A method for extracting suspicious points in the engineering management audit of construction projects based on 3D point clouds

Abstract

The present invention relates to the field of engineering auditing, and specifically relates to a method for extracting engineering management auditing doubts in construction projects based on 3D point clouds, including: Step 1: Construct an auditing data resource pool with the BIM model as the main body; Step 2: Place two types of non-homologous data in the same coordinate system; Step 3: Associate the geometric data of the BIM model with project management information; Step 4: Pre-train a laser point cloud semantic segmentation model to accurately identify key components in the actually measured point cloud during the auditing stage; Step 5: Process the point cloud data, identify key engineering components, perform geospatial matching and fusion of the identification results with the component point cloud of the BIM model, calculate and compare the geometric feature differences of the key components, and complete the doubt detection. By introducing advanced 3D point cloud and machine learning technologies, an efficient, accurate, and intelligent method for extracting engineering management auditing doubts in construction projects is provided, providing strong technical support for engineering management and auditing work.

Description

Technical Field

[0001] The present invention relates to the field of engineering management auditing, and particularly to a method for extracting audit doubts in construction project engineering management based on three-dimensional point clouds. Background Art

[0002] Construction project engineering management auditing refers to the process in which auditors, based on relevant laws, regulations, industry norms, and relevant design contract documents for engineering construction, audit the quality, safety, and progress of engineering construction to promote the smooth realization of project construction goals. The main contents of the audit mainly include reviewing the implementation of the construction project progress management plan and the quality acceptance situation during the construction process. In modern construction projects, with the continuous increase in project scale and complexity, the audit evidence collection work faces greater challenges. At present, many regions have begun to explore the use of laser point cloud and three-dimensional modeling technologies to carry out audit work. By constructing a three-dimensional model from the obtained point cloud data and comparing and analyzing it with the real scene model and BIM model, potential audit doubts can be preliminarily screened out. This technical method not only achieves full coverage of the project engineering, greatly reduces the workload of field verification, improves work efficiency, but also significantly reduces measurement errors, thus effectively ensuring the audit quality.

[0003] Although the above technical method has certain advantages in audit evidence collection, there are still several problems. For example, during the process of establishing a three-dimensional model based on point cloud data, the accuracy will be lost to some extent, which may lead to accuracy differences when comparing and analyzing with the BIM model subsequently. In addition, the modeling process usually requires a large amount of time and labor input, resulting in a time delay in the audit evidence collection work. Moreover, the current combined application of data mostly stays at the stage of visual display and simple comparative analysis, requiring a large amount of manual intervention. Not only is the degree of automation low, but also human errors may be introduced. So far, there is still a lack of methods that combine point cloud technology, BIM technology, and intelligent algorithms to achieve automated and intelligent audit evidence collection. Summary of the Invention

[0004] Aiming at the problems of low efficiency and insufficient accuracy in the existing construction project engineering management audit evidence collection methods, the purpose of the present invention is to provide a method for extracting audit doubts based on three-dimensional point clouds that directly compares the collected point cloud data with the BIM model parsed point cloud, reduces the accuracy loss and time consumption in the three-dimensional modeling link, and uses machine learning methods to automatically extract and identify audit doubts in engineering management, improving audit efficiency and accuracy.

[0005] To achieve the above purpose, the technical solution adopted by the present invention is: a method for extracting audit doubts in construction project engineering management based on three-dimensional point clouds, including the following steps:

[0006] Step 1: Audit data preparation. Carry out data collection, sorting, and storage work, construct an audit data resource pool based on the BIM model as the main body, and accurately construct three-dimensional portraits of different time stages during the project implementation process.

[0007] Step 2: BIM model geographical location definition and spatial reference conversion. Define the geographical and spatial location of the BIM model, and convert it to the laser scanning reference coordinate system to place the two types of non-homologous data in the same coordinate system.

[0008] Step 3: BIM model parsing of component point clouds and associated project management attributes. Parse the geometric data of the BIM model into three-dimensional point clouds and associate them with project management information.

[0009] Step 4: Train the laser point cloud semantic segmentation model. Pre-train the laser point cloud semantic segmentation model based on machine learning methods to accurately identify key components in the actually measured point clouds during the audit stage.

[0010] Step 5: Automatically detect doubts based on the actually measured point clouds and the BIM component point clouds. Use the semantic segmentation model obtained in Step 4 to process the point cloud data of each on-site measurement, automatically identify the key engineering components in the construction project, perform geographical and spatial matching and fusion of the identification results with the BIM model component point clouds, and complete the doubt detection by automatically calculating and comparing the geometric feature differences of the key components in the two sources of point cloud data.

[0011] The above method for extracting doubts in the project management audit of construction projects based on three-dimensional point clouds further includes Step 6: Doubt elimination, annotation, and visualization output. Conduct human-computer interaction verification and elimination for the automatically detected doubts. By checking relevant evidence or records, judge whether the doubts are valid. If it is found that the doubts are caused by design changes, data entry errors, or other normal business activities, then these doubts are eliminated.

[0012] The above method for extracting doubts in the project management audit of construction projects based on three-dimensional point clouds, Step 1 includes:

[0013] Step 1-1: Obtain relevant content of the project through business systems, regulatory platforms, and Internet channels, conduct data governance, record the key time nodes of the project, and convert the information of relevant content and key time nodes into structured data for storage as project feature inputs.

[0014] Step 1-2: Use the three-dimensional model constructed by BIM software during the project design stage, and supplement and add the material information of the components according to the design specifications and construction codes.

[0015] Step 1-3: Collect the early-stage laser point cloud data of the project as the training data source for the semantic segmentation model, record and save the multi-phase laser point cloud results, and form a complete data chain.

[0016] The above method for extracting audit doubts in construction project engineering management based on 3D point clouds, wherein step 2 includes:

[0017] Step 2-1: Define the projection coordinate system adopted in the project design and the geographical coordinate system to be converted.

[0018] Step 2-2: Set the planar coordinate information of the model data through BIM software to determine its longitude, latitude, and altitude.

[0019] Step 2-3: Obtain the angle between the north of the project and the true north through drawing measurement as the angle value between the BIM model and the true north.

[0020] Step 2-4: Use a conversion tool or conversion algorithm to convert the BIM model from planar coordinates to geospatial coordinates.

[0021] The above method for extracting audit doubts in construction project engineering management based on 3D point clouds, wherein step 3 includes:

[0022] Step 3-1: Screen the key component categories participating in the audit detection according to semantic information, extract the corresponding BIM model components and convert them into triangular surface models. After conversion, perform topological structure correction and data compression to reduce the number of triangles and lower the model complexity.

[0023] Step 3-2: Determine the sampling point density, capture the curvature changes on the surface of the geometry, and realize the 3D point cloud representation of the BIM model components through the closed contour point cloud filling method.

[0024] Step 3-3: Assign attributes to the sampling points, realize the association between the sampling points and the project management information, and export them in the standard point cloud file format.

[0025] The above method for extracting audit doubts in construction project engineering management based on 3D point clouds, in step 4, the pre-training includes dataset preparation and model training, and the specific steps are as follows:

[0026] Step 4-1: Manually annotate the previously collected point cloud data to generate point cloud samples covering important category components and their corresponding semantic labels, and integrate them into the existing 3D public dataset to form a high-quality training dataset covering diverse scenarios.

[0027] Step 4-2: Comprehensively consider the accuracy of local feature extraction and the efficiency of global semantic segmentation, and select a suitable training model framework.

[0028] Step 4-3: Preprocess the point cloud data and divide the dataset into a training set, a validation set, and a test set;

[0029] Step 4-4: Define the loss function and the optimization strategy, gradually iterate and optimize the model, adjust the hyperparameters using the validation set, improve the generalization ability and robustness of the BIM model, and evaluate the BIM model through the test set.

[0030] For the above method for extracting audit doubts in construction project engineering management based on 3D point cloud, in step 5, the contents to be detected include: detecting position and / or dimension deviation, detecting material deviation, and detecting unfinished parts.

[0031] For the above method for extracting audit doubts in construction project engineering management based on 3D point cloud, to detect the position and / or dimension deviation, calculate the symmetric distance between two clusters of point clouds of the same component in the adjacent space, and determine whether there is a position deviation. When this distance exceeds the set threshold, it can be determined that the position of the component has deviated. The specific method is as follows:

[0032] Step a: Assume two clusters of point clouds, where point cloud A contains the point set , and point cloud B contains the point set . For each point in point cloud A, find the nearest point in point cloud B and calculate its distance. Similarly, for each point in point cloud B, find the nearest point in point cloud A;

[0033] Step b: The average nearest point distance from point cloud A to point cloud B:

[0034] , where m is a constant;

[0035] Step c: The average nearest point distance from point cloud B to point cloud A:

[0036] , where n is a constant;

[0037] Step d: Calculate the symmetric distance:

[0038] .

[0039] For the above method for extracting audit doubts in construction project engineering management based on 3D point cloud, detecting the material deviation includes extracting the color features of each component in the measured point cloud, and finding the corresponding BIM model component point cloud based on its spatial position to obtain the corresponding material semantic information, and comparing it with the point cloud data with the same material label in the training dataset. The specific steps are as follows:

[0040] Step e: Discretize the color values into multiple color intervals, calculate the number or frequency of point clouds within each color interval, and obtain a color histogram as a description of the color feature;

[0041] Step f: Through the distance metric formula , compare the differences in the color histograms of two groups of point clouds to quantify the material deviation, where and respectively represent the histogram values of point cloud A and point cloud B within the color interval.

[0042] For the above method for extracting suspicious points in the engineering management audit of construction projects based on 3D point clouds, the detection of unfinished parts includes:

[0043] Step g: Screen out the components that should have been built at the current time in the BIM model point cloud by combining the semantic information of the construction period, and search for the corresponding measured point cloud based on the spatial position;

[0044] Step h: Quantify the point cloud coverage by calculating the intersection over union between the two. Perform voxelization on the two point clouds respectively, assign each point to the corresponding voxel, and calculate the intersection and union of the two sets of voxels. Its overlap degree is expressed as: , when the calculation result is lower than a certain threshold, it can be determined that the component is unfinished.

[0045] The beneficial effects of the method for extracting suspicious points in the engineering management audit of construction projects based on 3D point clouds according to the present invention are as follows: By constructing a 3D portrait, it realizes the reproduction of scenarios and data traceability at different times, and spatial reference conversion, laying a foundation for the precise integration and comparative analysis of the two in the spatial dimension during the audit stage. Through automated anomaly detection, the audit efficiency is greatly improved, and the time and cost of manual intervention are reduced; By using the method of combining 3D point cloud data and BIM models, high-precision geometric information can be provided, and by introducing machine learning technology, historical data can be trained and learned to establish an intelligent anomaly detection model. The model can continuously optimize itself, and with the increase in the amount and diversity of data, the detection accuracy and reliability will also continue to improve, enhancing the accuracy of audit evidence collection; Moreover, the present invention extracts suspicious points from multiple dimensions, which can not only detect dimensional deviations and material deviations during the construction process, but also identify unfinished parts, comprehensively evaluate the construction quality and progress from multiple dimensions, and improve the comprehensiveness and in-depthness of the audit; By realizing the three-dimensional visualization of the evidence collection results, the extracted suspicious point records are intuitive and easy to understand, including both the detected abnormal situations and specific locations, and can be combined with text and pictures for display, facilitating the rapid understanding and judgment of auditors.

[0046] The present invention provides an efficient, accurate, and intelligent method for extracting suspicious points in the engineering management audit of construction projects by introducing advanced three-dimensional point cloud and machine learning technologies, providing strong technical support for engineering management and audit work, and having significant application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the process for extracting suspicious points in the engineering management audit of construction projects based on three-dimensional point cloud of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] To enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention will be described below in conjunction with the specific embodiments and the accompanying drawings.

[0049] Figure 1 The processing process of extracting suspicious points in the engineering management audit of construction projects based on three-dimensional point cloud is fully elaborated. Based on Figure 1 , the extraction method of the present invention is as follows.

[0050] Embodiment 1

[0051] A method for extracting suspicious points in the engineering management audit of construction projects based on three-dimensional point cloud includes the following steps:

[0052] Step 1: Audit data preparation, carry out data collection, collation, and storage work, construct an audit data resource pool based on the BIM model as the main body, and accurately construct three-dimensional portraits at different time stages during the project implementation process.

[0053] Step 1-1: Obtain relevant content of the project through business systems, supervision platforms, and Internet channels, conduct data governance, record the key time nodes of the project, and convert the information of relevant content and key time nodes into structured data for storage as project feature inputs.

[0054] Step 1-2: Use the three-dimensional model constructed by BIM software during the project design stage, and supplement and add the material information of components according to the design description and construction specifications.

[0055] Step 1-3: Collect the early laser point cloud data of the project as the training data source for the semantic segmentation model, record and save the multi-period laser point cloud results, and form a complete data chain.

[0056] Step 2: Definition of the geographical location of the BIM model and spatial reference conversion, define the geographical spatial location of the BIM model, and convert it to the laser scanning reference coordinate system to place two types of non-homologous data in the same coordinate system.

[0057] Step 2-1: Clarify the projection coordinate system adopted in the project design and the geographical coordinate system to be converted.

[0058] Step 2-2: Set the planar coordinate information of the model data through BIM software to determine its longitude, latitude, and height.

[0059] Step 2-3: Obtain the angle between the north of the project and true north through drawing measurement as the angle value between the BIM model and true north.

[0060] Step 2-4: Use a conversion tool or conversion algorithm to convert the BIM model from planar coordinates to geospatial coordinates.

[0061] Step 3: The BIM model parses the component point cloud and associated project management attributes, parses the geometric data of the BIM model into a three-dimensional point cloud, and associates it with project management information.

[0062] Step 3-1: Screen the key component categories participating in the audit and detection according to semantic information, extract the corresponding BIM model components and convert them into a triangular mesh model. After conversion, perform topological structure correction and data compression to reduce the number of triangles and lower the model complexity.

[0063] Step 3-2: Determine the sampling point density, capture the curvature change of the geometric body surface, and realize the three-dimensional point cloud representation of the BIM model components through the closed contour point cloud filling method.

[0064] Step 3-3: Endow the sampling points with attributes, realize the association between the sampling points and project management information, and export them in the standard point cloud file format.

[0065] Step 4: Train the laser point cloud semantic segmentation model, pre-train the laser point cloud semantic segmentation model based on machine learning methods, and accurately identify the key components in the measured point cloud during the audit phase.

[0066] The pre-training includes dataset preparation and model training. The specific steps are as follows:

[0067] Step 4-1: Manually annotate the point cloud data collected in the early stage to generate point cloud samples covering important category components and their corresponding semantic labels, and integrate them into the existing three-dimensional public dataset to form a high-quality training dataset covering diverse scenarios.

[0068] Step 4-2: Comprehensively consider the accuracy of local feature extraction and the efficiency of global semantic segmentation, and select a suitable training model framework.

[0069] Step 4-3: Perform preprocessing operations on the point cloud data, and divide the dataset into a training set, a validation set, and a test set.

[0070] Step 4-4: Define the loss function and optimization strategy, gradually iterate and optimize the model, adjust the hyperparameters using the validation set, enhance the generalization ability and robustness of the BIM model, and evaluate the BIM model through the test set.

[0071] Step 5: Automatically detect suspicious points based on the measured point cloud and the BIM component point cloud. Process the point cloud data of each on-site measurement using the semantic segmentation model obtained in Step 4 to automatically identify the key engineering components in the construction project. Geospatially match and fuse the recognition results with the BIM model component point cloud, and compare the geometric feature differences of the key components in the two-source point cloud data through automated calculation means to complete the suspicious point detection.

[0072] The contents of the detection include: detecting position and / or size deviation, detecting material deviation, and detecting unfinished parts.

[0073] Step 6: Exclude, label, and visually output suspicious points. Conduct human-computer interaction verification and exclusion for the automatically detected suspicious points. By checking relevant evidence or records, judge whether the suspicious points are valid. If it is found that the suspicious points are caused by design changes, data entry errors, or other normal business activities, then exclude these suspicious points.

[0074] Embodiment 2

[0075] By comparing the differences between the three-dimensional scene of the current construction site and the design model before construction, discover and extract audit suspicious points. Among them, the three-dimensional scene of the current construction site is the point cloud data obtained through laser scanning technology, and the design model is the BIM model.

[0076] I. Audit data preparation.

[0077] Carry out data collection, collation, and storage work to construct an audit data resource pool with a three-dimensional model as the main body.

[0078] First, obtain the project's contracts, exploration reports, design drawings, construction plans, etc. through channels such as business systems, regulatory platforms, and the Internet, conduct data governance, record the key time nodes of the project (such as start date, completion date, phased acceptance date), and convert this information into structured data for storage as project feature inputs; second, prepare the project design BIM model as the basis for audit evidence collection. The three-dimensional model constructed using BIM software during the project design phase can be directly utilized, or a model can be rebuilt based on the design drawings, and the material information of the components can be supplemented and added according to the design specifications and construction codes. Third, collect the early laser point cloud data of the project as the training data source for the semantic segmentation model to improve the model training accuracy and overcome the limitations of existing public three-dimensional data sets.

[0079] Meanwhile, record and save the laser point cloud results of multiple periods to form a complete data chain, accurately construct the "3D portraits" at different time stages during the project implementation process, and based on this, realize the scenario reproduction and data traceability at different moments.

[0080] II. Definition of the geographical location of the BIM model and conversion of spatial references.

[0081] Define the geographical and spatial location of the BIM model and convert it to the laser scanning reference coordinate system, place the two types of non-homologous data in the same coordinate system, ensure the consistency of the spatial positions between the data, and lay a foundation for the accurate integration and comparative analysis of the two in the spatial dimension during the audit stage.

[0082] First, clarify the projection coordinate system adopted in the project design and the geographical coordinate system to be converted. Then, set the plane coordinate information of the model data through BIM software to determine its longitude, latitude, and altitude, which can be achieved by setting the model project base point or specifying shared coordinates. The specific operation is to select a certain determined point with known drawing coordinates on the model and set the actual coordinate values of this point, so as to reposition the model in the shared coordinate system and move it relative to the global positioning link. In addition, the angle between the model and due north needs to be set, and this angle value is the included angle between the project north and due north, which can be obtained by measuring on the drawing. Subsequently, use a conversion tool or conversion algorithm to convert the model from plane coordinates to geographical and spatial coordinates, and import the converted model into a geographic information system (GIS) for verification to ensure the accuracy and consistency of the conversion.

[0083] III. Analyze the component point cloud of the BIM model and associate project management attributes.

[0084] The BIM model contains rich geometric structures and semantic information, but there are significant differences in its geometric expression methods compared with laser point cloud data. Therefore, in order to achieve the efficient and automated matching and computational analysis of the BIM model and laser point cloud data, it is necessary to perform format conversion on the BIM geometric data, parse it into a 3D point cloud, and at the same time retain key attribute information such as component categories and materials. Moreover, the generated point cloud data will also be associated with project management information such as the construction period and time nodes extracted from project contracts and construction plans.

[0085] The specific process is as follows:

[0086] 1) Screen the key component categories participating in the audit and detection according to semantic information, such as walls, floors, doors, windows, etc., extract the corresponding BIM component models and convert them into triangular mesh models. After conversion, perform topological structure correction and data compression to reduce the number of triangles and lower the model complexity.

[0087] 2) Determine the sampling point density and uniformly sample within each triangular face. For curved surfaces, sample along the normal direction to accurately capture the curvature changes on the surface of the geometry, thereby realizing the three-dimensional point cloud representation of BIM components through the filling method of closed contour point clouds.

[0088] 3) Assign attributes to the sampling points, such as coordinates, normal vectors, etc., as well as custom attributes such as extended materials and construction periods, to realize the association with project management information and export it in the standard point cloud file format.

[0089] IV. Train the laser point cloud semantic segmentation model.

[0090] Pre-train the laser point cloud semantic segmentation model based on machine learning methods to achieve accurate identification of key components in the measured point cloud during the audit phase. The training process mainly includes two parts: dataset preparation and model training.

[0091] First, manually annotate the point cloud data collected in the early stage to generate point cloud samples covering important category components and their corresponding semantic labels, and integrate them into the existing three-dimensional public dataset to form a high-quality training dataset covering diverse scenarios to meet the point cloud segmentation requirements in complex indoor building environments; then, comprehensively consider the accuracy of local feature extraction and the efficiency of global semantic segmentation, and select a suitable training model framework.

[0092] Before model training, perform preprocessing operations such as normalization, downsampling, and filtering on the point cloud data, and divide the dataset into training set, validation set, and test set. During the model training process, gradually iterate and optimize the model by defining appropriate loss functions and optimization strategies, and use the validation set to adjust the hyperparameters to improve the generalization ability and robustness of the model. Finally, conduct model evaluation through the test set, verify the actual performance of the model based on the segmentation accuracy index, and ensure that it meets the expected accuracy requirements.

[0093] V. Automatically detect doubts based on the measured point cloud and the BIM component point cloud.

[0094] During the audit phase, use the semantic segmentation model trained in the previous steps to process the point cloud data measured on-site each time, automatically identify the key engineering components in the construction project, and then perform geospatial matching and fusion of the recognition results with the BIM component point cloud, and complete the doubt detection by automatically calculating and comparing the geometric feature differences of the key components in the two sources of point cloud data.

[0095] The specific detection content covers the following three aspects:

[0096] ①Position / dimension deviation: By calculating the symmetric distance between two clusters of point clouds of the same component in the adjacent space, it is judged whether there is a position deviation. When this distance exceeds the set threshold, it can be determined that the position of the component has deviated. Suppose there are two clusters of point clouds, where point cloud A contains the point set , and point cloud B contains the point set . For each point in point cloud A, find the nearest point in point cloud B, and calculate its distance. Similarly, for each point in point cloud B, find the nearest point in point cloud A, and calculate its distance.

[0097] The calculation process is as follows:

[0098] The average nearest point distance from point cloud A to point cloud B: ,

[0099] The average nearest point distance from point cloud B to point cloud A: ,

[0100] Symmetric distance calculation: .

[0101] ②Material deviation: Extract the color features of each component in the measured point cloud, and find the corresponding BIM component point cloud based on its spatial position to obtain the corresponding material semantic information. Then, compare it with the point cloud data with the same material label in the training dataset. The specific method is to discretize the color values into multiple color intervals, calculate the number or frequency of point clouds in each color interval, and obtain the color histogram as the description of the color features. Compare the differences in the color histograms of the two groups of point clouds through the distance metric formula, so as to quantify the material deviation.

[0102] The calculation formula is as follows: , where and respectively represent the histogram values of point cloud A and point cloud B in the color interval .

[0103] ③Detection of unfinished parts: Combine the semantic information of the construction period time to screen the components that should have been built at the current time in the BIM point cloud, and find the corresponding measured point cloud based on the spatial position. Quantify the point cloud coverage by calculating the intersection over union between the two. When the calculation result is lower than a certain threshold, it can be determined that the component is unfinished. The specific calculation steps are as follows: Perform voxelization processing on the two point clouds respectively, assign each point to the corresponding voxel, and calculate the intersection and union of the two sets of voxels.

[0104] Its overlap degree is expressed as: .

[0105] VI. Doubt Elimination, Annotation and Visualization Output.

[0106] For the doubts automatically detected, conduct human-computer interaction verification and elimination. By checking relevant evidence or records, determine whether the doubts are valid. If it is found that the doubts are caused by design changes, data entry errors or other normal business activities, then eliminate these doubts. Conduct three-dimensional visual annotation for the verified doubts, and estimate corresponding indicators through GIS functions such as spatial measurement, such as calculating the length, area, volume, etc. of each structure on the construction project, visually presenting the spatial location and abnormal characteristics of each doubt, and enhancing the transparency and interpretability of the results. Finally, batch output the doubt records in units of projects to support the subsequent audit work.

[0107] The above embodiments are only for illustrating the inventive concept and features of the present invention, and the purpose is to enable those of ordinary skill in the art to understand the content of the present invention and implement it accordingly, and it is not intended to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the essence of the content of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for extracting doubtful points in construction project engineering management audit based on three-dimensional point cloud, characterized in that: The following steps are involved: Step 1: Audit data preparation, data collection, organization and storage, building an audit data resource pool based on the BIM model, and accurately constructing a three-dimensional portrait of different time stages during the project implementation; Step 2: Define the geographic location of the BIM model and convert the spatial reference. Define the geographic location of the BIM model and convert it to the laser scanning reference coordinate system, placing the two types of non-homologous data in the same coordinate system, including: Step 2-1: Clarify the projection coordinate system used in the project design and the geographic coordinate system to be converted; Step 2-2: Set the model project base point or specify the shared coordinates to obtain the plane coordinate information of the data through the BIM software, determine its longitude, latitude and altitude, select a certain point on the model with known drawing coordinates, and set the actual coordinate value of the point, reposition the model in the shared coordinate system, and move it relative to the global positioning link; Step 2-3: Obtain the angle between the project north and true north through drawing measurement, which is used as the angle between the BIM model and true north; Step 2-4: Use conversion tools or conversion algorithms to convert the BIM model from plane coordinates to geospatial coordinates; Step 3: BIM model parsing component point cloud and associated project management attributes, parsing the geometric data of the BIM model into a 3D point cloud and associating it with project management information, including: Step 3-1: Filter the key component categories involved in the audit and detection based on semantic information, extract the corresponding BIM model components and convert them into triangular surface models, perform topological structure correction and data compression after conversion, reduce the number of triangles, and reduce the complexity of the model; Step 3-2: Determine the sampling point density, capture the curvature changes of the geometric surface, and realize the three-dimensional point cloud expression of the BIM model components through the closed contour point cloud filling method; Step 3-3: Assign sampling point attributes, associate sampling points with project management information, and export them to standard point cloud file format; Step 4: Train the laser point cloud semantic segmentation model. Pre-train the laser point cloud semantic segmentation model based on machine learning methods to accurately identify key components in the measured point cloud during the audit phase. Step 5: Perform automatic detection of suspicious points based on the measured point cloud and the BIM component point cloud. Use the semantic segmentation model obtained in step 4 to process the point cloud data of each on-site measurement, automatically identify the key engineering components in the construction project, match and fuse the identification results with the BIM model component point cloud in geographic space, and compare the geometric feature differences of the key components in the two source point cloud data by automated calculation methods to complete the suspicious point detection. The detection content includes: detecting position and / or size deviation, detecting material deviation, and detecting unfinished parts. The material deviation is measured by calculating the color histogram distance of two clusters of point clouds of the same component. The unfinished part is measured by calculating the voxel overlap intersection and union ratio of two clusters of point clouds of the same component. The position and / or size deviation is detected by calculating the symmetrical distance between two clusters of point clouds of the same component in the adjacent space to determine whether there is a position deviation. When the distance exceeds the set threshold, it can be determined that the component position has deviated. The specific method is as follows: Step a: Assume two clusters of point clouds, where point cloud A contains the point set , point cloud B contains the point set , for each point in point cloud A , find the nearest point in point cloud B , Similarly, for each point in point cloud B , find the nearest point in point cloud A ; Step b: The average closest point distance from point cloud A to point cloud B: , where m is a constant; Step c: The average closest point distance from point cloud B to point cloud A: , where n is a constant; Step d: Symmetric distance calculation: ； Step 6: Eliminate doubts, mark and visualize output. Perform human-computer interaction verification and elimination on the doubts detected automatically. By checking relevant evidence or records, determine whether the doubts are valid. If it is found that the doubts are caused by design changes, data entry errors or other normal business activities, these doubts will be eliminated.

2. The method for extracting doubtful points in construction project engineering management audit based on three-dimensional point cloud according to claim 1 is characterized in that: The step 1 comprises: Step 1-1: Obtain relevant content of the project through business systems, regulatory platforms and Internet channels, conduct data governance, record key time nodes of the project, and convert the relevant content and key time node information into structured data for storage as project feature input; Step 1-2: Use the 3D model built using BIM software during the project design phase to add component material information based on the design instructions and construction specifications; Step 1-3: Collect laser point cloud data from the early stages of the project as a training data source for the semantic segmentation model, record and save multiple laser point cloud results, and form a complete data chain.

3. The method for extracting doubtful points in construction project engineering management audit based on three-dimensional point cloud according to claim 1 is characterized in that: In step 4, pre-training includes data set preparation and model training, and the specific steps are as follows: Step 4-1: Manually annotate the point cloud data collected in the early stage, generate point cloud samples covering important categories of components and their corresponding semantic labels, and integrate them into the existing 3D public dataset to form a high-quality training dataset covering a variety of scenarios; Step 4-2: Consider the accuracy of local feature extraction and the efficiency of global semantic segmentation and select an appropriate training model framework; Step 4-3: Preprocess the point cloud data and divide the data set into training set, validation set and test set; Step 4-4: Define the loss function and optimization strategy, optimize the model step by step, adjust the hyperparameters using the validation set, improve the generalization ability and robustness of the BIM model, and evaluate the BIM model using the test set.

4. The method for extracting doubtful points in construction project engineering management audit based on three-dimensional point cloud according to claim 1 is characterized in that: The material deviation detection includes extracting the color features of each component in the measured point cloud, and searching for the corresponding BIM model component point cloud based on its spatial position to obtain the corresponding material semantic information, and comparing it with the point cloud data of the same material label in the training data set. The specific steps are as follows: Step e: discretize the color value into multiple color intervals, calculate the number or frequency of point clouds in each color interval, and obtain a color histogram as a description of the color feature; Step f: Use the distance metric formula Compare the color histogram differences of the two groups of point clouds to quantify the material deviation, where: and Respectively represent the histogram values ​​of point cloud A and point cloud B in the color range.

5. The method for extracting doubtful points in construction project engineering management audit based on three-dimensional point cloud according to claim 1 is characterized in that: The unfinished parts of the inspection include: Step g: Filter the components that should have been built at the current time in the BIM model point cloud by combining the construction period time semantics, and search for the measured point cloud that matches it based on the spatial position; Step h: quantify the coverage of the point clouds by calculating the intersection-union ratio between the two, voxelize the two point clouds respectively, assign each point to the corresponding voxel, and calculate the intersection and union of the two sets of voxel sets. The overlap is expressed as: , when the calculation result is lower than a certain threshold, it can be determined that the component is not completed.

Images