimensional point cloud dataset is a three-dimensional point cloud dataset obtained after registering the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM; the target three-dimensional point cloud dataset is paired with the points in the three-dimensional point cloud dataset of the BIM; wherein the registration of the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM to obtain a registration result includes: generating a rigid transformation matrix according to the three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM, generating a transformed three-dimensional point cloud dataset of the construction scene according to the rigid transformation matrix, and when the Euclidean distance between the transformed three-dimensional point cloud dataset of the construction scene and the three-dimensional point cloud dataset of the BIM reaches a preset convergence condition, using the transformed three-dimensional point cloud dataset of the construction scene as the target three-dimensional point cloud dataset; A first input unit is used to input the target three-dimensional point cloud data set into a first semantic segmentation model, and output a three-dimensional point cloud model set of the construction scene; wherein the three-dimensional point cloud model set of the construction scene includes a three-dimensional point cloud model of each construction component in the construction scene and semantic information corresponding to the construction component; the first semantic segmentation model is obtained by training a deep learning network model with a first training sample set; the first training sample set includes at least one training sample; the training sample includes training sample three-dimensional point cloud data and a three-dimensional point cloud model of a construction component corresponding to the training sample three-dimensional point cloud data, and real semantic information corresponding to the construction component; The quality inspection unit is used to generate quality inspection results of the construction components according to the three-dimensional point cloud model of the construction components in the construction scene and the three-dimensional point cloud model of the corresponding construction components in the BIM three-dimensional point cloud data set.

Images