A BIM-based 360° full-scene construction progress management method and system

By adopting a BIM-based 360° full-scene construction progress management method, the problem of heterogeneous progress observation inside and outside the building was solved, and the unified evaluation and adaptive judgment of progress data were realized, reducing false alarms and omissions, thus forming a closed loop of progress management.

CN123118010APending Publication Date: 2026-11-17CHINA CONSTR THIRD ENG BUREAU GRP CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611197623.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-07
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, the observation of the building's internal and external structures and the indoor construction progress are heterogeneous, and the reliability varies with the data collection conditions and time. Simple averaging and merging can easily lead to physical inconsistencies in the merged progress. Relying on deviation judgment with fixed thresholds can easily cause false or missed progress reports, and the correction process has not formed a closed management loop.

Method used

A BIM-based 360° full-scene construction progress management method is adopted. By acquiring external and internal progress observations of the building, calculating the reliability of each observation and applying weights that decay with time intervals, weighted fusion and bottom-up convolution operations are performed. Combined with time monotonicity and process sequence constraints, deviation judgment and alarm thresholds are dynamically adjusted to form an adaptive progress management closed loop.

Benefits of technology

It enables the merging, evaluation, and adaptive quantification of multi-source heterogeneous internal and external field observation data on a unified BIM benchmark, reducing false and missed progress reports, ensuring the physical consistency and reliability of progress assessment, and forming a closed-loop management system from deviation identification to resource allocation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118010A_ABST
    Figure CN123118010A_ABST
Patent Text Reader

Abstract

This invention relates to the field of construction progress management technology, and discloses a 360° full-scene construction progress management method and system based on BIM, used for automatic progress perception and collaborative decision-making in complex projects. Addressing the problem of physical inconsistency in fusion evaluation and open-loop correction caused by heterogeneous internal and external progress observations, this invention labels heterogeneous internal and external observations to unified BIM nodes, and performs weighted fusion based on data quality and time decay calculation reliability; through bottom-up convolutional harmonics and dual temporal constraints, a physically consistent progress benchmark is generated; based on aggregate confidence, the alarm threshold judgment deviation is adaptively adjusted, and a quantitative resource allocation plan is output. This invention, through confidence weighting, effectively suppresses false alarms and missed alarms in progress management, achieving closed-loop correction under a single benchmark.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of construction progress management technology, and in particular to a 360° full-scene construction progress management method and system based on BIM. Background Technology

[0002] Construction schedule management is crucial throughout all stages of building construction and serves as a key basis for on-site resource procurement and scheduling. Whether the project can be completed on time directly impacts cost control and contract fulfillment. For a long time, construction schedule management has relied heavily on Gantt charts, network planning, and BIM-based schedule planning. While these methods can streamline process logic and clarify milestone arrangements during the planning phase, obtaining schedule information during the dynamic implementation phase still primarily depends on manual on-site recording and hierarchical reporting, resulting in limitations such as high latency and limited coverage.

[0003] To improve the automation of progress data acquisition, the industry has introduced on-site image acquisition and 3D perception methods. Externally, high-definition cameras mounted on tower cranes and perimeter cameras are used to collect structural construction images. Point cloud reconstruction and comparison with the BIM model are then used to determine structural progress. For indoor spaces that are difficult for external cameras to cover, manual photography or lightweight 360° panoramic scanning combined with image recognition is used to assess the progress of interior masonry, electromechanical installation, and decoration processes. BIM serves as a digital twin platform to unify the organization of models, plans, and images.

[0004] However, structural progress observations from outside the building and indoor progress observations from inside the building are heterogeneous in terms of acquisition methods, scale, and content. Their reliability varies not only depending on the data source but also over time: external observations are affected by observation distance, motion blur, and component occlusion, while internal observations are affected by camera pose calibration residuals and feature recognition confidence. Furthermore, both observations become outdated as construction progresses. When such observations are managed separately, the overall building progress is difficult to assess on a unified benchmark. When they are combined using coarse methods such as single-source readings or simple averaging without considering reliability, the resulting progress is prone to physical inconsistencies, such as a higher-level node showing completion while its subordinate components are not yet completed, or a subsequent process showing completion while its preceding process has not yet been reached. The fusion result is unreliable. Using a fixed threshold for deviation judgment on such progress will frequently result in false alarms for unreliable readings and omissions for critical tasks. Moreover, subsequent corrections remain in an open-loop state, merely issuing alarms without quantitatively correlating the deviations with their resource causes and feasible resource adjustments. Therefore, how to obtain a single, physically consistent progress that quantifies the reliability from heterogeneous internal and external progress observations where the reliability varies with the source and time, and how to make deviation judgment and resource correction adaptive with the reliability, thereby suppressing false alarms and false negatives and forming a closed loop, has become a core technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to overcome the technical defects in the existing technology, such as the heterogeneity of the observation of the internal and external structure of the building and the indoor construction progress, the fact that the reliability varies with the data collection conditions and time, the tendency of simple averaging and merging to lead to physical inconsistencies in the merged progress, and the tendency of relying on fixed threshold deviation judgment to easily cause false alarms and omissions in the progress, and the failure to form a management closed loop for correction. The invention provides a 360° full-scene construction progress management method and system based on BIM.

[0006] This invention provides a BIM-based 360° full-scene construction progress management method, comprising the following steps:

[0007] Obtain structural progress observations from the exterior of the building and interior progress observations from the interior of the building, and mark each progress observation on the leaf node of the corresponding component and work package binary in the BIM work breakdown structure;

[0008] The reliability of each progress observation is calculated based on the data source and observation quality, and a weight that decays with time interval is applied to obtain the comprehensive weight of each progress observation.

[0009] Based on the comprehensive weight, all progress observations under the same leaf node are weighted and fused to obtain the fusion completion ratio of the leaf node, and the aggregation confidence of the leaf node is calculated based on the comprehensive weight of each progress observation.

[0010] The convolution operation is performed from bottom to top using the fusion completion ratio of child nodes and the corresponding workload weights. The resulting convolution value is then weighted and harmonicized with the direct observation of the parent node itself based on confidence, to obtain the initial completion ratio of each node after hierarchical consistency projection.

[0011] By simultaneously applying time monotonicity constraints and process sequence constraints to the initial completion ratio, the final completion ratio of each node is obtained.

[0012] The progress deviation is calculated based on the final completion ratio and the planned completion ratio of each node, and the deviation alarm threshold is dynamically adjusted according to the aggregate confidence level to perform adaptive deviation judgment and progress warning.

[0013] Through the above steps, the combined evaluation and adaptive quantitative determination of multi-source heterogeneous internal and external field observation data are realized on a unified building information model benchmark.

[0014] As a preferred embodiment of the present invention, the method for calculating the comprehensive weight and the method for calculating the fusion completion ratio of the leaf nodes specifically includes:

[0015] External structural progress monitoring Its credibility calculation model is as follows:

[0016]

[0017] in, The distance from the component to the camera. The occlusion ratio of the component in the model. For residual ambiguity, For distance normalization scale, The fuzzy sensitivity coefficient;

[0018] For internal indoor progress monitoring Its credibility calculation model is as follows:

[0019]

[0020] in, To calibrate the residuals of the acquired pose, To segment the mask confidence level, To determine the dispersion of the local cluster, and For the corresponding precision scale;

[0021] The combined weight of the decay of each progress observation over time is: ,in , For the assessment time, For the time of data collection, The decay time constant;

[0022] The leaf node Integration completion rate and aggregate confidence The calculation model is as follows:

[0023]

[0024] in, For the evaluation time Previous leaf nodes The accumulated set of observations The original completion percentages are as follows.

[0025] This quantification model ensures that data with high confidence and timeliness dominates the progress fusion process, avoiding conflicting readings from affecting overall accuracy.

[0026] As a preferred embodiment of the present invention, the method for calculating the bottom-up hierarchical consistency projection specifically includes:

[0027] Through child nodes completion rate and its workload weight Calculate the parent node The actual completion ratio of the convolution value :

[0028]

[0029] If the parent node has a directly observed fusion completion rate and corresponding aggregate confidence level Then, the harmonic completion ratio is obtained by harmonic adjustment of the convolution value with the corresponding confidence level. :

[0030]

[0031] in, The convolution confidence is obtained by weighting the confidence of child nodes by workload.

[0032] This eliminates potential logical inconsistencies in project quantities when merging progress data at the individual unit, floor, and sub-component levels, ensuring the physical consistency of the overall progress assessment.

[0033] As a preferred embodiment of the present invention, the specific constraint rules for the time monotonicity constraint and the process sequence constraint include:

[0034] Use historical fusion values ​​to assess the completion rate of the current node. Apply time monotonicity lower bound clamp:

[0035]

[0036] in, This represents the fusion completion rate of this node in the previous cycle;

[0037] Based on the set of preceding work packages Apply an upper limit constraint, controlled by parallel preceding processes, to the completion rate of the current node:

[0038]

[0039] in, Pre-work package The completion rate, This is to account for the tolerance of misjudgment during the process.

[0040] This dual constraint effectively eliminates abnormal rollbacks in progress data caused by local data obstruction, as well as misjudgments that violate the objective laws governing the sequence of construction.

[0041] As a preferred embodiment of the present invention, the alarm triggering condition for the adaptive deviation determination is as follows: ,in To establish confidence in the multiplier, To incorporate the schedule deviation between the completed and planned completion dates, the effective uncertainty is... With key tightening factors The calculation models are as follows:

[0042]

[0043] in, The calculated value is the aggregate confidence level after taking the effective lower bound. Basic observation standard deviation The standard deviation of the historical deviation of the component's stage. For critical path identification, This is the tightening factor;

[0044] When a progress warning is triggered, the cause of the deviation is attributed, and if the cause is insufficient labor, a quantitative personnel allocation plan is output with the goal of minimizing the weighted deficit.

[0045]

[0046] in, Lagging region The required working hours For supply area Surplus working hours For the region Transferred to the region The number of people to be deployed, The remaining working days. For single-person daily working hours, Priority is given to areas with lagging development.

[0047] This significantly reduces the false alarm rate for unreliable observation data and the underreporting rate for construction delays on the critical path, and extends schedule correction to quantified resource allocation instructions, forming a closed loop for schedule management.

[0048] As a preferred embodiment of the present invention, the specific steps for obtaining structural progress observations of the building exterior include:

[0049] Collect high-altitude video from tower cranes and ground monitoring video, minimizing adjacent frames. and Between residuals Solve for the motion vectors and perform deconvolution deblurring compensation on the high-altitude frames of the tower crane high-altitude video;

[0050] The ground surveillance video is reconstructed using super-resolution and the perspective deviation is corrected using a radial distortion model.

[0051] Using building corner points and site benchmark control points as shared feature points, the constraints are minimized. The similarity transformation parameters between high-altitude video and ground monitoring video are solved to achieve alignment in the same world coordinate system. Semantic segmentation is used to embed the components identified by ground monitoring into the occluded areas of the tower crane high-altitude model, generating a seamless real-scene model as the data basis for external structure progress observation.

[0052] This allows the details of ground-based images to compensate for the blind spots caused by the tower crane's high altitude, achieving semantic connectivity and filling the visual blind spots in traditional external reality perception.

[0053] As a preferred embodiment of the present invention, the specific steps for obtaining indoor progress observations inside the building include:

[0054] A panoramic scan is performed along the indoor inspection path to generate a stitched panoramic space, and the target component elements are identified based on a semantic segmentation model.

[0055] The same component pixels identified in panoramic images from different acquisition points are back-projected into spatial rays originating from the camera's optical center. The midpoint of the common perpendicular segment of any two rays is then used as a candidate intersection point.

[0056]

[0057] in, and The coordinates of the optical center of the acquisition point. and This is the unit direction vector corresponding to the pixel. and The solution is obtained from a system of linear equations formed by the ray direction and the coordinates of the optical center;

[0058] Clustering of all candidate intersection points is performed using a nearest neighbor density-based clustering algorithm. The centroid of the cluster with the most points is extracted as the true coordinates of the component in three-dimensional space, and the dispersion of points within the cluster is characterized as the localization cluster dispersion of this internal observation.

[0059] This solution effectively overcomes the difficulty of performing spatial distance measurement directly on complex construction sites using monocular panoramic vision, greatly improving the accuracy and reliability of three-dimensional positioning of components.

[0060] This invention also provides a BIM-based 360° full-scene construction progress management system, which uses the above-mentioned method and includes:

[0061] The perception access module is used to acquire structural progress observations outside the building and indoor progress observations inside the building, and to map each progress observation to the leaf nodes of the corresponding component and work package tuple in the BIM work breakdown structure.

[0062] The confidence quantification module is used to calculate the confidence and comprehensive weight of each progress observation based on data source characteristics, observation quality indicators and time decay constant.

[0063] The weighted fusion module is used to perform weighted fusion of multi-source heterogeneous progress observations at the same leaf node level, using the comprehensive weight as the weight parameter, and outputs the fusion completion ratio and aggregation confidence of the leaf node.

[0064] The hierarchical harmonization module is used to perform convolution projection on the fusion completion ratio of child nodes according to the workload weight from bottom to top according to the BIM tree structure, and harmonize it with the completion ratio directly observed by the parent node according to their respective confidence levels.

[0065] The dual-clamp module is used to synchronously apply a time monotonicity lower limit clamp based on historical data to the reconciled schedule status, and a process sequence upper limit clamp controlled by the preceding parallel processes, to establish a physically self-consistent schedule benchmark.

[0066] The adaptive control module is used to dynamically set the deviation alarm threshold based on the aggregate confidence level, historical fluctuations of the stage, and critical path identification to perform deviation judgment, and generate a closed-loop deviation correction and allocation scheme attributable to the corresponding resource shortage after the deviation alarm is triggered.

[0067] The system effectively integrates on-site sensing hardware and software facilities, enabling fully automated workflow of progress tracking and intelligent decision-making within the digital twin platform.

[0068] As a preferred embodiment of the present invention, the sensing access module includes an internal panoramic inspection subsystem, which is equipped with an indoor construction environment panoramic scanning device for roving scanning. This device specifically includes:

[0069] The main support module includes an adjustable boom with an adjustable acquisition angle, a mounting column, and a high-definition PTZ camera;

[0070] The transmission protection module includes a micro motor, a light-shielding plate connecting ring connected to the micro motor, and a light-shielding plate installed on the light-shielding plate connecting ring; wherein, when the system is exposed to strong light and triggers overexposure of the image, the micro motor is controlled to start and drive the light-shielding plate connecting ring and the light-shielding plate to rotate to the optimal blocking angle to eliminate glare.

[0071] The data processing module, including an image sensor and a low-latency Bluetooth module, is used to transmit digitized acquired images to the nearest edge gateway at the construction site and offload and forward them to the cloud.

[0072] This ensures the clarity of image acquisition within the construction work area under extreme strong light, and improves the efficiency of unloading and processing large-capacity multi-source image data through a hierarchical transmission mechanism.

[0073] As a preferred embodiment of the present invention, the system is further configured with a lightweight modular module that interacts with the BIM digital twin base, specifically used for:

[0074] Based on the semantic segmentation network, the global scene point cloud is split into two levels of hierarchical fine segmentation according to the main building blocks and component units, and semantic clusters with independent construction semantic attributes are extracted and generated.

[0075] Density-priority thinning is performed using voxels as units, and the results are evaluated using calculation formulas. Calculate any point cloud points within each semantic block. The local density is calculated, and only the single representative point with the highest density estimate is retained within each individual voxel, where... To select the nearest neighbor number, For the first The distance between the nearest neighbors is limited by a very small positive lower bound to avoid division by zero singularity;

[0076] Spatial indexes are established for 3D model units that have undergone precision-controlled and lightweight processing. In the front-end interaction, corresponding block models are issued and rendered online as needed based on view coordinates or BIM component click commands.

[0077] This solution specifically reduces the massive size of the entire site cloud and shortens the terminal model loading time to the second level, providing a smooth operational foundation for on-site mobile online management and control.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] This invention unifies structural progress observations from outside the building and indoor progress observations from inside the building into the same BIM work breakdown structure. It then weights and fuses these observations with a confidence level that reflects the quality of each observation and decays over time. This is followed by bottom-up convolution harmonics weighted by workload and constrained by time monotonicity and process sequence constraints. This allows internal and external progress to be evaluated on the same benchmark. Conflicting readings are selected based on reliability rather than data source, thus avoiding physical inconsistencies such as higher-level nodes exceeding the capacity of their subordinate components or subsequent processes exceeding their predecessors. The fused progress is more closely aligned with actual site conditions than using a single data source or a simple average without considering reliability. The error in determining the progress can be controlled within 3%, lower than the error level of manual statistics.

[0080] Based on this, by setting the deviation alarm threshold as an adaptive quantity that varies with the fusion confidence level, the volatility of the construction phase, and the criticality of the components, observations with lower reliability require larger deviations to trigger an alarm, while work on the critical path is marked with smaller deviations. This reduces false alarms for unreliable readings and missed alarms for critical work compared to fixed thresholds. Furthermore, since the triggered deviations are further attributed to quantified labor or material shortages and transformed into personnel allocation plans constrained by surplus resources, progress management forms a closed loop from deviation identification and cause attribution to feasible adjustments, rather than stopping at alarm prompts. Attached Figure Description

[0081] Figure 1 This is a schematic diagram of the system structure of a preferred embodiment of the present invention;

[0082] Figure 2 This is a schematic diagram of the process of multi-source video real-scene reverse modeling in a preferred embodiment of the present invention;

[0083] Figure 3 This is a schematic diagram of the process for fusing internal and external multi-source progress evidence in a preferred embodiment of the present invention;

[0084] Figure 4 This is a flowchart illustrating the adaptive deviation determination and resource correction process in a preferred embodiment of the present invention.

[0085] Figure 5 This is a schematic diagram of the structure of the panoramic scanning device for indoor construction environment acquisition in a preferred embodiment of the present invention.

[0086] Explanation of reference numerals in the attached diagram: 100, External Real-Scene Perception Subsystem; 110, Tower Crane PTZ Camera; 120, Perimeter Fixed Camera; 130, Mobile Inspection Camera; 140, Vehicle Entrance / Exit Bullet Camera; 150, Weighbridge Sensor; 200, Internal Panoramic Inspection Subsystem; 210, Panoramic Scanning Device; 220, Main Support Module; 221, Adjustable Boom; 222, Mounting Column; 223, High-Definition PTZ Camera; 230, Transmission Protection Module; 231, Sunshade Connecting Ring; 232, Sunshade; 240, Data Processing Module; 241, Image Sensor; 2 42. Low-latency Bluetooth module; 250. Edge gateway; 260. Camera mounting plate; 300. Cloud data processing and service layer; 310. Reverse modeling module; 320. Lightweight modular module; 330. Element recognition module; 340. Progress fusion module; 350. Deviation judgment and correction module; 360. Labor statistics module; 370. Digital archive module; 400. BIM digital twin base; 500. Control and display layer; 510. Progress monitoring module; 520. Image management module; 530. Resource management module; 540. Early warning and optimization module. Detailed Implementation

[0087] The following is a detailed description of the 360° full-scene construction progress management method and system based on BIM provided by this invention. This solution is applicable to super high-rise buildings, large complexes, stadiums and residential projects with overlapping internal and external operations and complex process logic, and is used to automatically perceive, uniformly evaluate and collaboratively decide on the construction progress of the building's external structure and internal space.

[0088] See Figure 1This invention comprises an external real-scene perception subsystem 100, an internal panoramic inspection subsystem 200, a cloud data processing and service layer 300, a unified BIM digital twin base 400, and a multi-terminal control and display layer 500. The external real-scene perception subsystem 100 and the internal panoramic inspection subsystem 200 are responsible for collecting on-site data from the exterior and interior of the building, respectively. The collected multi-source data is analyzed, modeled, identified, and fused by the cloud data processing and service layer 300. Using the BIM digital twin base 400 as a unified spatial coordinate and semantic reference, the control and display layer 500 ultimately provides managers with an interactive interface for progress viewing, deviation warnings, and resource allocation. The cloud-based data processing and service layer 300 includes a reverse modeling module 310, a lightweight modularization module 320, an element recognition module 330, a progress fusion module 340, a deviation judgment and correction module 350, a labor statistics module 360, and a digital archive module 370. The control and display layer 500 includes a progress monitoring module 510, an image management module 520, a resource management module 530, and an early warning and optimization module 540. The control and display layer 500 is implemented using a browser and server architecture. The server is developed in Java and deployed on a web server, requiring no additional installation programs on the terminal. It can be accessed on computers, tablets, and mobile phones to meet the needs of mobile office work on construction sites. BIM models, panoramic images, progress plans, and personnel, material, and machinery data are uniformly incorporated into the backend database management and kept synchronized in real time.

[0089] The external real-scene perception subsystem 100 relies on existing hardware resources to build a data acquisition network that emphasizes both full coverage and detailed supplementation. Tower crane PTZ cameras 110 and perimeter fixed cameras 120 simultaneously acquire images of the building structure and site facilities from high-altitude and full-area perspectives. Mobile inspection cameras 130 supplement detailed images for confined spaces such as manholes and equipment rooms. Vehicle entrance / exit cameras 140, in conjunction with weighbridge sensors 150, register and track vehicles entering and leaving the site using license plate numbers and colors, and also register the materials carried by transport vehicles. Managers can access corresponding real-time images by clicking on point icons on the floor plan in the control display layer 500, and can quickly locate key areas and blind spots such as deep foundation pits and material storage yards using the floor plan. AI video analysis provides second-level alarms for risks such as not wearing safety helmets, crossing boundaries, and open flames / smoke, and transmits the operating status of machinery such as tower cranes and construction elevators back to the central platform in real time.

[0090] In the real-world reconstruction phase of the external progress, the reverse modeling module 310 first performs cross-scale adaptive preprocessing on multi-source video streams from tower crane aerial and ground monitoring, and then unifies them into the same world coordinate system to complete the reverse modeling. The process is as follows: Figure 2As shown. To address the motion blur generated by the tower crane camera during rotation and amplitude changes, the module estimates the motion vectors between adjacent frames frame by frame and compensates accordingly (S101). The motion vectors are obtained by minimizing the inter-frame residuals.

[0091]

[0092] in , For two adjacent frames, The pixel coordinates are used; deconvolving this frame with the point spread function spanned by the cumulative displacement within a single frame exposure time can suppress motion blur. For low-resolution ground videos, deep learning-based super-resolution reconstruction is used to upscale the resolution to at least twice the original value, and a radial distortion model is used to correct perspective deviation. The correction relationship is as follows:

[0093]

[0094] In the formula These are distorted image points. For the corrected image points, , The radial distortion coefficient is used. After preprocessing, the building corner points and the reference control points set up on site are used as shared feature points for the aerial video and the ground video. The similarity transformation between the two is solved to achieve automatic alignment in a unified coordinate system (S102).

[0095]

[0096] in , These are the coordinates of the same feature point in the camera coordinate system and the world coordinate system. , , These are the scale factor, rotation matrix, and translation vector, respectively. Based on geometric consistency, motion reconstruction structures and multi-view stereo vision algorithms are used to align feature points, remove redundancy, and filter noise in multi-source point clouds, generating dense point clouds with centimeter-level accuracy. Then, using tower crane aerial panoramic images, a macroscopic geometric framework and real coordinates of the site are rapidly constructed. Ground monitoring images are treated as dynamic patches, and semantic segmentation is used to identify components such as walls, windows, beams, and columns, accurately embedding them into the occluded areas of the building's base and internal space within the tower crane model. This high-to-low compensation and semantic integration fill in the blind spots of traditional modeling, resulting in a complete and seamless computable real-world model. This step corresponds to... Figure 2 Dense reconstruction and occlusion compensation (S103) are performed. The resulting model is synchronized to the service layer in real time. Administrators can retrieve the 3D model for any time period through the timeline control, download it as needed for offline analysis or multi-party collaboration, and automatically identify building height and measure the distance between any two points in real time on the model.

[0097] For full-site real-world models with massive point clouds reaching billions of points, the lightweight modular module 320 employs hierarchical modularization based on construction semantics and lightweight processing with controllable precision (corresponding to...). Figure 2 The semantic lightweighting step S104 addresses the bottlenecks of slow loading and difficulty in online application of traditional terminals. The block segmentation abandons the octree voxel partitioning that ignores semantic boundaries, and instead proceeds with a two-level expansion: coarse segmentation of building blocks and fine segmentation of component units. Based on the site division and building height stratification, the entire site's point cloud is divided into independent building units and near-ground grid blocks. Then, the Superpoint Transformer deep learning algorithm is introduced to perform component-level semantic segmentation on each coarse block, identifying components such as beams, columns, slabs, and pipelines, and forming semantic clusters. This ensures that the point cloud within each cluster is highly consistent geometrically and semantically, preventing subsequent processing from damaging the integrity of the components. The lightweighting process follows the principle of density priority and controllable accuracy. Based on the positive correlation between point cloud density and reconstruction accuracy in image modeling, high-density point clouds from the optimal viewpoint are automatically retained within each semantic cluster, while low-precision noise caused by occlusion or long-distance shooting is removed. The local density according to its first Nearest neighbor distance estimated as

[0098]

[0099] in From this point to the... The distance between the nearest neighbors is used to calculate the density on the set of points that have undergone redundancy removal, and the distance between the nearest neighbors is used to calculate the density on the set of points that have undergone redundancy removal. Take a very small positive lower limit To avoid overlapping points caused by splicing, which would result in a zero denominator, the model is thinned using voxels as units. Each voxel retains only the representative point with the highest density, ensuring that the modeling error of key components does not exceed 2 mm. This reduces the model size by more than 70%, and the size of a single plot model can be reduced from approximately 5 gigabytes to less than 1.5 gigabytes. The lightweight block module 320 establishes a spatial index for the block-based model and links it with the BIM digital twin base 400 and progress data. The front end does not need to download the complete model. Managers can load the corresponding area's 3D data as needed by entering coordinates in the browser or clicking on the BIM component, reducing the loading time to less than 3 seconds. Clicking on any area of ​​the construction drawing will link to the corresponding 3D model and pop up a panel showing the component material, specifications, process requirements, and responsible person. This, combined with real-time distance measurement, allows for verification of reserved opening dimensions and component spacing.

[0100] The internal panoramic inspection subsystem 200 uses an indoor construction environment panoramic scanning device 210 as the data acquisition terminal, the structure of which is described in the reference document. Figure 5The device consists of three parts: a main support module 220, a transmission and protection module 230, and a data processing module 240. The main support module 220 includes an adjustable boom 221, a mounting column 222, a camera mounting plate 260, and a high-definition PTZ camera 223. The mounting column 222 is fixedly installed on the ground. The adjustable boom 221 is rotatably installed on the mounting column 222. The camera mounting plate 260 is fixedly installed at the end of the adjustable boom 221. The high-definition PTZ camera 223 is installed on the camera mounting plate 260 through a camera bracket. The adjustable boom 221 can flexibly adjust the acquisition angle according to the construction space to cover concealed areas such as manholes and ceilings.

[0101] The transmission protection module 230 includes a micro motor, a light-shielding plate connecting ring 231 connected to the micro motor, and a light-shielding plate 232 mounted on the light-shielding plate connecting ring 231. The micro motor is installed inside the camera housing. The light-shielding plate connecting ring 231 is rotatably located outside the camera housing via shafts at both ends. The shafts on the light-shielding plate connecting ring 231 and the output shaft of the micro motor are coaxially and fixedly connected. The light-shielding plate 232 is fixed at the center position on the light-shielding plate connecting ring 231 and is located directly above the camera. The micro motor drives the light-shielding plate connecting ring 231 and the light-shielding plate 232 to rotate, thereby controlling the angle of the light-shielding plate 232 (rotation direction as shown in the image). Figure 5 (As shown in the diagram). The data processing module 240 incorporates a high-sensitivity image sensor 241 and a low-latency Bluetooth module 242. Under normal lighting conditions, the high-definition PTZ camera 223 captures 4K resolution images to restore the texture of walls and pipelines. When the image is overexposed due to direct sunlight or strong light, the system triggers a micro motor to rotate the light shield 232 to the optimal shading angle to eliminate glare and ensure image clarity. The image sensor 241 decomposes the captured image into pixel-level digital signals. The low-latency Bluetooth module 242 transmits these digital signals via a short-range wireless link to the edge gateway 250 on the inspection side. The edge gateway 250, which has wireless LAN or cellular network capabilities, then forwards them to the cloud data processing and service layer 300. This allows for timely data unloading during the acquisition process, avoiding data stagnation on-site, and improving the efficiency of subsequent 3D modeling by approximately three times compared to manual photography. During the inspection, the device 210 automatically performs a 360° panoramic scan of the wall, floor, ceiling, beam and column joints, pipeline interfaces, and other component junctions along the path, achieving data acquisition during inspection.

[0102] After the scanned data is uploaded, the cloud-based stitching algorithm first identifies the overlapping features of different acquisition points to eliminate viewing angle deviations and light and shadow interference. Then, a seamless panoramic space is generated through spherical projection. Users can achieve 360° roaming, zoom in to view component dimensions, and switch between different acquisition points to form a path-based spatial browsing on the control and display layer 500. To make the panoramic images measurable, the device 210 incorporates an improved multi-beacon point pose calibration method. By setting up reference points such as wall QR codes with known coordinates on site, the device automatically identifies the reference points and calculates the camera's precise pose in the construction site coordinate system during acquisition, ensuring that the positional error does not exceed 5 cm and the angle error does not exceed 1°, thereby assigning each panoramic image a spatiotemporal coordinate of time and place.

[0103] The feature recognition module 330 addresses the challenges of scarce data, dim lighting, and component occlusion at construction sites by building its own specialized dataset for construction features. This dataset covers 12 types of components, including masonry structures, electromechanical pipelines, and decorative surfaces, comprising approximately 100,000 labeled images. Based on the Mask2Former semantic segmentation model, parameters are optimized through transfer learning. This enables pixel-level contour delineation of features such as walls and pipelines even in complex environments, achieving an average intersection-union ratio of 0.9536 and a false negative rate of less than 2%. The recognition results are further used for 3D component localization: pixels of the same component identified in panoramic images from different perspectives are back-projected into spatial rays originating from the camera's optical center.

[0104]

[0105] in For the first The optical center of each collection point Let be the unit direction vector corresponding to the pixel. For any two rays, find the midpoint of their common perpendicular segment as a candidate intersection point. The candidate point is obtained by solving the following system of linear equations, obtaining the parameters, and then taking the midpoint.

[0106]

[0107] Subsequently, an improved DBSCAN density clustering method was used to cluster all candidate intersection points. The centroid of the cluster with the most points was taken as the 3D coordinates of the component in the real world. The dispersion of points within the cluster reflects the reliability of the localization, thus overcoming the problem that monocular vision cannot measure distance. Cluster neighborhood radius Take a distance of 1 to 3 centimeters and a minimum number of neighboring points of 3 to 5, which can be determined according to the density of points on site.

[0108] The structural progress observations obtained from the reverse modeling module 310 and the installation and completion progress observations obtained from the feature recognition module 330 must ultimately be converged into a unified, reliable, and synchronized progress judgment in the progress fusion module 340. This convergence is completed using the BIM work breakdown structure as a common framework. (See process details below.) Figure 3The progress fusion module 340 organizes the BIM digital twin base 400 into a tree structure by project, plot, building, floor, and component, and further distinguishes work packages such as masonry, plastering, and electromechanical installation at each component node. Each on-site observation is marked on the leaf node corresponding to the unique component and work package binary (S201). The reverse modeling module 310 converts the height measurements of buildings and floors into the completion ratio of structural work packages, and the element identification module 330 converts the wall masonry height, pipe installation length, ceiling and paint construction area, etc., into the completion ratio of the corresponding work packages. The two fall on different leaf nodes but roll up to the same parent node, thus naturally realizing the combined evaluation of internal and external progress at the floor and building levels, rather than separate management and isolation.

[0109] Let a leaf node At the assessment time Previously accumulated observation set Each observation in the set Give the original completion rate With the time of collection The reliability is calculated based on its source (S202). For external observations, the reliability decreases with increasing observation distance, enhanced residual motion blur, and increased proportion of component obstruction.

[0110]

[0111] In the formula The distance from the component to the camera, The occlusion ratio of components in the model, To address residual blur after image stabilization, To normalize the distance to approximately 50 meters, This represents the fuzziness sensitivity coefficient. For internal observations, its reliability increases as the pose calibration residual decreases, the recognition confidence increases, and the dispersion of the backprojection cluster decreases.

[0112]

[0113] In the formula To calibrate the pose residuals of this acquisition, For segmentation mask confidence, To determine the dispersion of the local cluster, Take 5 cm A 2mm margin is chosen to match the modeling accuracy. Considering the ongoing construction, the representativeness of older observations should decrease; therefore, each observation is weighted with a weight that decays exponentially with time intervals.

[0114]

[0115] in Assuming a decay time constant and taking one inspection cycle of approximately 7 days, The confidence level of the corresponding source is used. Accordingly, the completion rate of the leaf nodes is obtained by confidence-weighted fusion, and the aggregate confidence level is accumulated in the form of noise based on the mutual independence of each observation (S203).

[0116]

[0117] When external and internal observations give conflicting readings for the same leaf node, weighted fusion allows the more reliable one to dominate, avoiding simply overriding one data source with another; multiple mutually corroborating observations collectively elevate the reading. This confidence level will be passed on to subsequent judgment stages.

[0118] Based on leaf node fusion, the schedule fusion module 340 performs a bottom-up hierarchical consistency projection (S204) to eliminate physically inconsistent schedule combinations that may occur when merging internal and external data. The actual completion of any parent node should be equal to the convolution of the completion of its child nodes weighted by planned workload or planned time.

[0119]

[0120] in for The set of child nodes This represents the workload weight for child nodes. If the parent node also has its own direct observations, such as the direct measurement of floor height by the reverse modeling module 310, then the direct observation and the convolution value of the child nodes are weighted and harmonicized again according to their respective confidence levels.

[0121]

[0122] when and If the value is zero, meaning that the node has neither its own direct observations nor any observations of its child nodes, it is marked as having no valid observations and will not participate in subsequent bias determinations until observations are accumulated. In numerical implementation, a very small positive lower limit is taken for the denominator of the above formula to avoid singularities.

[0123] After reconciliation, two constraints are applied (S205). The first is a time monotonicity constraint; since completed entities do not revert, the current value is bounded by the historical fused value. ,in This represents the fusion completion rate of the node in the previous inspection cycle, thus mitigating the abnormal drop in progress caused by a noise observation that is too low due to obstruction or poor viewing angle. Secondly, there is the constraint of process sequence. Considering that in a construction network, a work package is often controlled by several parallel preceding work packages, let's denote... The set of prerequisite work packages is For example, if plastering of the same object is controlled by its masonry work, or electromechanical installation on the same work surface can be controlled by multiple preceding processes, then the completion of subsequent work packages must not significantly exceed the completion of their most tightly controlled preceding process.

[0124]

[0125] Tolerance A value of approximately 0.1 is used to correct situations where the identification process misjudges a subsequent process as completed when its preceding process has not yet been reached. The combined effects of credibility weighting, time decay, hierarchical convolutional harmonics, monotonicity, and process-based dual constraints anchor observations from two separate data sources onto the same BIM component tree, prioritizing them based on credibility rather than source. This results in a consistent, objective, and real-time progress across the entire scene, from component to individual unit.

[0126] The node-by-node completion rate output by the above fusion Its confidence level The deviation determination and correction module 350 drives the automatic determination of deviations and resource correction. See the flowchart below. Figure 4 The progress visualization uses colors to distinguish between floors and work areas that are in normal, lagging, and ahead of schedule. The planned completion rate is determined by the planned start and end times of each component in the 4D BIM, with the deviation being the difference between the merged completion and the planned completion (S301).

[0127]

[0128] In the formula , For the start and end times of the plan, Negative values ​​indicate lag, and positive values ​​indicate lead. The judgment no longer uses a fixed 5% deviation threshold; instead, the alarm threshold is adaptively adjusted based on the observation confidence level, the progress fluctuation of the component's construction stage, and whether the component is on the critical path (S302). The effective uncertainty of the deviation is first expressed as a synthesis of the observation uncertainty and the historical fluctuations of the stage.

[0129]

[0130] in Basic observation standard deviation The stage to which the component belongs The historical deviation standard deviation is used, with values ​​calibrated separately for each stage: foundation, structure, electromechanical, and decoration; to ensure calculability, values ​​are used in the above formula. The lower limit is a very small positive number. , that is to Substituting, when there is no on-site observation of the component and... When the hysteresis uncertainty approaches zero, its effective uncertainty becomes very large, and a warning is not triggered accordingly, consistent with the expectation that false alarms should not be triggered when there is no reliable data. Under this premise, the lower the confidence level, the greater the uncertainty, meaning that unreliable observations require a larger deviation to trigger an alarm, thus suppressing false alarms; when the hysteresis deviation exceeds several times its effective uncertainty and still exceeds the limit after key tightening, a warning is triggered.

[0131]

[0132] In the formula To establish confidence, we take approximately 1.64 to correspond to approximately 95% confidence on one side. Characterized by criticality; 1 is assigned when the path is critical, otherwise 0. By setting the threshold to approximately 0.5, the alarm thresholds for critical path components are tightened, allowing problems to be detected earlier. Leaf node alarms are rolled up to the work surface and plot level according to workload weights. When the convolution deviation in a certain area is continuous... When an inspection cycle, approximately 3 days, continues to exceed the boundary, an upgrade notification is sent. The deviation judgment and correction module 350 notifies the management personnel via platform pop-ups and SMS messages and simultaneously marks the deviation area. At the same time, the original images, recognition reports, and progress data are archived to the digital archive module 370 by timestamp for traceability.

[0133] After the warning is triggered, the deviation judgment and correction module 350 checks the lagging area. The cause of the deviation is attributed (S303), and correlated with the statistical results of the labor statistics module 360 ​​and the weekly material report data. The actual working hours of each job type are obtained by the work site camera and AI algorithm. The actual quantity of various materials entering the site is obtained from the material ledger. The shortage rate was obtained by comparing it with the planned value.

[0134]

[0135] The dominant cause is the product of the shortage rate and the share of that resource in the region's remaining workload, whichever is greater. ,in For job type In the lagging region The remaining work's planned percentage and material costs are similarly assessed to determine whether the delay stems from insufficient labor or delayed material delivery, thus generating a weekly construction report. When attributing the delay to insufficient labor, the module further provides quantifiable personnel allocation suggestions (S304). Record the current time. To the next milestone The remaining working days between The lagging area At The additional working hours required to return to the plan, and the supply areas that are either prior to or on schedule. The available surplus working hours are estimated as follows:

[0136]

[0137] In the formula , Each is a job type within the corresponding area. The planned total man-hours, As a transferable proportion, and taken to be approximately 0.3, both equations are in the dimension of time. From the region... Transferred to the region Jobs The number of working hours that personnel can contribute during the remaining construction period is ,in For single-person daily working hours, To allocate personnel, the objective of finding a personnel transfer plan is to minimize the weighted deficit under the constraint of surplus supply.

[0138]

[0139] in The priority is set for lagging regions, and the value is higher for critical path regions. This indicates that the non-negative part is taken; the units of both the objective function and the constraints are time, and the time contributed by the allocation is based on the daily output. With remaining working days The decision is made jointly. In implementation, a greedy algorithm can be used, prioritizing lagging areas from highest to lowest priority. Manpower is then allocated sequentially from the supply area with the largest surplus and closest proximity until the deficit is filled or the surplus is exhausted, thus providing managers with specific personnel to allocate based on job type. In this way, the perception, fusion, judgment, and correction of progress form a closed loop within the platform. For example, when the structural construction of a certain plot is several days behind schedule, the system automatically links the area's imagery with personnel statistics, identifies a shortage of steelworkers causing the delay, and pushes quantitative personnel allocation suggestions. Managers can then implement adjustments after confirming the site status through panoramic images. The effective uncertainty relied upon by the aforementioned adaptive judgment is directly taken from the confidence level output by the progress fusion module 340. The regional progress on which deviations and attributions are based is also taken from the same fusion result harmonized by hierarchical convolution, ensuring that reliability information permeates the judgment and decision-making process. The two work together to transform the original progress management—characterized by manual recording, delayed reporting, internal and external disconnect, and separation of planning and execution—into an automated, objectively weighted, internally and externally unified, and closed-loop corrective progress management system.

[0140] To support the above judgments and attributions, the labor force statistics module 360 ​​completes dynamic labor force statistics and performance monitoring. The module establishes a hat color classification model through sample training. After connecting to the work surface camera, it performs pixel analysis on the head area of ​​personnel in the image, distinguishing between the special-colored hats of management personnel and the safety helmets of ordinary workers. It also combines human posture recognition to confirm identity and avoid misidentification. When the target moves on the work surface, it uses edge computing to match the hat color and height / body type features in the images of adjacent cameras in real time, automatically switching the tracking lens. Even with brief obstruction, it can re-lock based on movement trajectory prediction, achieving uninterrupted tracking of the work surface. When management personnel enter the work surface and meet the special dress requirements to begin performing their duties, the module captures the first image and marks the start time stamp and the work surface location in associated BIM coordinates. At the end of their duties, it captures the last image and marks the end point. These two images are bound and stored with the timing data for verification, preventing substitute work and falsified records. By combining automatic job identification and work area division, the module can statistically analyze the number of laborers and job composition in each area in real time and perform weekly trend analysis, providing data for deviation judgment and resource allocation of the 350 correction module.

[0141] The panoramic images, recognition results, and progress data of the entire construction process are structured and stored by the digital archive module 370 according to the time and space dimensions, and are linked and bound to the BIM model, progress data, and responsible person information of the corresponding stage to build a visualized digital archive that supports full-process backtracking. The panoramic view of the foundation stage can trace the foundation excavation status, the panoramic view of the main structure stage can compare the beam and slab pouring quality, and the panoramic view of the decoration stage can verify the wall plastering effect. It not only provides intuitive evidence for tracing the source of quality problems, but also serves as a site image archive for the digital delivery of the project and subsequent operation and maintenance. The control and display layer 500 organizes the above capabilities into a progress monitoring module 510, an image management module 520, a resource management module 530, and an early warning and optimization module 540. The progress monitoring module 510 supports overall progress overview by plot and building, tracking of sub-projects, and monitoring of the planned and actual time differences of key nodes. The image management module 520 supports image classification and storage, multi-time-point comparison, and bidirectional association of images with BIM components. The resource management module 530 implements personnel allocation assessment, equipment operation monitoring, and material inventory early warning. The early warning and optimization module 540 is responsible for deviation early warning push, resource allocation simulation, and optimization suggestion output. The progress monitoring module 510 generates a dynamic progress model along the timeline on a monthly basis, constructing progress models for the core tube, outer frame, and podium. The core tube model presents the pouring height and construction nodes at each stage, the outer frame model shows the beam and column installation sequence and completion range, and the podium model marks the progress of masonry and roofing works. Managers can slide the timeline to view the progress status of each part at different times, intuitively grasping differences such as the core tube being ahead of schedule and the outer frame lagging behind.

[0142] In specific implementations, this invention can have various equivalent modifications. External data acquisition, besides the tower crane PTZ camera 110 and the perimeter fixed camera 120, can be replaced or supplemented by drone patrol images, which, after undergoing the aforementioned cross-scale preprocessing and coordinate alignment, participate in reverse modeling. The angle of the light-shielding plate 232 of the panoramic scanning device 210, in addition to being triggered by overexposure detection, can also be continuously adjusted in a closed loop based on the ambient illuminance measured by the image sensor 241. In addition to uploading image data via a two-level link between the low-latency Bluetooth module 242 and the edge gateway 250, when a stable wireless LAN or cellular coverage is available on-site, the forwarding capability of the edge gateway 250 can be integrated into inspection handheld terminals and other devices, without changing the implementation path of first short-range aggregation and then forwarding to the cloud via a wide area network. The semantic segmentation model used by the element recognition module 330 can be equivalently replaced by other pixel-level segmentation networks, and the density clustering used for component 3D positioning can also be replaced by other clustering methods based on nearest-neighbor density, without changing the implementation path of determining the component's 3D coordinates through multi-view ray intersection. The credibility sub-item in the schedule fusion module 340 can be added or removed according to on-site conditions. For example, data latency can be introduced as an additional reduction factor in areas with limited network conditions. Parameters such as confidence multipliers, critical tightening coefficients, decay time constants, process tolerances, and minimum lower limits of each denominator in the deviation judgment and correction module 350 can be recalibrated using historical data from different projects. When the schedule is changed from a linear assumption to an S-curve assumption, only the expression of the planned completion percentage needs to be replaced accordingly, while the judgment and correction logic remains unchanged. With the above approach, modeling can achieve centimeter-level accuracy and is approximately three times more efficient than traditional methods. The schedule judgment error can be controlled within 3%, lower than the error level of manual statistics. The model size is compressed by more than 70%, and the loading time is shortened to less than 3 seconds. This enables automated perception, unified evaluation, and collaborative decision-making of schedules in complex projects with overlapping internal and external operations.

[0143] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A BIM-based 360° full-scene construction progress management method, characterized in that, Includes the following steps: Obtain structural progress observations from the exterior of the building and interior progress observations from the interior of the building, and mark each progress observation on the leaf node of the corresponding component and work package binary in the BIM work breakdown structure; The reliability of each progress observation is calculated based on the data source and observation quality, and a weight that decays with time interval is applied to obtain the comprehensive weight of each progress observation. Based on the comprehensive weight, all progress observations under the same leaf node are weighted and fused to obtain the fusion completion ratio of the leaf node, and the aggregation confidence of the leaf node is calculated based on the comprehensive weight of each progress observation. The convolution operation is performed from bottom to top using the fusion completion ratio of child nodes and the corresponding workload weights. The resulting convolution value is then weighted and harmonicized with the direct observation of the parent node itself based on confidence, to obtain the initial completion ratio of each node after hierarchical consistency projection. By simultaneously applying time monotonicity constraints and process sequence constraints to the initial completion ratio, the final completion ratio of each node is obtained. The progress deviation is calculated based on the final completion ratio and the planned completion ratio of each node, and the deviation alarm threshold is dynamically adjusted according to the aggregate confidence level to perform adaptive deviation judgment and progress warning.

2. The method according to claim 1, characterized in that, The methods for calculating the overall weight and the fusion completion ratio of the leaf nodes specifically include: External structural progress monitoring Its credibility calculation model is as follows: ; in, The distance from the component to the camera. The occlusion ratio of the component in the model. For residual ambiguity, For distance normalization scale, The fuzzy sensitivity coefficient; For internal indoor progress monitoring Its credibility calculation model is as follows: ; in, To calibrate the residuals of the acquired pose, To segment the mask confidence level, To determine the dispersion of the local cluster, and For the corresponding precision scale; The combined weight of the decay of each progress observation over time is: ,in , For the assessment time, For the time of data collection, The decay time constant; The leaf node Integration completion rate and aggregate confidence The calculation model is as follows: ; in, For the evaluation time Previous leaf nodes The accumulated set of observations The original completion percentages are as follows.

3. The method according to claim 1, characterized in that, The calculation method for bottom-up hierarchical consistency projection specifically includes: through child nodes completion rate and its workload weight Calculate the parent node The actual completion ratio of the convolution value : ; If the parent node has a directly observed fusion completion rate and corresponding aggregate confidence level Then, the harmonic completion ratio is obtained by harmonic adjustment of the convolution value with the corresponding confidence level. : ; in, The convolution confidence is obtained by weighting the confidence of child nodes by workload.

4. The method according to claim 1 or 3, characterized in that, The specific constraint rules for the time monotonicity constraint and the process sequence constraint include: using historical fusion values ​​to measure the completion ratio of the current node. Apply time monotonicity lower bound clamp: ; in, This represents the fusion completion rate of this node in the previous cycle; based on the previous work package set. Apply an upper limit constraint, controlled by parallel preceding processes, to the completion rate of the current node: in, Pre-work package The completion rate, This is to account for the tolerance of misjudgment during the process.

5. The method according to claim 1, characterized in that, The alarm triggering condition for the adaptive deviation determination is as follows: ,in To establish confidence in the multiplier, To incorporate the schedule deviation between the completed and planned completion dates, the effective uncertainty is... With key tightening factors The calculation models are as follows: ; in, The calculated value is the aggregate confidence level after taking the effective lower bound. Basic observation standard deviation The standard deviation of the historical deviation of the component's stage. For critical path identification, This is the tightening coefficient; when a progress warning is triggered, the cause of the deviation is attributed, and if the cause is insufficient labor, a quantitative personnel allocation plan is output with the goal of minimizing the weighted deficit: ; in, Lagging region The required working hours For supply area Surplus working hours For the region Transferred to the region The number of people to be deployed, The remaining working days. For single-person daily working hours, Priority is given to areas with lagging development.

6. The method according to claim 1, characterized in that, The specific steps for obtaining structural progress observations from the exterior of the building include: Collect high-altitude video from tower cranes and ground monitoring video, minimizing adjacent frames. and Between residuals Solve for the motion vectors and perform deconvolution deblurring compensation on the high-altitude frames of the tower crane high-altitude video; The ground surveillance video is reconstructed using super-resolution and the perspective deviation is corrected using a radial distortion model. Using building corner points and site benchmark control points as shared feature points, the constraint is minimized. The similarity transformation parameters between high-altitude video and ground monitoring video are solved to achieve alignment in the same world coordinate system. Semantic segmentation is used to embed the components identified by ground monitoring into the occluded areas of the tower crane high-altitude model, generating a seamless real-scene model as the data basis for external structure progress observation.

7. The method according to claim 1, characterized in that, The specific steps for obtaining indoor progress observations within the building include: performing a panoramic scan along the indoor inspection path and generating a stitched panoramic space; identifying target component elements based on a semantic segmentation model; back-projecting the same component pixels identified in the panoramic images from different acquisition points into spatial rays originating from the camera's optical center; and solving for the midpoint of the common perpendicular segment of any two rays as candidate intersection points. ; in, and The coordinates of the optical center of the acquisition point. and This is the unit direction vector corresponding to the pixel. and The linear equations formed by the ray direction and the coordinates of the optical center are solved; all the candidate intersection points are clustered using a clustering algorithm based on nearest neighbor density, and the centroid of the cluster with the most points is extracted as the true coordinates of the component in three-dimensional space. The dispersion of the points in the cluster is characterized as the localization cluster dispersion of the internal observation.

8. A BIM-based 360° full-scene construction progress management system, characterized in that, The system uses the method as described in any one of claims 1-7, the system comprising: The perception access module is used to acquire structural progress observations outside the building and indoor progress observations inside the building, and to map each progress observation to the leaf nodes of the corresponding component and work package tuple in the BIM work breakdown structure. The confidence quantification module is used to calculate the confidence and comprehensive weight of each progress observation based on data source characteristics, observation quality indicators and time decay constant. The weighted fusion module is used to perform weighted fusion of multi-source heterogeneous progress observations at the same leaf node level, using the comprehensive weight as the weight parameter, and outputs the fusion completion ratio and aggregation confidence of the leaf node. The hierarchical harmonization module is used to perform convolution projection on the fusion completion ratio of child nodes according to the workload weight from bottom to top according to the BIM tree structure, and harmonize it with the completion ratio directly observed by the parent node according to their respective confidence levels. The dual-clamp module is used to synchronously apply a time monotonicity lower limit clamp based on historical data to the reconciled schedule status, and a process sequence upper limit clamp controlled by the preceding parallel processes, to establish a physically self-consistent schedule benchmark. The adaptive control module is used to dynamically set the deviation alarm threshold based on the aggregate confidence level, historical fluctuations of the stage, and critical path identification to perform deviation judgment, and generate a closed-loop deviation correction and allocation scheme attributable to the corresponding resource shortage after the deviation alarm is triggered.

9. The system according to claim 8, characterized in that, The sensing access module includes an internal panoramic inspection subsystem, which is equipped with an indoor construction environment acquisition panoramic scanning device (210) for roving scanning. This device specifically includes: The main support module (220) includes an adjustable boom (221) with an adjustable acquisition angle, a mounting column (222) and a high-definition PTZ camera (223). The transmission protection module (230) includes a micro motor, a light shield connecting ring (231) connected to the micro motor, and a light shield (232) installed on the light shield connecting ring (231); wherein, when the system is exposed to strong light and triggers overexposure of the screen, the micro motor is controlled to start and drive the light shield connecting ring (231) and the light shield (232) to rotate to the optimal blocking angle to eliminate glare; The data processing module (240), including an image sensor (241) and a low-latency Bluetooth module (242), is used to transmit the digitized acquired images to the edge gateway (250) at the construction site and offload them to the cloud.

10. The system according to claim 8, characterized in that, The system is also equipped with a lightweight modular module that interacts with the BIM digital twin base. This module is specifically used for: Based on the semantic segmentation network, the global scene point cloud is split into two levels of hierarchical fine segmentation according to the main building blocks and component units, and semantic clusters with independent construction semantic attributes are extracted and generated. Density-priority thinning is performed using voxels as units, and the results are evaluated using calculation formulas. Calculate any point cloud points within each semantic block. The local density is calculated, and only the single representative point with the highest density estimate is retained within each independent voxel, where... To select the nearest neighbor number, For the first The distance between the nearest neighbors is limited by a very small positive lower bound to avoid division by zero singularity; Spatial indexes are established for 3D model units that have undergone precision controllable and lightweight processing. In the front-end interaction, corresponding block models are issued and rendered online as needed based on view coordinates or BIM component click commands.