Three-dimensional point cloud segmentation method and system for bim-based assembly monolithic steel structure

Abstract

The application relates to a three-dimensional point cloud segmentation method and system for a BIM-based assembly monolithic steel structure. The method combines BIM prior knowledge and a deep learning model to convert manual annotation data into automatic segmentation capability, reducing the dependence on manual intervention. At the same time, the model can adapt to the diversity of interference point clouds in different construction scenes, and improve the segmentation robustness through feature learning. The method realizes the automatic segmentation of assembly monolithic steel structure point clouds in the construction scene, effectively solves the problems of low efficiency and high error rate of manual operation, and improves the accuracy of distinguishing target point clouds and interference point clouds by using BIM data to guide model training, providing reliable input for the accurate matching of BIM models and point cloud data.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the technical field of three-dimensional visualization, and particularly relate to a three-dimensional point cloud segmentation method and system for assembly monolithic steel structure based on BIM. BACKGROUND

[0002] As a modern construction technology, assembly monolithic steel structure combines factory prefabricated components with on-site assembly, fully utilizing the advantages of high strength and fast construction of steel structure.

[0003] With the in-depth application of BIM (Building Information Modeling) technology in engineering construction, building a BIM model for assembly monolithic steel structure can improve the quality and efficiency of design and construction of assembly monolithic steel structure. The precision requirements of BIM for three-dimensional point cloud data are increasingly high. In actual construction scenarios, laser scanners cannot automatically identify target objects and selectively scan, resulting in point cloud data obtained containing not only assembly monolithic steel structure but also a large amount of interference point cloud such as personnel, construction equipment and site, etc. The characteristics of these interference point clouds are complex and the volume is huge. Current point cloud processing technology cannot accurately distinguish assembly monolithic steel structure from construction environment impurities, which seriously restricts the accurate matching of BIM model and field situation, affecting construction quality control and progress management. SUMMARY

[0004] The following is a summary of the subject matter of the detailed description herein. This summary is not intended to limit the scope of the claims.

[0005] The main purpose of the embodiments of the present application is to provide a three-dimensional point cloud segmentation method and system for assembly monolithic steel structure based on BIM, which can extract three-dimensional point cloud data through BIM technology and combine with a point cloud segmentation model trained by labeling to accurately distinguish target steel structure from interference point cloud in a construction scene, has the advantages of improving the efficiency and accuracy of steel structure point cloud segmentation, and realizing automatic identification and segmentation of target components in a construction scene.

[0006] To achieve the above purpose, a first aspect of the embodiments of the present application provides a three-dimensional point cloud segmentation method for assembly monolithic steel structure based on BIM, the method comprising:

[0007] extracting three-dimensional point cloud data of a plurality of assembly monolithic steel structure samples in a construction scene based on BIM;

[0008] labeling three-dimensional point clouds related to the assembly monolithic steel structure samples and three-dimensional point clouds unrelated to the assembly monolithic steel structure samples in the three-dimensional point cloud data;

[0009] The labeled 3D point cloud data is used to train a point cloud segmentation model to obtain a trained point cloud segmentation model.

[0010] Based on the trained point cloud segmentation model, the target assembled monolithic steel structure is segmented from the 3D point cloud corresponding to the construction scene of the target assembled monolithic steel structure.

[0011] This application provides a BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures, which has at least the following beneficial effects:

[0012] This method combines BIM prior knowledge with a deep learning model to transform manually labeled data into automated segmentation capabilities, reducing reliance on manual intervention. Simultaneously, the model can adapt to the diversity of interfering point clouds in different construction scenarios, improving segmentation robustness through feature learning. This method achieves automated segmentation of point clouds for prefabricated steel structures in construction scenarios, effectively solving the problems of low efficiency and high error rate in manual operations. By using BIM data to guide model training, the accuracy of distinguishing between target and interfering point clouds is improved, providing reliable input for subsequent accurate matching of BIM models and point cloud data.

[0013] In some embodiments, extracting global features at multiple scales and local features at multiple scales from the input image includes:

[0014] The image to be input is input into the first branch network to obtain global features at multiple scales output by the first branch network; wherein, the first branch network includes multiple cascaded Transformer units and a downsampling unit is included between every two Transformer units;

[0015] The image to be input is input into the second branch network to obtain local features at multiple scales output by the second branch network; wherein the second branch network includes multiple cascaded CNN units and a downsampling unit is included between every two CNN units.

[0016] In some embodiments, the point cloud segmentation model includes:

[0017] A global feature extraction unit is used to extract global features from the three-dimensional point cloud data.

[0018] A local feature extraction unit is used to extract local features from the three-dimensional point cloud data.

[0019] A cross-attention mechanism unit is used to perform cross-attention fusion of the global features and the local features to obtain fused features;

[0020] A point cloud segmentation unit is used to segment the assembled monolithic steel structure sample based on the fusion features.

[0021] In some embodiments, the process by which the global feature extraction unit extracts global features from the three-dimensional point cloud data of any one of the assembled monolithic steel structure samples includes:

[0022] The first global feature extracted from 3D point cloud data based on a multilayer perceptron;

[0023] The first global feature is subjected to max pooling to obtain the second global feature;

[0024] Global features are extracted from the second global features using a multilayer perceptron.

[0025] In some embodiments, the local feature extraction unit shown is a 3D U-Net.

[0026] In some embodiments, the process by which the cross-attention mechanism unit extracts fused features from the global features and the local features includes:

[0027] The global features are subjected to average pooling to obtain the first global intermediate features; the local features are subjected to average pooling to obtain the first local intermediate features.

[0028] After the first global intermediate feature is forward serialized, it is mapped to the Q function, and after the first local intermediate feature is forward serialized, it is mapped to the K function and the V function.

[0029] The attention score is obtained by querying the K function according to the Q function, and the attention score is then activated by Softmax to obtain the first attention feature;

[0030] The first attention feature is multiplied pixel by pixel by the V function to obtain the second attention feature;

[0031] After deserializing the second attention feature, the fused feature is obtained.

[0032] In some embodiments, the process by which the point cloud segmentation unit segments the assembled monolithic steel structure sample from the fused features includes:

[0033] Perform max pooling on the fused features in the spatial dimension to obtain the first intermediate feature;

[0034] The fused feature is subjected to max pooling in the channel dimension to obtain the second intermediate feature;

[0035] The fused features are subjected to average pooling in the spatial dimension to obtain the third intermediate feature;

[0036] The fused feature is subjected to average pooling along the channel dimension to obtain the fourth intermediate feature;

[0037] The first feature and the second feature are multiplied pixel by pixel to obtain the fifth intermediate feature;

[0038] The third feature and the fourth feature are multiplied pixel by pixel to obtain the sixth intermediate feature;

[0039] The fused feature and the fifth intermediate feature are added element by element to obtain the seventh intermediate feature;

[0040] Subtract the seventh intermediate feature from the sixth intermediate feature element by element to obtain the eighth intermediate feature;

[0041] The assembled monolithic steel structure sample is segmented from the eighth intermediate feature based on the segmentation layer.

[0042] In some embodiments, the segmentation layer is a multilayer perceptron.

[0043] To achieve the above objectives, a second aspect of this application provides a BIM-based three-dimensional point cloud segmentation system for prefabricated monolithic steel structures, the system comprising:

[0044] The 3D point cloud acquisition module is used to extract 3D point cloud data of multiple prefabricated steel structure samples in the construction scene based on BIM.

[0045] A 3D point cloud annotation module is used to annotate the 3D point cloud data with respect to the assembled monolithic steel structure sample and with respect to the 3D point cloud data with respect to the assembled monolithic steel structure sample.

[0046] The segmentation model training module is used to train a point cloud segmentation model using the labeled 3D point cloud data to obtain the trained point cloud segmentation model.

[0047] The 3D point cloud segmentation module is used to segment the target assembled steel structure from the 3D point cloud corresponding to the construction scene of the target assembled steel structure based on the trained point cloud segmentation model.

[0048] To achieve the above objectives, a third aspect of this application provides an electronic device, comprising: at least one control processor and a memory for communicatively connecting to the at least one control processor; the memory stores instructions executable by the at least one control processor, the instructions being executed by the at least one control processor to enable the at least one control processor to perform the above-described BIM-based three-dimensional point cloud segmentation method for assembled monolithic steel structures.

[0049] To achieve the above objectives, a fourth aspect of this application provides a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the above-described BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures.

[0050] It is understood that the beneficial effects of the second to fourth aspects compared with the related technologies are the same as the beneficial effects of the first aspect compared with the related technologies. Please refer to the relevant description in the first aspect above, which will not be repeated here. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart illustrating a BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures, provided in one embodiment of this application.

[0053] Figure 2 This is a structural schematic diagram of a three-dimensional point cloud segmentation system for a BIM-based prefabricated monolithic steel structure provided in one embodiment of this application;

[0054] Figure 3 This is a schematic diagram of the structure of an electronic device provided in one embodiment of this application. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0056] As a modern construction technology, prefabricated steel structures fully leverage the advantages of high strength and fast construction by combining factory-prefabricated components with on-site assembly.

[0057] With the in-depth application of BIM (Building Information Modeling) technology in engineering construction, building BIM models of prefabricated steel structures can improve the quality and efficiency of prefabricated steel structure design and construction.

[0058] BIM demands increasingly higher accuracy from 3D point cloud data. In actual construction scenarios, laser scanners cannot automatically identify target objects and perform selective scanning, resulting in the acquisition of point cloud data containing not only prefabricated steel structures but also a large number of interfering point clouds, such as personnel, construction equipment, and sites. These interfering point clouds are complex in characteristics and enormous in size.

[0059] Current point cloud processing technology struggles to accurately distinguish between prefabricated steel structures and impurities in the construction environment, severely hindering the precise matching of BIM models with actual site conditions and impacting construction quality control and schedule management.

[0060] like Figure 1 To address the shortcomings of current methods, one embodiment of this application provides a BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures. The method includes the following steps S110 to S140:

[0061] Step S110: Extract three-dimensional point cloud data of multiple prefabricated steel structure samples in the construction scenario based on BIM.

[0062] Step S120: Label the three-dimensional point cloud data related to the assembled monolithic steel structure sample, as well as the three-dimensional point cloud data unrelated to the assembled monolithic steel structure sample.

[0063] Step S130: Train the point cloud segmentation model using the labeled 3D point cloud data to obtain the trained point cloud segmentation model.

[0064] Step S140: Based on the trained point cloud segmentation model, the target assembled monolithic steel structure is segmented from the 3D point cloud corresponding to the construction scene of the target assembled monolithic steel structure.

[0065] In step S110, BIM is a tool used in architecture, engineering, and civil engineering. 3D point cloud data extraction refers to exporting the geometric information of the prefabricated steel structure sample through the BIM software interface and combining it with the actual construction scene point cloud data of the prefabricated steel structure sample obtained by a laser scanner. Specifically, data matching can be achieved by aligning the component coordinates of the BIM model with the point cloud coordinates.

[0066] In step S120, the annotation tool is used to annotate the 3D point cloud data related to the prefabricated steel structure sample (the 3D point cloud of the prefabricated steel structure sample itself) and the 3D point cloud unrelated to the prefabricated steel structure sample (such as people, equipment, etc.). Each point in the dataset is assigned a category label, for example, points belonging to the prefabricated steel structure are labeled as "target", and the rest are labeled as "background".

[0067] In step S130, a point cloud segmentation model is constructed and trained using labeled 3D point cloud data to obtain a trained point cloud segmentation model. During subsequent training, the point cloud segmentation model extracts spatial features of the point cloud through a multilayer perceptron and enhances its ability to identify target regions using an attention mechanism. After training, the model can automatically analyze newly acquired construction scene point clouds and output the segmented steel structure point cloud.

[0068] In step S140, the target assembled monolithic steel structure is the assembled monolithic steel structure that needs to be segmented. A 3D laser scanner is used to collect the 3D point cloud of the construction scene corresponding to the target assembled monolithic steel structure. Then, based on the trained point cloud segmentation model, the target assembled monolithic steel structure can be segmented from it.

[0069] The BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures provided in this embodiment has the following beneficial effects:

[0070] Existing methods rely on manual point cloud screening, which is inefficient due to operator experience and struggles to handle large-scale, complex scenarios. This method combines BIM prior knowledge with a deep learning model to transform manually labeled data into automated segmentation capabilities, reducing reliance on manual intervention. Simultaneously, the model can adapt to the diversity of interfering point clouds in different construction scenarios, improving segmentation robustness through feature learning. This method achieves automated segmentation of point clouds for prefabricated steel structures in construction scenarios, effectively solving the problems of low efficiency and high error rates associated with manual operations. By using BIM data to guide model training, the accuracy of distinguishing between target and interfering point clouds is improved, providing reliable input for subsequent accurate matching of BIM models and point cloud data.

[0071] Furthermore, the point cloud segmentation model includes:

[0072] The global feature extraction unit is used to extract global features from 3D point cloud data. The global feature extraction unit performs feature transformation on the input point cloud using a multilayer perceptron, and outputs a feature vector representing the overall shape after max pooling to compress redundant information.

[0073] The Local Feature Extraction Unit (LSI) is used to extract local features from 3D point cloud data. The LSI employs a 3D U-Net encoder-decoder structure, where the 3D U-Net preserves spatial details at different scales through skip connections.

[0074] The cross-attention mechanism unit is used to fuse global and local features through cross-attention to obtain fused features. The cross-attention mechanism achieves adaptive feature fusion by calculating attention weights.

[0075] The point cloud segmentation unit is used to segment prefabricated steel structure samples based on fused features. The point cloud segmentation unit performs pooling operations on the fused features in both spatial and channel dimensions, enhances feature contrast through element-wise multiplication and subtraction operations, and finally outputs the segmentation result through a multilayer perceptron.

[0076] Traditional point cloud segmentation methods often employ single feature extraction approaches. This method utilizes a cross-attention mechanism to dynamically fuse global and local features, automatically adjusting feature contribution based on the characteristics of different point cloud regions. For example, it enhances global shape features in the main steel beam region and strengthens local geometric features in bolt connection areas, thereby improving segmentation accuracy in complex construction scenarios. This method achieves efficient differentiation between steel structures and interfering point clouds, solving the problem of low efficiency in manual screening. The cross-attention mechanism effectively integrates the advantages of features at different scales, avoiding missegmentation caused by single feature extraction.

[0077] Furthermore, the process of the global feature extraction unit extracting global features from the 3D point cloud data of any assembled monolithic steel structure sample includes the following steps S210 to S230:

[0078] Step S210: Extract the first global feature from the 3D point cloud data based on the multilayer perceptron.

[0079] Step S220: Perform max pooling on the first global feature to obtain the second global feature.

[0080] Step S230: Extract global features from the second global features based on the multilayer perceptron.

[0081] The 3D point cloud data is first input into a multilayer perceptron for preliminary feature transformation. Through the combination of linear mapping and nonlinear activation functions in the fully connected layer, the basic patterns of geometric shape and spatial distribution in the point cloud are captured to generate the first global features.

[0082] Then, max pooling is performed on the first global feature to extract the maximum value in each feature channel, eliminate redundant information, and generate a second global feature with reduced dimensions.

[0083] The second global feature is then input into the multilayer perceptron for higher-order feature extraction. Through deeper adjustment of weight parameters, a global feature that can characterize the overall form of the assembled steel structure is finally generated.

[0084] This method achieves a progressive deepening of features from low-order geometric features to high-order semantic features through alternating processing of two multilayer perceptrons and pooling operations. At the same time, the pooling operation effectively reduces computational complexity, making the feature extraction process both comprehensive and efficient. It effectively extracts global features containing multi-level geometric information, providing highly discriminative feature representations for subsequent point cloud segmentation, thereby improving the separation accuracy of the steel structure body and the interfering point cloud.

[0085] Furthermore, the local feature extraction unit is a 3D U-Net.

[0086] Specifically, in the 3D point cloud segmentation process of a construction scene, the point cloud data is input into the encoder stage of 3D U-Net, for example, through four layers of 3D convolutional operations, with each convolutional kernel size set to 3×3×3 and a stride of 1. The pooling layers use max pooling to progressively reduce the feature map size to 1 / 8 of the original input. In the decoder stage, deconvolutional layers gradually restore the feature map to its original resolution. After each deconvolution operation, the feature map is concatenated with the corresponding output layer of the encoder, enhancing the local features of minute structures. The final 3D convolutional layer outputs the local feature vector for each point cloud.

[0087] Furthermore, the process by which the cross-attention mechanism unit extracts fused features from global and local features includes the following steps S310 to S350:

[0088] Step S310: Perform average pooling on the global features to obtain the first global intermediate features; perform average pooling on the local features to obtain the first local intermediate features;

[0089] Step S320: After forward serialization of the first global intermediate feature, it is mapped to the Q function; and after forward serialization of the first local intermediate feature, it is mapped to the K function and the V function.

[0090] Step S330: Query the K function based on the Q function to obtain the attention score, and then perform Softmax activation on the attention score to obtain the first attention feature;

[0091] Step S340: Multiply the first attention feature by the V function pixel by pixel to obtain the second attention feature;

[0092] Step S350: After deserializing the second attention feature, the fused feature is obtained.

[0093] Specifically, during the fusion process, global features undergo average pooling to reduce dimensionality, forming the first global intermediate feature; local features undergo the same operation to form the first local intermediate feature. Both are then converted into vector form through forward serialization and mapped to Q, K, and V functions. An attention score is obtained by calculating the dot product of Q and K, which is then normalized using Softmax and multiplied by the V function in a weighted manner to generate the second attention feature. This feature is then deserialized to restore its spatial structure, forming the fused feature while preserving both global contextual information and local detail features.

[0094] This method dynamically calculates the association weights between global and local features through a cross-attention mechanism, enabling the fusion process to adaptively focus on key areas and effectively solve the problem of interference from point clouds in complex construction scenarios. This method significantly improves the ability to distinguish between prefabricated steel structures and interfering point clouds, suppressing irrelevant noise while preserving complete structural features, ensuring a high degree of matching between the segmentation results and the BIM model, and providing reliable technical support for automated 3D point cloud processing.

[0095] Furthermore, the process of segmenting the assembled monolithic steel structure sample from the fused features by the point cloud segmentation unit includes the following steps S410 to S490:

[0096] Step S410: Perform max pooling operation on the fused features in the spatial dimension to obtain the first intermediate feature.

[0097] Step S420: Perform max pooling on the channel dimension of the fused features to obtain the second intermediate features.

[0098] Step S430: Perform average pooling operation on the spatial dimension of the fused features to obtain the third intermediate feature;

[0099] Step S440: Perform average pooling on the channel dimension of the fused features to obtain the fourth intermediate feature.

[0100] Step S450: Multiply the first feature and the second feature pixel by pixel to obtain the fifth intermediate feature.

[0101] Step S460: Multiply the third feature and the fourth feature pixel by pixel to obtain the sixth intermediate feature.

[0102] Step S470: Add the fused feature and the fifth intermediate feature element by element to obtain the seventh intermediate feature.

[0103] Step S480: Subtract the seventh intermediate feature from the sixth intermediate feature element by element to obtain the eighth intermediate feature.

[0104] Step S490: Segment the assembled monolithic steel structure sample from the eighth intermediate feature according to the segmentation layer.

[0105] Specifically, by performing max pooling and average pooling on the fused features in both the spatial and channel dimensions, the spatial distribution characteristics and channel response patterns of the steel structure point cloud can be captured simultaneously. Multiplying the spatial max pooling features and channel max pooling features pixel-by-pixel highlights the salient areas of the steel structure in its spatial distribution and their corresponding key channel features. Multiplying the spatial average pooling features and channel average pooling features pixel-by-pixel preserves the overall morphological characteristics of the steel structure and its global channel response patterns. By superimposing the fused features with the spatial-channel cross features and then performing a difference operation with the global-local cross features, discriminative edge features are preserved while smoothing features are eliminated, effectively suppressing dynamic interference in the construction scene.

[0106] This method employs dual-path pooling operations across spatial and channel dimensions, combined with a cross-multiplication feature enhancement mechanism, to enhance the feature differences between the steel structure and background noise while preserving local geometric details. It addresses the challenge of automatically separating steel structures from dynamic interference point clouds in construction scenario point cloud data. By eliminating interference point clouds generated by temporary material stacking through multi-dimensional feature cross-calculation, it ensures accurate matching between the segmented point cloud data and the BIM model. This method can replace traditional manual annotation, avoiding component missegmentation due to visual fatigue, such as misidentifying tower crane cantilever point clouds as steel structure components, thereby improving the efficiency of quality control in prefabricated steel structure construction.

[0107] Furthermore, the segmentation layer is a multilayer perceptron.

[0108] Specifically, in the point cloud segmentation stage, the fused features are input into the multilayer perceptron, and the output value is transformed into a probability distribution through the Softmax function. The category with the highest probability is selected as the segmentation result for each point.

[0109] This method automatically learns the complex relationships between features using a multilayer perceptron, eliminating the need for manual feature engineering. It can also directly process high-dimensional point cloud features, significantly improving segmentation accuracy and efficiency. This method achieves automated and accurate segmentation of 3D point cloud data in construction scenarios, overcoming the low efficiency and error-prone nature of manual removal of interfering point clouds. Furthermore, it can adapt to scenarios where target structures and interfering point cloud features are mixed in complex construction environments, effectively improving the reliability of point cloud segmentation for BIM-based prefabricated steel structures.

[0110] Reference Figure 2 One embodiment provides a BIM-based three-dimensional point cloud segmentation system for prefabricated monolithic steel structures. The system includes:

[0111] The 3D point cloud acquisition module 1100 is used to extract 3D point cloud data of multiple prefabricated steel structure samples in the construction scene based on BIM.

[0112] The 3D point cloud annotation module 1200 is used to annotate 3D point clouds related to the assembled monolithic steel structure sample and 3D point clouds unrelated to the assembled monolithic steel structure sample in 3D point cloud data.

[0113] The segmentation model training module 1300 is used to train a point cloud segmentation model using labeled 3D point cloud data to obtain a trained point cloud segmentation model.

[0114] The 3D point cloud segmentation module 1400 is used to segment the target assembled steel structure from the 3D point cloud corresponding to the construction scene of the target assembled steel structure based on the trained point cloud segmentation model.

[0115] It should be noted that the BIM-based three-dimensional point cloud segmentation system for prefabricated steel structures provided in this embodiment is based on the same inventive concept as the aforementioned BIM-based three-dimensional point cloud segmentation method for prefabricated steel structures. Therefore, the relevant content of the aforementioned BIM-based three-dimensional point cloud segmentation method for prefabricated steel structures is also applicable to the content of the BIM-based three-dimensional point cloud segmentation system for prefabricated steel structures. Therefore, it will not be repeated here.

[0116] Existing methods rely on manual point cloud screening, which is inefficient due to operator experience and struggles to handle large-scale, complex scenarios. This system combines BIM prior knowledge with a deep learning model to transform manually labeled data into automated segmentation capabilities, reducing reliance on manual intervention. Simultaneously, the model adapts to the diversity of interfering point clouds in different construction scenarios, improving segmentation robustness through feature learning. This system achieves automated segmentation of point clouds for prefabricated steel structures in construction scenarios, effectively solving the problems of low efficiency and high error rates associated with manual operations. By using BIM data to guide model training, the system improves the accuracy of distinguishing between target and interfering point clouds, providing reliable input for subsequent accurate matching of BIM models and point cloud data.

[0117] like Figure 3 This application also provides an electronic device, which includes:

[0118] At least one memory;

[0119] At least one processor;

[0120] At least one program;

[0121] The program is stored in memory, and the processor executes at least one program to implement the above-described BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures.

[0122] This electronic device can be any smart terminal, including mobile phones, tablets, personal digital assistants (PDAs), and in-vehicle computers.

[0123] The electronic devices according to embodiments of this application will now be described in detail.

[0124] The processor 1600 can be implemented using a general-purpose central processing unit (CPU), microprocessor, application specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0125] The memory 1700 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 1700 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1700 and is called and executed by the processor 1600 to implement the BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures according to the embodiments of this application.

[0126] The input / output interface 1800 is used to implement information input and output.

[0127] The communication interface 1900 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0128] Bus 2000 transmits information between various components of the device (e.g., processor 1600, memory 1700, input / output interface 1800, and communication interface 1900);

[0129] The processor 1600, memory 1700, input / output interface 1800 and communication interface 1900 are connected to each other within the device via bus 2000.

[0130] This application also provides a storage medium, which is a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the above-described BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures.

[0131] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0132] The embodiments described in this application are intended to more clearly illustrate the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will know that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0133] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0134] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0135] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0136] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0137] It should be understood that the data used in this way can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than those illustrated or described herein. Furthermore, the terms “comprising” and “having”, and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or apparatus.

[0138] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0139] "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0140] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0141] Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0142] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0143] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0144] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0145] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0146] The above is a detailed description of the preferred embodiments of this application. However, the embodiments of this application are not limited to the above-described implementation methods. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the embodiments of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of the embodiments of this application.

Claims

1. A BIM-based method for segmenting three-dimensional point clouds of prefabricated monolithic steel structures, characterized in that, The method includes: Based on BIM, three-dimensional point cloud data of multiple prefabricated steel structure samples were extracted in the construction scenario. The three-dimensional point cloud data includes the three-dimensional point cloud related to the assembled monolithic steel structure sample and the three-dimensional point cloud unrelated to the assembled monolithic steel structure sample. A point cloud segmentation model is trained using the labeled 3D point cloud data to obtain a trained point cloud segmentation model; the point cloud segmentation model includes: A global feature extraction unit is used to extract global features from the three-dimensional point cloud data. A local feature extraction unit is used to extract local features from the three-dimensional point cloud data. A cross-attention mechanism unit is used to perform cross-attention fusion of the global features and the local features to obtain fused features; the process of the cross-attention mechanism unit extracting the fused features from the global features and the local features includes: The global features are subjected to average pooling to obtain a first global intermediate feature; the local features are subjected to average pooling to obtain a first local intermediate feature; the first global intermediate feature is forward serialized and mapped to a Q function, and the first local intermediate feature is forward serialized and mapped to a K function and a V function; the K function is queried based on the Q function to obtain an attention score, and the attention score is activated with Softmax to obtain a first attention feature; the first attention feature is multiplied pixel-by-pixel by the V function to obtain a second attention feature; the second attention feature is deserialized to obtain a fused feature; A point cloud segmentation unit is used to segment prefabricated steel structure samples based on the fusion features; the process by which the point cloud segmentation unit segments prefabricated steel structure samples from the fusion features includes: Perform max pooling on the fused features in the spatial dimension to obtain the first intermediate feature; The fused feature is subjected to max pooling in the channel dimension to obtain the second intermediate feature; The fused features are subjected to average pooling in the spatial dimension to obtain the third intermediate feature; The fused feature is subjected to average pooling along the channel dimension to obtain the fourth intermediate feature; The first intermediate feature and the second intermediate feature are multiplied pixel by pixel to obtain the fifth intermediate feature; The third intermediate feature and the fourth intermediate feature are multiplied pixel by pixel to obtain the sixth intermediate feature; The fused feature and the fifth intermediate feature are added element by element to obtain the seventh intermediate feature; Subtract the seventh intermediate feature from the sixth intermediate feature element by element to obtain the eighth intermediate feature; The assembled monolithic steel structure sample is segmented from the eighth intermediate feature based on the segmentation layer; Based on the trained point cloud segmentation model, the target assembled monolithic steel structure is segmented from the 3D point cloud corresponding to the construction scene of the target assembled monolithic steel structure.

2. The BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures according to claim 1, characterized in that, The process by which the global feature extraction unit extracts global features from the three-dimensional point cloud data of any one of the assembled monolithic steel structure samples includes: The first global feature extracted from 3D point cloud data based on a multilayer perceptron; The first global feature is subjected to max pooling to obtain the second global feature; Global features are extracted from the second global features using a multilayer perceptron.

3. The BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures according to claim 1, characterized in that, The local feature extraction unit is a 3D U-Net.

4. The BIM-based three-dimensional point cloud segmentation method for prefabricated monolithic steel structures according to claim 1, characterized in that, The segmentation layer is a multilayer perceptron.

5. A BIM-based three-dimensional point cloud segmentation system for prefabricated monolithic steel structures, characterized in that, The system includes: The 3D point cloud acquisition module is used to extract 3D point cloud data of multiple prefabricated steel structure samples in the construction scene based on BIM. A 3D point cloud annotation module is used to annotate the 3D point cloud data with respect to the assembled monolithic steel structure sample and with respect to the 3D point cloud data with respect to the assembled monolithic steel structure sample. A segmentation model training module is used to train a point cloud segmentation model using the labeled 3D point cloud data to obtain a trained point cloud segmentation model; the point cloud segmentation model includes: A global feature extraction unit is used to extract global features from the three-dimensional point cloud data. A local feature extraction unit is used to extract local features from the three-dimensional point cloud data. A cross-attention mechanism unit is used to perform cross-attention fusion of the global features and the local features to obtain fused features; the process of the cross-attention mechanism unit extracting the fused features from the global features and the local features includes: The global features are subjected to average pooling to obtain a first global intermediate feature; the local features are subjected to average pooling to obtain a first local intermediate feature; the first global intermediate feature is forward serialized and mapped to a Q function, and the first local intermediate feature is forward serialized and mapped to a K function and a V function; the K function is queried based on the Q function to obtain an attention score, and the attention score is activated with Softmax to obtain a first attention feature; the first attention feature is multiplied pixel-by-pixel by the V function to obtain a second attention feature; the second attention feature is deserialized to obtain a fused feature; A point cloud segmentation unit is used to segment prefabricated steel structure samples based on the fusion features; the process by which the point cloud segmentation unit segments prefabricated steel structure samples from the fusion features includes: Perform max pooling on the fused features in the spatial dimension to obtain the first intermediate feature; The fused feature is subjected to max pooling in the channel dimension to obtain the second intermediate feature; The fused features are subjected to average pooling in the spatial dimension to obtain the third intermediate feature; The fused feature is subjected to average pooling along the channel dimension to obtain the fourth intermediate feature; The first intermediate feature and the second intermediate feature are multiplied pixel by pixel to obtain the fifth intermediate feature; The third intermediate feature and the fourth intermediate feature are multiplied pixel by pixel to obtain the sixth intermediate feature; The fused feature and the fifth intermediate feature are added element by element to obtain the seventh intermediate feature; Subtract the seventh intermediate feature from the sixth intermediate feature element by element to obtain the eighth intermediate feature; The assembled monolithic steel structure sample is segmented from the eighth intermediate feature based on the segmentation layer; The 3D point cloud segmentation module is used to segment the target assembled steel structure from the 3D point cloud corresponding to the construction scene of the target assembled steel structure based on the trained point cloud segmentation model.

6. An electronic device, characterized in that, include: At least one control processor and a memory for communicatively connecting to the at least one control processor; The memory stores instructions that can be executed by the at least one control processor, which, when executed by the at least one control processor, enables the at least one control processor to perform the three-dimensional point cloud segmentation method for BIM-based prefabricated monolithic steel structures as described in any one of claims 1 to 5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions for causing a computer to execute the three-dimensional point cloud segmentation method for BIM-based prefabricated monolithic steel structures as described in any one of claims 1 to 5.

Images