A method, device and medium for monitoring the construction of ultra-large span high-precision special-shaped structures

Abstract

The present invention discloses a method, device and medium for monitoring the construction of ultra-large span high-precision special-shaped structures, and relates to the field of image processing technology. A construction monitoring and optimization process based on a construction data model library and real-time image data is proposed. By extracting building skeleton point cloud fragments from the real-time image data, the actual situation of the construction site can be monitored more accurately to ensure that the construction accuracy, progress and quality meet expectations. By using a feature extraction backbone network, in-depth feature extraction can be performed on the building skeleton point cloud fragments. The building skeleton point cloud fragments are fused based on the extracted local features to obtain more comprehensive overall features, thereby enhancing the extraction accuracy of the building structure. The building skeleton generated according to the overall features is compared with the construction data model library, which can guide the rod processing for a second time, improve the practicality and accuracy of the model library, and improve the accuracy, efficiency and quality of the building construction.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method, a device and a medium for monitoring the construction of an ultra-large span high-precision special-shaped structure. Background Art

[0002] The current large-scale wind tunnel construction technology is difficult to obtain. The construction process of this type of structure is complex and requires extremely high precision. Therefore, in the design stage, it is necessary to explore a construction plan that can ensure the effective realization of the wind tunnel test function, the realization of the process requirements, and the safety and reliability of the structural design. On this basis, it is necessary to initially refine it to guide the subsequent preparation of detailed construction plans. Therefore, it is necessary to first analyze and calculate the construction model based on digital virtual construction, and guide the construction of the upper grid and upper plate based on the construction model. After the completion of this part of the construction, the data information of the grid positioning is obtained by using three-dimensional scanning technology to guide the processing and installation of the lower suspension system to improve the construction accuracy and efficiency. In the process of building this high-precision special-shaped model, the traditional monitoring method for the construction of ultra-large span high-precision special-shaped structures is gradually unable to meet the needs. In order to improve the monitoring efficiency, the monitoring method is changing from contact measurement to non-contact measurement, but the detection efficiency of traditional detection methods is low and the detection effect is poor. Specifically, there are the following problems:

[0003] Due to the structural complexity of the wind tunnel, the structure has complex surfaces that cannot be accurately described by analytical geometry and cannot be controlled with high precision. In the construction of a high-precision solution model for deformation monitoring based on the GPS / BDS combination, the existing model monitoring process cannot accurately identify the skeleton contour structure, and thus cannot accurately generate the skeleton structure, resulting in inaccurate optimization of the later construction data model library, which will affect the stability and solution accuracy of the building structure model. In addition, the traditional data fusion method is simple and cannot dynamically reflect the deformation characteristics of the building structure, resulting in insufficient data use, low deformation feature accuracy, and real-time problems, which makes it impossible to provide accurate guidance during secondary processing and installation. Summary of the invention

[0004] The technical problem to be solved by the present invention is that the existing model monitoring accuracy is low and the extraction effect of the building skeleton is poor. The purpose is to provide a method, equipment and medium for monitoring the construction of ultra-large span high-precision special-shaped structures, and propose a construction monitoring and optimization process based on a construction data model library and real-time image data, that is, how to efficiently create a model library containing construction design data, and how to optimize and update these models according to actual construction conditions. By extracting building skeleton point cloud fragments from real-time image data, the actual situation of the construction site can be more accurately monitored to ensure that the construction accuracy, progress and quality meet expectations. Using a feature extraction backbone network, the construction In-depth feature extraction is performed on the building skeleton point cloud fragments, including local features and overall features. The building skeleton point cloud fragments are fused based on the extracted local features to obtain more comprehensive overall features, which enhances the extraction accuracy of the building structure. The building skeleton generated according to the overall features is compared with the construction data model library to improve the practicality and accuracy of the model library. By combining the construction data model library and real-time image data processing technology, the accuracy, efficiency and quality of building construction can be improved. Through secondary construction guidance, the digitization and intelligence of the construction process are promoted, which provides strong technical support for construction management and realizes efficient real-time monitoring and management of the construction process.

[0005] The present invention is achieved through the following technical solutions:

[0006] The first aspect of the present invention provides a method for monitoring the construction of an ultra-large span high-precision special-shaped structure, comprising the following specific steps:

[0007] Generate a construction data model library, perform construction based on the construction data model library, and obtain real-time image data according to the actual construction situation;

[0008] Preprocess the live image data to obtain the RGB image of the physical construction structure;

[0009] Convert the RGB image into HSL three-channel, divide the hue thresholds of the three channels respectively, and obtain the building skeleton image according to the hue thresholds;

[0010] Binarize the building skeleton image to obtain a foreground binary image;

[0011] Use skeleton extraction algorithm to extract the outline of building skeleton from the foreground binary image;

[0012] Combine the contour information of the two-dimensional plane with the three-dimensional point cloud data to generate a point cloud distribution matrix;

[0013] According to the point cloud distribution matrix and the outline information of the building skeleton, the building skeleton point cloud fragments are extracted;

[0014] Construct a feature extraction backbone network to extract the features of the building skeleton point cloud fragments and obtain the local features of the current building skeleton point cloud fragments;

[0015] Based on the extracted local features, the building skeleton point cloud fragments are fused to obtain the overall features;

[0016] Generate the building skeleton according to the overall characteristics;

[0017] Based on the comparison between the building skeleton and the construction data model library, secondary guidance on processing and installation is provided, and the construction data model library is updated and iterated.

[0018] Furthermore, the use of a skeleton extraction algorithm to extract the outline of the building skeleton from the foreground binary image specifically includes:

[0019] Iterate the erosion and opening operations on the foreground binary image. In each iteration:

[0020] Perform an erosion operation on the image and then perform an opening operation;

[0021] Get the difference between the erosion result and the opening result, and add the difference to the foreground binary image until the foreground binary image is completely eroded into an empty set. Output a binary image with the same size as the building skeleton image but containing only a single-pixel width skeleton.

[0022] Furthermore, when extracting the building skeleton point cloud fragment, the method further includes:

[0023] Find the point cloud distortion data, identify and mark the distorted parts in the point cloud data, including: obtaining the input image, fitting the input image multiple times, generating a segmented polynomial curve, finding the point coordinates of the hollow part according to the polynomial curve, and drawing the incomplete parts of the building skeleton point cloud fragment;

[0024] The distorted part of the mark is removed to obtain the dedistorted point cloud data;

[0025] The dedistorted point cloud data is adjusted through multiple optimizations, the coordinates of the measuring station are corrected using nonlinear optimization methods, and the building skeleton point cloud fragments are output.

[0026] Furthermore, the construction of the feature extraction backbone network to extract the local features of the current building skeleton point cloud fragment specifically includes:

[0027] Based on the U-net network, a multi-scale feature combination module and a multi-scale feature enhancement module are introduced to extract multi-scale features of the building skeleton point cloud fragments;

[0028] A dual attention module is introduced to learn the feature importance of channels and spatial positions, process the differences between different deep features, and output the local features of the current building skeleton point cloud fragment.

[0029] Furthermore, the multi-scale feature combination module includes a multi-scale input and a multi-scale output. By adding a multi-scale enhancement module between the encoder and the decoder, context information of different scales is captured to obtain a wider receptive field, specifically including:

[0030] Project the point cloud onto the image plane and stretch it in the vertical direction to compensate for the sparsity of the point cloud;

[0031] A two-level residual structure attention fusion module is constructed, and a parallel attention mechanism module is used to replace direct channel stacking. The building skeleton point cloud fragments and the physical construction structure image data are fused to obtain features under multiple receptive fields.

[0032] Furthermore, the fusing of the building skeleton point cloud fragments based on the extracted local features specifically includes:

[0033] Extract the image coordinates in the building skeleton point cloud fragment, and extract the corresponding point cloud coordinates in the image coordinates;

[0034] According to the point cloud features corresponding to the image coordinates, the semantically aligned features are obtained by learning the offset of the point cloud features;

[0035] Cross-attention is used to fuse the semantically aligned features to obtain the overall features.

[0036] Furthermore, the process of generating the building skeleton specifically includes:

[0037] Generate the initial building skeleton based on the overall characteristics;

[0038] Perform median downsampling on the initial building skeleton to obtain the source point set;

[0039] Perform voxel downsampling on the source point set to obtain an initial contraction point set;

[0040] For each point in the initial contraction point set, iterative contraction is performed in its local neighborhood. A new projection point set is obtained in each iteration. The iteration is performed until the projection point set is redistributed to the center of the source point set, and the final building skeleton is output.

[0041] Furthermore, the comparison based on the building skeleton and the construction data model library specifically includes:

[0042] Randomly sample sample data points in the building point cloud model and the construction data model for fitting, and determine the differences between the building point cloud model and the construction data model;

[0043] According to the preset fitting threshold, the data points with fitting errors within the threshold range are regarded as reference points;

[0044] Calculate the difference between the reference point set and all sampling point sets to obtain the non-reference point set;

[0045] Based on the set of non-reference points, the deviations in the construction data model are eliminated, and the construction data model is iteratively updated. In each iteration, the model with the largest number of reference points is selected as the optimal output.

[0046] A second aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a method for monitoring the construction of ultra-large span, high-precision, special-shaped structures is implemented.

[0047] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for monitoring the construction of ultra-large-span, high-precision, special-shaped structures.

[0048] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0049] By extracting building skeleton point cloud fragments from live image data, the actual situation of the construction site can be monitored more accurately to ensure that the construction accuracy, progress and quality meet expectations. By using the feature extraction backbone network, in-depth feature extraction of building skeleton point cloud fragments can be performed, including local features and overall features. By fusing the building skeleton point cloud fragments based on the extracted local features, more comprehensive overall features can be obtained, which enhances the extraction accuracy of the building structure. The building skeleton generated according to the overall features is compared with the construction data model library to improve the practicality and accuracy of the model library. By combining the construction data model library and live image data processing technology, the accuracy, efficiency and quality of building construction can be improved. Through secondary construction guidance, the digitization and intelligence of the construction process are promoted, providing strong technical support for construction management and realizing efficient real-time monitoring and management of the construction process. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0051] Figure 1 A monitoring process for model construction in an embodiment of the present invention;

[0052] Figure 2 It is a guiding construction process in the embodiment of the present invention;

[0053] Figure 3 This is the construction process in the embodiment of the present invention. DETAILED DESCRIPTION

[0054] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0055] As a possible implementation, Figure 1 As shown, the present embodiment provides a method for monitoring the construction of ultra-large span high-precision special-shaped structures, including the following specific steps: generating a construction data model library, performing construction based on the construction data model library, and obtaining real-time image data according to the actual construction situation; preprocessing the real-time image data to obtain an RGB image of the physical construction structure; converting the RGB image into HSL three-channels, dividing the hue thresholds of the three channels respectively, and obtaining a building skeleton image according to the hue threshold screening; binarizing the building skeleton image to obtain a foreground binary image; extracting the outline of the building skeleton from the foreground binary image using a skeleton extraction algorithm; combining the outline information of the two-dimensional plane with the three-dimensional point cloud data to generate a point cloud distribution matrix; extracting the building skeleton point cloud fragments according to the point cloud distribution matrix and the outline information of the building skeleton; constructing a feature extraction backbone network, extracting the features of the building skeleton from the building skeleton point cloud fragments, and obtaining the local features of the current building skeleton point cloud fragments; fusing the building skeleton point cloud fragments based on the extracted local features to obtain the overall features; generating the building skeleton according to the overall features; comparing the building skeleton with the construction data model library, secondary guiding processing and installation, and updating the iterative construction data model library. This embodiment can more accurately monitor the actual situation of the construction site by extracting building skeleton point cloud fragments from real-time image data to ensure that the construction accuracy, progress and quality meet expectations. By using the feature extraction backbone network, the building skeleton point cloud fragments can be deeply extracted, including local features and overall features. The building skeleton point cloud fragments are fused based on the extracted local features to obtain more comprehensive overall features, thereby enhancing the extraction accuracy of the building structure. The building skeleton generated according to the overall features is compared with the construction data model library to improve the practicality and accuracy of the model library. By combining the construction data model library and the real-time image data processing technology, the lower suspension rod system is subjected to secondary precise cutting and processing to improve the accuracy, efficiency and quality of the building construction. Through secondary construction guidance, the digitization and intelligence of the construction process are promoted, and strong technical support is provided for construction management, thereby realizing efficient feedback iteration, real-time monitoring and management of the construction process.

[0056] In some possible implementations, the generated construction data model library may be a model constructed according to construction requirements, or may be an existing constructed model.

[0057] In some possible implementations, extracting the building skeleton from the point cloud data specifically includes: preprocessing the live image data to obtain an RGB image of the physical construction structure; converting the RGB image into three HSL (hue, saturation, brightness) channels, dividing the hue thresholds of the three channels respectively, and obtaining the building skeleton image according to the hue thresholds. The accuracy of identifying the building skeleton can be improved by screening the hue thresholds and the characteristics of the HSL color space, especially when the contrast between the building color and the background color is not high; binarizing the building skeleton image, that is, highlighting the building skeleton as the foreground and eliminating the background, thereby obtaining a foreground binary image; extracting the outline of the building skeleton from the binarized image using a skeleton extraction algorithm; combining two-dimensional image processing and three-dimensional point cloud data, that is, combining the outline information of the two-dimensional plane with the three-dimensional point cloud data to generate a point cloud distribution matrix; according to the point cloud distribution matrix and the outline information of the building skeleton, accurately separating the part related to the building skeleton from the overall point cloud data to obtain the building skeleton point cloud fragment. The extracted building skeleton point cloud fragment can be used to support complex structural analysis.

[0058] In some possible implementations, when using a skeleton extraction algorithm to extract the outline of a building skeleton from a binary image, it is necessary to iteratively perform erosion and opening operations on the foreground binary image, and in each iteration, it is necessary to first perform an erosion operation on the image and then perform an opening operation to obtain the difference between the erosion result and the opening result, and accumulate the difference to the foreground binary image until the foreground binary image is completely eroded into an empty set, and output a binary image of the same size as the building skeleton image but containing only a skeleton of a single pixel width.

[0059] In some possible implementations, when extracting the building skeleton point cloud fragment, it also includes: searching for point cloud distortion data, identifying and marking the distorted part in the point cloud data, including: obtaining an input image, fitting the input image multiple times, fitting using the RANSAC algorithm for robust fitting, generating a segmented polynomial curve after fitting, obtaining the point coordinates of the hollow part according to the polynomial curve, using the principal component analysis (PCA) to analyze the directional consistency of the skeleton points, thereby obtaining the skeleton curve segment, and then merging through breadth-first search (BFS) to obtain the merged curve segment, using Laplace smoothing to optimize the merged curve segment and obtain the final curve skeleton, thereby drawing the incomplete part of the building skeleton point cloud fragment; the coordinate value after removing the distortion is basically consistent with the true value, the de-distorted point cloud data is adjusted through multiple optimizations, and the coordinates of the measuring station are corrected using a nonlinear optimization method. After the above steps are processed, a building skeleton point cloud fragment with high accuracy and completeness can be obtained.

[0060] In some possible implementations, a feature extraction backbone network is constructed to extract local features of the current building skeleton point cloud fragment, specifically including: based on the U-net network, a multi-scale feature combination module and a multi-scale feature enhancement module are introduced to extract multi-scale features of the building skeleton point cloud fragment; the multi-scale feature combination module includes multi-scale input and multi-scale output, and a multi-scale enhancement module is added between the encoder and the decoder to capture contextual information of different scales to obtain a wider receptive field, specifically including: projecting the point cloud onto the image plane and stretching it in the vertical direction to compensate for the sparsity of the point cloud; constructing an attention fusion module of a two-level residual structure, using a parallel attention mechanism module instead of direct channel stacking, fusing the building skeleton point cloud fragment and the physical construction structure image data, and obtaining features under multiple receptive fields; introducing a dual attention module to learn the feature importance of channels and spatial positions, processing the differences between different depth features, and outputting the local features of the current building skeleton point cloud fragment. In the above content, a multi-scale input is constructed in the encoding part of the network, and after convolution processing of input images of different scales, the superimposed input is input into the decoder corresponding to the network to enhance the acquisition of multi-scale information and realize the fusion of receptive fields at different levels. In the decoding part of the model, a multi-scale output is constructed, and the feature map of each scale is output and combined through convolution and bilinear interpolation upsampling, which effectively supplements the information of feature maps at different levels and realizes the accurate extraction of multi-scale buildings. A multi-scale enhancement module is added between the encoder and decoder of the model. By extracting features in multiple different hole convolutions, contextual information of different scales is captured to obtain a wider receptive field and improve the network's perception of multi-scale features. Among them, the point cloud is projected onto the image plane and stretched in the vertical direction to compensate for the sparsity of the point cloud. It helps to improve the density and quality of point cloud data and provide richer information for subsequent feature extraction. A two-level residual structure attention fusion module is constructed, and a parallel attention mechanism module is used to replace direct channel stacking to fuse the building skeleton point cloud fragments and the physical construction structure image data to obtain features under multiple receptive fields. The introduction of the dual attention module can adjust the weight of the feature according to the importance of different channels and different spaces in the feature map, helping the network to more accurately capture the features of the building and suppress background interference, thereby improving the accuracy of building extraction. According to the feature importance of learning channels and spatial positions by the dual attention module, the differences between different depth features are processed, and the local features of the current building skeleton point cloud fragment are output. The above feature extraction method can effectively extract multi-scale features of the building skeleton point cloud fragment, and use the attention mechanism to improve the expression ability of the features, and finally output more accurate building skeleton point cloud fragments.

[0061] In some possible implementations, the building skeleton point cloud segments are fused based on the extracted local features, specifically including:

[0062] By projecting the point cloud onto the image plane, the corresponding pixel position of each point cloud in the image is obtained, that is, the image coordinates in the building skeleton point cloud fragment are extracted, and the corresponding point cloud coordinates in the image coordinates are extracted; after obtaining the image coordinates, the original point cloud coordinates corresponding to these coordinates need to be extracted in reverse. This usually involves the mapping relationship between the point cloud and the image. According to the point cloud features corresponding to the image coordinates, the point cloud features can be semantically aligned by calculating the offset of the point cloud features, so that the point cloud features and the image features are more semantically consistent, and then the cross-attention mechanism is used to fuse the semantically aligned features. The cross-attention mechanism can handle the dependency between two different sequences. For example, in multimodal tasks, the effective fusion of image features and point cloud features can help the model better understand the relationship between different modalities, thereby extracting a more comprehensive feature representation.

[0063] In some possible implementations, the initial building skeleton is optimized, which is usually used to extract the center line or skeleton of the structure from the three-dimensional data. In order to reduce the complexity of the data set, the data is smoothed by applying a median filter and the resolution of the data is reduced to effectively remove noise while retaining edge information. The initial building skeleton needs to be median downsampled to obtain a source point set, and the source point set is voxel downsampled to obtain an initial contraction point set. Voxel downsampling reduces the number of voxels in the three-dimensional data set, thereby further simplifying the model. On this basis, the skeleton is further refined, and each point in the initial contraction point set is iteratively contracted within its local neighborhood. A new projection point set is obtained in each iteration, and it is iterated until the projection point set is redistributed to the center of the source point set, and the final building skeleton is output.

[0064] In some possible implementations, the comparison is performed based on the building skeleton and the construction data model library, specifically including:

[0065] Randomly sample sample data points from the building point cloud model and the construction data model, and use a fitting algorithm to fit the randomly sampled sample data points from the building point cloud model and the construction data model to determine the difference between the building point cloud model and the construction data model;

[0066] According to the preset fitting threshold, the data points with fitting errors within the threshold range are regarded as reference points. When the positions of these points in the two models are close enough, they can be considered to be matched;

[0067] Calculate the difference between the reference point set and all sampling point sets to obtain the non-reference point set. The non-reference point set has a large difference in position between the two models, which may be caused by construction errors or other factors.

[0068] Based on the non-reference point set, the deviation in the construction data model is eliminated, and the construction data model is iteratively updated. In each iteration, the model with the largest number of reference points is selected as the optimal output. After obtaining the optimal output, it is usually necessary to verify the iteratively updated construction data model to ensure its accuracy and reliability.

[0069] As a possible implementation, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, a method for monitoring the construction of ultra-large span, high-precision, special-shaped structures is implemented.

[0070] As a possible implementation, this embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for monitoring the construction of ultra-large-span, high-precision, special-shaped structures.

[0071] This embodiment provides a specific monitoring process during the construction process. In the construction of the wind tunnel structure of this embodiment, Figure 2 and Figure 3 As shown, this embodiment needs to be based on Figure 2 Guide the construction process and Figure 3 The construction process is constantly optimized, including:

[0072] (1) Construction of foundation or pile foundation engineering;

[0073] (2) Construction of the main structure base plate;

[0074] (3) Construction of vertical structural concrete inner hole wall and rib wall;

[0075] (4) Construction of main steel grid of roof;

[0076] (5) Construction of steel grid upper chord roof concrete slab and its associated metal roof system;

[0077] (6) Construction of the concrete inner protection plate of the lower chord of the steel grid. According to the comparison results in the monitoring method, guide the secondary processing and ensure that the inner protection plate is adjusted horizontally before construction;

[0078] (7) Construction of secondary poured concrete slab for the main base plate.

[0079] The wind tunnel constructed in this embodiment is very different from the general building structure, one of which is the particularity of the main structure of the roof and its internal maintenance structure. When the wind tunnel is in operation, due to the airflow passing through, a smooth, high-precision, and airtight inner wall, namely the internal maintenance structure, is required. In the selection of the tunnel body, the vertical side walls and the bottom plate are all made of reinforced concrete wall panels, which can directly serve as the internal maintenance structure. For the roof, due to many reasons such as large span, heavy load and high precision, a large-span steel structure + hanging steel secondary beam system + reinforced concrete internal maintenance structure system is adopted. The internal maintenance plate is a structural member that is directly in contact with the internal airflow and needs to withstand high pressure and high temperature. Its load action mode also has typical tension and compression cycles, shear cycles, vibration, and fatigue characteristics. Therefore, the entire roof system is not only novel in structure, but also has no design and construction experience to refer to, and the load action is complex and the process requirements are high. How to ensure the safety and high precision of the roof, and optimize the construction sequence of the construction process so that the construction accuracy meets the design and process requirements is the difficulty of design and construction integration. Therefore, when the construction is carried out according to the above construction process, when the upper grid and upper plate construction is completed, it is necessary to adopt the super-large span high-precision special-shaped structure construction monitoring method for real-time monitoring, extract the building skeleton point cloud fragments from the live image data, and use the feature extraction backbone network to perform in-depth feature extraction on the building skeleton point cloud fragments. First, extract the local features, and fuse the building skeleton point cloud fragments based on the extracted local features to obtain more comprehensive overall features, thereby enhancing the extraction accuracy of the building structure. By combining the construction data model library and live image data processing technology, the subsequent lower hanging rod system is accurately cut and processed to improve the accuracy, efficiency and quality of building construction, while promoting the digitization and intelligence of the construction process. By analyzing and calculating the construction process based on digital virtual construction, the accuracy of the target building construction is improved. At the same time, the structure can be continuously monitored and maintained to achieve high accuracy throughout the life cycle, which can effectively control the design quality of the hanging board project and improve the efficiency and quality of the hanging board project construction.

[0080] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for monitoring the construction of ultra-large span, high-precision, special-shaped structures, characterized in that: The specific steps include: Generate a construction data model library, perform construction based on the construction data model library, and obtain real-time image data according to the actual construction situation; Preprocess the live image data to obtain the RGB image of the physical construction structure; Convert the RGB image into HSL three-channel, divide the hue thresholds of the three channels respectively, and obtain the building skeleton image according to the hue thresholds; Binarize the building skeleton image to obtain a foreground binary image; The skeleton extraction algorithm is used to extract the outline of the building skeleton from the foreground binary image, specifically including: iteratively performing corrosion operation and opening operation on the foreground binary image, in each iteration: performing corrosion operation on the image, and then performing opening operation; obtaining the difference between the corrosion result and the opening operation result, and accumulating the difference to the foreground binary image until the foreground binary image is completely corroded into an empty set, and outputting a binary image with the same size as the building skeleton image but containing only a skeleton with a single pixel width; Combine the contour information of the two-dimensional plane with the three-dimensional point cloud data to generate a point cloud distribution matrix; According to the point cloud distribution matrix and the outline information of the building skeleton, the building skeleton point cloud fragments are extracted; Construct a feature extraction backbone network to extract the features of the building skeleton point cloud fragments and obtain the local features of the current building skeleton point cloud fragments, specifically including: introducing a multi-scale feature combination module and a multi-scale feature enhancement module based on the U-net network to extract multi-scale features of the building skeleton point cloud fragments; introducing a dual attention module to learn the feature importance of channels and spatial positions, processing the differences between different depth features, and outputting the local features of the current building skeleton point cloud fragments; Based on the extracted local features, the building skeleton point cloud fragments are fused to obtain the overall features, specifically including: extracting the image coordinates in the building skeleton point cloud fragments, extracting the corresponding point cloud coordinates in the image coordinates; according to the point cloud features corresponding to the image coordinates, by learning the offset of the point cloud features, the semantically aligned features are obtained; cross-attention is used to fuse the semantically aligned features to obtain the overall features; Generate a building skeleton according to the overall features, specifically including: generate an initial building skeleton according to the overall features; perform median downsampling on the initial building skeleton to obtain a source point set; perform voxel downsampling on the source point set to obtain an initial contraction point set; perform iterative contraction on each point in the initial contraction point set within its local neighborhood, obtain a new projection point set in each iteration, iterate until the projection point set is redistributed to the center of the source point set, and output the final building skeleton; Compare the building skeleton and the construction data model library to optimize the construction data model library.

2. The method for monitoring the construction of ultra-large span, high-precision, special-shaped structures according to claim 1 is characterized in that: When extracting the building skeleton point cloud fragment, the method further includes: Find the point cloud distortion data, identify and mark the distorted parts in the point cloud data, including: obtaining the input image, fitting the input image multiple times, generating a segmented polynomial curve, finding the point coordinates of the hollow part according to the polynomial curve, and drawing the incomplete parts of the building skeleton point cloud fragment; The distorted part of the mark is removed to obtain the dedistorted point cloud data; The dedistorted point cloud data is adjusted, the coordinates of the measuring station are corrected using a nonlinear optimization method, and the building skeleton point cloud fragments are output.

3. The method for monitoring the construction of ultra-large span, high-precision, special-shaped structures according to claim 1 is characterized in that: The multi-scale feature combination module includes a multi-scale input and a multi-scale output. By adding a multi-scale enhancement module between the encoder and the decoder, context information of different scales is captured to obtain a wider receptive field, specifically including: Project the point cloud onto the image plane and stretch it in the vertical direction to compensate for the sparsity of the point cloud; A two-level residual structure attention fusion module is constructed, and a parallel attention mechanism module is used to replace direct channel stacking. The building skeleton point cloud fragments and the physical construction structure image data are fused to obtain features under multiple receptive fields.

4. The method for monitoring the construction of ultra-large span, high-precision, special-shaped structures according to claim 1, characterized in that: The comparison based on the building skeleton and the construction data model library specifically includes: Randomly sample sample data points in the building point cloud model and the construction data model for fitting, and determine the differences between the building point cloud model and the construction data model; According to the preset fitting threshold, the data points with fitting errors within the threshold range are regarded as reference points; Calculate the difference between the reference point set and all sampling point sets to obtain the non-reference point set; Based on the set of non-reference points, the deviations in the construction data model are eliminated, and the construction data model is iteratively updated. In each iteration, the model with the largest number of reference points is selected as the optimal output.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the ultra-large span high-precision special-shaped structure construction monitoring method as described in any one of claims 1 to 4 is implemented.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for monitoring the construction of an ultra-large span, high-precision, special-shaped structure as described in any one of claims 1 to 4 is implemented.

Images