Methods, systems, equipment, and media for analyzing project progress based on BIM models and AI video

Abstract

The present invention discloses a method, system, equipment and medium for analyzing engineering progress based on BIM model and AI video, which relates to the technical field of construction engineering management, including collecting regional image data, performing semantic segmentation and environmental modeling. A multi-scale spatial feature network and a dynamic skeleton modeling engine are constructed, and the three-dimensional environmental map, the structurally identifiable area map and the obstacle probability distribution map are input into the multi-scale spatial feature network and the dynamic skeleton modeling engine, and the output performs structural recognition and defect anomaly labeling. The engineering response mechanism is triggered based on the risk event scoring function, and the construction path is remotely and dynamically adjusted. The method described in the present invention realizes high-precision structural recognition of construction site images by combining semantic segmentation with three-dimensional environmental modeling, effectively solving the problems of unclear structural recognition and blurred component boundaries in traditional video surveillance.

Description

Technical Field

[0001] The present invention relates to the technical field of construction project management, and specifically to a method, system, equipment and medium for analyzing project progress based on a BIM model and AI video. Background Art

[0002] With the continuous development of building information modeling technology, its application in the construction industry is becoming increasingly widespread. In particular, BIM models have become an important means of achieving visualization, collaboration, and information management in the entire process of project planning, design, construction, and operation and maintenance. At the same time, the maturity of artificial intelligence and computer vision technologies has also provided a technical foundation for automatic monitoring and intelligent analysis of construction sites, making it possible to shift from traditional manual inspections to intelligent, automated, and data-driven monitoring methods.

[0003] During the construction phase of a project, project progress management, as one of the core elements of project management, plays a key role in ensuring on-schedule project completion, rationally allocating resources, and controlling costs. Current mainstream project progress management methods rely on progress management software such as Project, combined with manual photography and reporting to confirm project milestones and compare plans. However, this approach has significant limitations: On the one hand, manual recording of project progress suffers from high subjectivity, delayed updates, and difficulty acquiring on-site data, seriously impacting the real-time and accuracy of project management and control. On the other hand, although some projects have attempted to use BIM models for progress planning and visual comparison, this process still relies heavily on manual maintenance of BIM progress status and lacks automated data collection and dynamic perception capabilities.

[0004] The existing project progress management methods have the problem that progress data acquisition mainly relies on manual collection and manual entry, lacks intelligence and real-time performance, AI recognition results are difficult to accurately match BIM model components, lack component-level semantic alignment mechanisms, lack engineering response capabilities based on structural recognition and risk scoring mechanisms, and cannot dynamically adjust and optimize decisions based on recognition results. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is: the existing technical methods for project progress management and monitoring have the problems of relying on manual labor for data collection, low progress identification accuracy, and untimely information updates; as well as how to automatically identify the construction structure status based on on-site video data and achieve accurate matching with the BIM model, so as to dynamically analyze the actual construction progress and perform path optimization adjustments.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for analyzing engineering progress based on BIM model and AI video, including collecting regional image data, performing semantic segmentation and environmental modeling.

[0008] Build a multi-scale spatial feature network and dynamic skeleton modeling engine.

[0009] The three-dimensional environment map, structure identifiable area map and obstacle probability distribution map are input into the multi-scale spatial feature network and dynamic skeleton modeling engine, and the output performs structure recognition and defect anomaly labeling.

[0010] The engineering response mechanism is triggered based on the risk event scoring function and the construction path is adjusted remotely and dynamically.

[0011] The three-dimensional environment map includes generating a three-dimensional environment map by modeling the environment and converting the structure boundary information, the location of the blocked area and the construction path.

[0012] The structure-recognizable area map and the obstacle probability distribution map include inputting the collected area image data into a constructed deep convolutional neural network model to obtain the structure-recognizable area map and the obstacle probability distribution map.

[0013] Constructing a multi-scale spatial feature network and a dynamic skeleton modeling engine includes identifying target components and structural unit nodes and outputting a structural map through a multi-scale residual graph neural network and a skeleton fitting engine.

[0014] As a preferred solution of the method for analyzing engineering progress based on BIM model and AI video according to the present invention, the acquisition of regional image data includes performing regional coverage scanning of the target building structure within a set time interval using infrared thermal imaging equipment, visible light high-definition cameras, and lidar sensing components installed at the construction site, and performing unified time series numbering and spatial registration operations on the multimodal images and point cloud data collected in the scanning area.

[0015] The spatial registration operation includes using a sparse feature matching algorithm based on laser SLAM to perform coarse alignment to maintain structural consistency between image frames, performing fine-grained registration based on the point cloud, and binding each frame of data to the building digital map model through timestamps and positioning information to form a perception frame cluster.

[0016] As a preferred solution of the method for analyzing engineering progress based on BIM model and AI video described in the present invention, the execution of semantic segmentation and environmental modeling includes using a deep convolutional neural network model to perform multi-channel feature extraction on the collected image data, constructing a semantic segmentation map, marking structures, personnel, equipment, and abnormal areas as semantic category layers, and using a strategy based on spatial consistency and grayscale voting to perform misclassification correction.

[0017] Based on the segmentation map, the environment is modeled and the structural boundary information, occlusion area location and passage path are used to generate a three-dimensional environment map.

[0018] Generating a three-dimensional environment map includes: constructing the three-dimensional environment map by using an edge surface fitting method based on dense point cloud interpolation.

[0019] Voxel sparsification operation is introduced to compress the data volume, and the output structure can identify the area map and obstacle probability distribution map.

[0020] As a preferred solution of the method for analyzing engineering progress based on BIM model and AI video described in the present invention, the construction of a multi-scale spatial feature network and a dynamic skeleton modeling engine includes multi-source processing of a three-dimensional environment map, a structure-recognizable area map, and an obstacle probability distribution map, fusing the three-dimensional environment map, the structure-recognizable area map, and the obstacle probability distribution map after multi-source fusion processing, and fusing the displacement change data on the time series to construct a dynamic graph structure, and identify target components and structural unit nodes.

[0021] Identifying target components and structural unit nodes involves adopting a feature aggregation mechanism based on K-hop neighborhood relationships, taking each candidate node as the center, propagating the surrounding topological and spatial features through multi-order graph convolution operations, outputting the embedding vector of each node, and classifying the nodes into beam, plate, and column structural unit types by setting a classification discriminant head.

[0022] A skeleton connection diagram of building edge lines and key nodes is established using a structural line model based on skeleton fitting. The sparse node areas are corrected using an adaptive curvature enhancement strategy. The continuity and integrity of the overall structure recognition are enhanced through a graph regularization algorithm, and a structural map is output.

[0023] Establishing a skeleton connection diagram of building edge lines and key nodes includes selecting key points with high boundary response intensity and significant curvature changes in the structural feature map as skeleton fitting nodes, generating a preliminary skeleton diagram through an initial connection method based on the spatial minimum spanning tree, and combining the directional gradient and normal vector consistency between nodes to construct an optimized skeleton connection relationship.

[0024] As a preferred solution of the method for analyzing engineering progress based on BIM model and AI video, the defect anomaly labeling includes extracting the visual feature distribution and thermal anomaly layer in the potential abnormal area according to the optimized skeleton connection relationship, performing matching calculations against known defect templates in the anomaly knowledge base, and obtaining the defect anomaly labeling results.

[0025] Known defect templates include crack propagation patterns, equipment shedding patterns, and thermal concentration anomalies.

[0026] Matching includes using an abnormal matching algorithm based on the joint judgment of cosine distance and spatial clustering to realize matching calculation.

[0027] As a preferred solution of the method for analyzing engineering progress based on BIM model and AI video, the method includes triggering an engineering response mechanism based on a risk event scoring function, constructing a risk event vector sequence according to the defect anomaly labeling results, taking continuous abnormal data windows during the inspection process as input, and using the LSTM-Attention two-layer network structure to predict the risk evolution trend in the future time period, and performing a risk quantitative assessment of the current status to prepare an actual engineering progress analysis report.

[0028] As a preferred solution of the engineering progress method based on BIM model and AI video analysis described in the present invention, the remote dynamic adjustment of the construction path includes recalculating the optimal construction path based on the current building structure map, defect location coordinates, wind field, and occlusion.

[0029] Recalculating the optimal construction path includes adopting an improved multi-objective A algorithm based on heuristic guidance.

[0030] Another object of the present invention is to provide an engineering progress system based on BIM models and AI video analysis, which can achieve high-precision recognition and semantic graph generation of construction structures by constructing a multi-scale graph neural network and a dynamic skeleton modeling engine, solving the problems of current BIM+AI combined progress management technology that relies on manual progress updates and lacks component-level recognition and risk response capabilities.

[0031] As a preferred solution of the engineering progress system based on BIM model and AI video analysis described in the present invention, it includes: a data acquisition module, a defect marking module, and an engineering adjustment module.

[0032] The data acquisition module is used to collect regional image data and perform semantic segmentation and environment modeling.

[0033] The defect annotation module is used to construct a multi-scale spatial feature network and a dynamic skeleton modeling engine, input the three-dimensional environment map, the structure identifiable area map and the obstacle probability distribution map into the multi-scale spatial feature network and the dynamic skeleton modeling engine, and output the execution structure recognition and defect anomaly annotation.

[0034] The engineering adjustment module is used to trigger the engineering response mechanism based on the risk event scoring function and remotely and dynamically adjust the construction path.

[0035] A computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a method for analyzing engineering progress based on a BIM model and AI video.

[0036] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a method for analyzing engineering progress based on a BIM model and AI video.

[0037] Beneficial effects of the present invention: The method for analyzing construction progress based on a BIM model and AI video provides high-precision structural recognition of construction site images by combining semantic segmentation with three-dimensional environmental modeling, effectively resolving the issues of unclear structural recognition and fuzzy component boundaries in traditional video surveillance. By constructing a multi-scale spatial feature network and a dynamic skeleton modeling engine, a structural map is formed and component-level defect anomaly areas are accurately labeled, enabling automatic identification of construction structures and precise defect location. This overcomes the difficulty of automatically identifying construction status in existing BIM model progress management. The LSTM-Attention network structure is utilized to predict and quantitatively assess risk events, automatically triggering the project response mechanism and achieving an intelligent closed loop from structural identification to risk decision-making, resolving the issue of delayed discovery of progress anomalies. The introduction of a multi-objective improved algorithm combined with environmental dynamic factors to calculate the optimal construction path enhances the flexibility and real-time performance of construction management path adjustments, avoiding path redundancy and safety risks caused by manual adjustments. The present invention achieves improved results in terms of structural identification accuracy, construction progress response efficiency, and intelligent project site management capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 This is an overall flow chart of the engineering progress analysis method based on the BIM model and AI video provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0040] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0041] Example 1, with reference to Figure 1 , which is an embodiment of the present invention, provides a method for analyzing project progress based on a BIM model and AI video, including:

[0042] S1: Collect regional image data and perform semantic segmentation and environment modeling.

[0043] Using infrared thermal imaging equipment, visible light high-definition cameras, and lidar sensing components installed at the construction site, the target building structure is scanned in an area-covering manner within a set time interval, and the multimodal images and point cloud data collected in the scanning area are uniformly time-sequentially numbered and spatially aligned.

[0044] A preferred solution for unified temporal numbering and spatial registration operations is:

[0045] ,

[0046] in, 、 、 Indicates the coordinate value of the scanning point at the construction site in three-dimensional space, The horizontal coordinate of the pixel representing the construction feature point, Indicates the vertical coordinate of the pixel point of the construction feature point, Indicates that the construction equipment camera is The focal length of the direction, Indicates that the construction equipment camera is The focal length of the direction, 、 Indicates the coordinate position of the main point of the construction equipment camera on the image, Indicates the depth data corresponding to the construction feature point.

[0047] The spatial registration operation includes using a sparse feature matching algorithm based on laser SLAM to perform coarse alignment to maintain structural consistency between image frames and perform fine-grained registration based on the point cloud.

[0048] An optimal solution for fine-grained registration based on point clouds is:

[0049] ,

[0050] in, represents the mean square error of the registration between the construction point cloud and the BIM component anchor point, Represents the number of feature points involved in registration in the construction point cloud, Indicates the first Scan point coordinates, Indicates the target point cloud The corresponding BIM component node anchor coordinates, represents the construction site registration rotation matrix, represents the construction site registration translation vector, represents the Euclidean norm.

[0051] Each frame of data is bound to the building digital map model through timestamp and positioning information to form a perception frame cluster.

[0052] A deep convolutional neural network model is used to extract multi-channel features from the collected image data and construct a semantic segmentation map.

[0053] A preferred solution for constructing a semantic segmentation map is:

[0054] ,

[0055] in, represents the normalized pixel grayscale value of the construction site image, Represents the grayscale value of the pixel in the original construction image, Indicates the minimum grayscale value in the construction image, Indicates the maximum grayscale value in the construction image.

[0056] Structures, personnel, equipment, and abnormal areas are marked as semantic category layers, and misclassification correction is performed using a strategy based on spatial consistency and grayscale voting.

[0057] Based on the segmentation map, the environment is modeled and the structural boundary information, occlusion area location and construction path are generated into a three-dimensional environment map.

[0058] Generating a three-dimensional environment map includes: constructing the three-dimensional environment map by using an edge surface fitting method based on dense point cloud interpolation.

[0059] Voxel sparsification operation is introduced to compress the data volume, and the output structure can identify the area map and obstacle probability distribution map.

[0060] A preferred solution for outputting a 3D environment map, a structure-recognizable area map, and an obstacle probability distribution map is:

[0061] ,

[0062] in, Indicates the structural strength of the construction area, Indicates that the construction image is The gradient value of the direction, Indicates that the image is The gradient value of the direction.

[0063] The obtained structural strength of the construction area is used as input for the three-dimensional modeling of the construction environment and the identification of obstructions.

[0064] S2: Build a multi-scale spatial feature network and dynamic skeleton modeling engine.

[0065] A skeleton connection diagram of building edge lines and key nodes is established using a structural line model based on skeleton fitting. The sparse node areas are corrected using an adaptive curvature enhancement strategy. The continuity and integrity of the overall structure recognition are enhanced through a graph regularization algorithm, and a structural map is output.

[0066] An optimal solution for establishing a skeleton connection diagram of building edge lines and key nodes is:

[0067] ,

[0068] in, Represents the angle between the connection vector and the local normal vector between construction nodes, Indicates construction node and The connection vector of Indicates the normal vector of the construction node corresponding to the direction of the connecting line, represents the dot product operation, represents the Euclidean norm.

[0069] Establishing a skeleton connection diagram of building edge lines and key nodes includes selecting key points with high boundary response intensity and significant curvature changes in the structural feature map as skeleton fitting nodes, generating a preliminary skeleton diagram through an initial connection method based on the spatial minimum spanning tree, and combining the directional gradient and normal vector consistency between nodes to construct an optimized skeleton connection relationship.

[0070] S3: Input the 3D environment map, structure identifiable area map and obstacle probability distribution map into the multi-scale spatial feature network and dynamic skeleton modeling engine, and output the execution structure recognition and defect anomaly labeling.

[0071] Multi-source processing is performed on the three-dimensional environment map, the structure identifiable area map and the obstacle probability distribution map. The three-dimensional environment map, the structure identifiable area map and the obstacle probability distribution map after multi-source fusion processing are integrated with the displacement change data on the time series to construct a dynamic graph structure and identify the target components and structural unit nodes.

[0072] Identifying target components and structural unit nodes involves adopting a feature aggregation mechanism based on K-hop neighborhood relationships, taking each candidate node as the center, propagating the surrounding topological and spatial features through multi-order graph convolution operations, outputting the embedding vector of each node, and classifying the nodes into beam, plate, and column structural unit types by setting a classification discriminant head.

[0073] An optimal solution for the discriminant head to classify nodes into beam, plate, and column structural element types is:

[0074] ,

[0075] in, Indicates in After the layer The embedding vector of the construction node, Indicates the The learnable weight matrix of the layer construction graph, represents the activation function, Indicates the The construction node feature aggregation function of the layer, Indicates construction node The adjacent node of The embedding representation of the layer, Indicates construction node The set of neighbors.

[0076] The structural characteristics of each construction node are propagated through the characteristics of adjacent nodes, enhancing the discrimination ability of local components.

[0077] According to the optimized skeleton connection relationship, the visual feature distribution and thermal anomaly layer in the potential abnormal area are extracted, and matching calculations are performed against the known defect templates in the anomaly knowledge base to obtain the defect anomaly labeling results.

[0078] An optimal solution for defect anomaly marking is:

[0079] ,

[0080] in, represents two feature vectors for comparison, express The cosine similarity between .

[0081] Known defect templates include crack propagation patterns, equipment shedding patterns, and thermal concentration anomalies.

[0082] Matching includes using an abnormal matching algorithm based on the joint judgment of cosine distance and spatial clustering to realize matching calculation.

[0083] S3: Trigger the engineering response mechanism based on the risk event scoring function and remotely and dynamically adjust the construction path.

[0084] Based on the defect anomaly labeling results, a risk event vector sequence is constructed. Continuous abnormal data windows during the inspection process are used as input. The LSTM-Attention two-layer network structure is used to predict the risk evolution trend in the future time period, and a risk quantitative assessment of the current status is performed to produce an analysis report on the actual project progress.

[0085] A preferred solution for risk quantitative assessment is:

[0086] ,

[0087] ,

[0088] ,

[0089] ,

[0090] in, Indicates the The risk scoring results of the construction nodes are: represents the nonlinear feature adjustment factor, represents the Gaussian penalty factor, represents the status difference factor of the construction node, Indicates the The weighted coefficient of the construction index, Indicates the The construction node is at The value under the construction index, Indicates the The global mean of the construction index, Indicates the The standard deviation of the construction index, represents a small constant that avoids division by zero, Indicates the Penalty coefficient of construction index, Indicates the Ideal reference values ​​for construction indicators of this type, Indicates the The construction node is at The value under the construction index, Indicates the total number of nodes.

[0091] Recalculate the optimal construction path based on the current building structure map, defect location coordinates, wind field, and obstruction.

[0092] Recalculating the optimal construction path includes adopting a multi-objective improved algorithm based on heuristic guidance.

[0093] An optimal solution for the multi-objective improved algorithm based on heuristic guidance is:

[0094] ,

[0095] in, Indicates construction node The total cost, Indicates the starting point to the construction node The actual cost, Indicates that from the construction node Heuristic estimate of the cost to the endpoint.

[0096] Example 2 is an embodiment of the present invention, which provides an engineering progress system based on BIM model and AI video analysis, including a data acquisition module, a defect marking module, and an engineering adjustment module.

[0097] Among them: the data acquisition module is used to collect regional image data and perform semantic segmentation and environment modeling.

[0098] It should also be noted that the data acquisition module, as the front-end perception unit of the system, collects multimodal raw data from the construction site, and through spatial registration and time series numbering, constructs multi-source synchronous data frames, image semantic segmentation, generates multi-category semantic layers, reconstructs and fits point clouds, outputs three-dimensional environment maps, performs voxel sparsification processing, and generates structure-recognizable area maps and obstacle probability distribution maps.

[0099] The processed structural diagram, obstacle diagram, and semantic layer are used as input data for the defect annotation module, providing the multi-source foundation required for building graph neural networks and skeleton structure modeling.

[0100] Among them: the defect annotation module is used to build a multi-scale spatial feature network and dynamic skeleton modeling engine, input the three-dimensional environment map, structure identifiable area map and obstacle probability distribution map into the multi-scale spatial feature network and dynamic skeleton modeling engine, and output execution structure recognition and defect anomaly annotation.

[0101] It should also be noted that the defect annotation module undertakes the core recognition and analysis functions of the system, outputs the structural unit node map under the multi-scale spatial feature network, the structural contour map based on skeleton connection, the defect anomaly annotation results generated by the fusion and matching of the thermal map layer and image features, and the embedding vector, status indicator and node type label of each node component.

[0102] The results are used as input for risk event modeling in the engineering adjustment module, which makes intelligent adjustments to construction paths and work plans by analyzing abnormal distribution and risk trends.

[0103] Among them: the engineering adjustment module is used to trigger the engineering response mechanism based on the risk event scoring function and remotely and dynamically adjust the construction path.

[0104] It should also be noted that after completing the risk scoring and path recalculation, the engineering adjustment module will feed back the adjusted optimal construction path, risk level spatial distribution map, node priority processing order, and designated area re-sampling request to the data acquisition module, thereby realizing dynamic optimization control of the acquisition range and frequency, ensuring that the subsequently collected data can perform higher-frequency and more precise scanning and sampling of risk areas, forming a closed loop of data-analysis-feedback.

Claims

1. A method for analyzing engineering progress based on BIM model and AI video, characterized by: include: Collect regional image data and perform semantic segmentation and environment modeling; Build a multi-scale spatial feature network and dynamic skeleton modeling engine; The 3D environment map, the structure identifiable area map and the obstacle probability distribution map are input into the multi-scale spatial feature network and the dynamic skeleton modeling engine, and the output performs structure recognition and defect anomaly annotation; Trigger the engineering response mechanism based on the risk event scoring function and remotely and dynamically adjust the construction path; The three-dimensional environment map includes, generating a three-dimensional environment map by modeling the environment and including the structure boundary information, the location of the occluded area and the construction path; The structure-recognizable region map and the obstacle probability distribution map include inputting the collected region image data into the constructed deep convolutional neural network model to obtain the structure-recognizable region map and the obstacle probability distribution map; Building a multi-scale spatial feature network and dynamic skeleton modeling engine includes identifying target components and structural unit nodes and outputting a structural map through a multi-scale residual graph neural network and a skeleton fitting engine; The acquisition area image data includes: Using infrared thermal imaging equipment, visible light high-definition cameras, and lidar sensing components installed at the construction site, the target building structure is scanned in an area-wide manner within a set time interval. The multimodal images and point cloud data collected within the scanned area are uniformly time-sequentially numbered and spatially registered. The spatial registration operation includes using a sparse feature matching algorithm based on laser SLAM to perform coarse alignment to maintain structural consistency between image frames, and performing fine-grained registration based on the point cloud. Each frame of data is bound to the building digital map model through timestamps and positioning information to form a perception frame cluster. The execution of semantic segmentation and environment modeling includes: A deep convolutional neural network model is used to extract multi-channel features from the collected image data, construct a semantic segmentation map, and label structures, personnel, equipment, and abnormal areas as semantic category layers. A strategy based on spatial consistency and grayscale voting is used to correct misclassifications. Based on the segmentation map, the environment is modeled to generate a three-dimensional environment map with the structure boundary information, the location of the occluded area and the construction path; Generating a three-dimensional environment map includes: constructing the three-dimensional environment map using an edge surface fitting method based on dense point cloud interpolation; The voxel sparsification operation is introduced to compress the data volume, and the output structure can identify the area map and the obstacle probability distribution map; The construction of the multi-scale spatial feature network and dynamic skeleton modeling engine includes: Perform multi-source processing on the 3D environment map, the structural identifiable area map, and the obstacle probability distribution map. The 3D environment map, the structural identifiable area map, and the obstacle probability distribution map after multi-source fusion processing are integrated with the displacement change data in the time series to construct a dynamic graph structure and identify the target components and structural unit nodes. Identifying target components and structural unit nodes involves adopting a feature aggregation mechanism based on K-hop neighborhood relationships, taking each candidate node as the center, propagating the surrounding topological and spatial features through multi-order graph convolution operations, outputting the embedding vector of each node, and classifying the nodes into beam, plate, and column structural unit types by setting a classification discriminant head; Using a structural line model based on skeleton fitting, a skeleton connection diagram of building edge lines and key nodes is established. An adaptive curvature enhancement strategy is used to correct sparse node areas. A graph regularization algorithm is used to enhance the continuity and integrity of overall structural recognition and output a structural map. Establishing a skeleton connection diagram of building edge lines and key nodes involves selecting key points with high boundary response intensity and significant curvature changes in the structural feature map as skeleton fitting nodes, generating a preliminary skeleton diagram through an initial connection method based on a spatial minimum spanning tree, and constructing an optimized skeleton connection relationship by combining the directional gradient and normal vector consistency between nodes; The defect anomaly marking includes: Based on the optimized skeleton connection relationship, the visual feature distribution and thermal anomaly layer in the potential abnormal area are extracted, and matching calculations are performed against the known defect templates in the anomaly knowledge base to obtain the defect anomaly labeling results; Known defect templates include crack propagation patterns, equipment shedding patterns, and thermal concentration anomalies; Matching includes,using an abnormal matching algorithm based on the joint judgment of cosine distance and spatial clustering to achieve matching calculation; The engineering response mechanism triggered based on the risk event scoring function includes: Based on the defect anomaly labeling results, a risk event vector sequence is constructed. Continuous abnormal data windows during the inspection process are used as input. The LSTM-Attention two-layer network structure is used to predict the risk evolution trend in the future time period, and a risk quantitative assessment of the current status is performed to produce an analysis report on the actual project progress.

2. The method for analyzing project progress based on a BIM model and AI video according to claim 1, characterized in that: The remote dynamic adjustment of the construction path includes: Recalculate the optimal construction path based on the current building structure map, defect location coordinates, wind field, and obstruction; Recalculating the optimal construction path includes adopting a multi-objective improved algorithm based on heuristic guidance.

3. A system for analyzing project progress based on a BIM model and AI video, using a method for analyzing project progress based on a BIM model and AI video as described in any one of claims 1 to 2, characterized in that: Including data acquisition module, defect marking module, and engineering adjustment module; The data acquisition module is used to collect regional image data and perform semantic segmentation and environment modeling; The defect annotation module is used to construct a multi-scale spatial feature network and a dynamic skeleton modeling engine, input the three-dimensional environment map, the structure identifiable area map and the obstacle probability distribution map into the multi-scale spatial feature network and the dynamic skeleton modeling engine, and output the execution structure recognition and defect anomaly annotation; The engineering adjustment module is used to trigger the engineering response mechanism based on the risk event scoring function and remotely and dynamically adjust the construction path.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for analyzing engineering progress based on BIM model and AI video are implemented as described in any one of claims 1 to 2.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for analyzing engineering progress based on a BIM model and AI video are implemented as described in any one of claims 1 to 2.

Images