Highway subgrade receiver collaborative metering system based on live-action three-dimensional model

Abstract

The invention relates to the technical field of engineering metering, in particular to a highway subgrade receiver collaborative metering system based on a live-action three-dimensional model, and the system comprises a data collection module which is used for the data collection of multi-source equipment, so as to obtain subgrade basic data; the data preprocessing module carries out standardization processing on the roadbed basic data to obtain preprocessed data; the live-action three-dimensional model generation module generates a roadbed model through an SfM algorithm and an MVS algorithm; the metering platform performs feature extraction on the roadbed model to obtain point location engineering quantity metering data, length engineering quantity metering data and area engineering quantity metering data; and the report module is used for acquiring data of the metering platform to generate a project quantity report. According to the highway subgrade receiver collaborative metering system based on the live-action three-dimensional model, point location, linearity and area engineering quantity metering can be integrated, and the problems that a traditional metering method is low in precision and low in efficiency are solved.

Description

A Collaborative Measurement System for Highway Subgrade Measurement Based on Real-Scene 3D Model Technical Field

[0001] This invention relates to the field of engineering measurement technology, and in particular to a collaborative measurement system for highway subgrade measurement based on a real-world 3D model. Background Technology

[0002] Highway quantity measurement is a crucial aspect of highway construction, directly impacting project cost control and overall efficiency. Current highway construction quantity measurement technologies primarily rely on manual measurement or semi-automatic tools. For example, traditional point measurement depends on manual counting; linear length measurement (such as guardrail lines) is unsuitable for on-site changes; and area measurement (such as pavement paving) is susceptible to large errors and inefficiency due to terrain variations. In subgrade measurement, manual counting of point facilities (such as anchor bolts and settlement monitoring points) is prone to omissions due to their dispersed distribution; measurement of linear facilities (such as subgrade ditches and guardrail foundations) is delayed due to dynamic adjustments in fill / cut; and for area measurements (such as fill / cut area and slope surface area), the large topographical undulations lead to manual measurement errors exceeding 15%. Furthermore, existing measurement methods lack point-line-area data linkage, failing to meet the precise and efficient requirements of subgrade measurement. These methods are time-consuming, inaccurate, and easily affected by dynamic construction changes, leading to settlement delays and cost control issues.

[0003] Compared to existing technologies, such as the CN 112288848 B patent, which uses drone aerial photography to generate 3D models, the measurement function still requires users to manually select points or delineate boundaries, essentially making it an advanced "digital ruler." Another example is the drone-based automatic measurement platform, which requires users to log in and manually measure distances, areas, and volumes, or click to select slope ranges in the 3D view (lacking fully automated processing). These systems cannot achieve intelligent identification and automated batch measurement of engineering objects, especially for roadbed measurement scenarios. Existing solutions do not have dedicated logic for roadbed slope area segmentation and point-to-linear data linkage measurement design, requiring manual adjustment of the measurement range. This makes it difficult to adapt to dynamic adjustments in roadbed construction (such as changes in temporary filling areas), resulting in low efficiency and unsuitability for complex dynamic construction scenarios. In particular, it still relies on manual intervention in multi-task lists and result viewing. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a collaborative measurement system for highway subgrade measurement based on a real-world 3D model. This system integrates point, linear, and area measurement of engineering quantities, solving the problems of low accuracy and low efficiency in traditional measurement methods.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A collaborative measurement system for highway subgrade measurement based on a real-scene 3D model includes a data acquisition module, a data preprocessing module, a real-scene 3D model generation module, a measurement platform, and a reporting module.

[0007] The data acquisition module is used to acquire data from multiple sources to obtain roadbed foundation data, which includes point data, linear path data, and area data.

[0008] The data preprocessing module is used for data acquisition by the data acquisition module, and the data preprocessing module performs standardization processing on the roadbed foundation data to obtain preprocessed data;

[0009] The real-scene 3D model generation module is used for data acquisition by the data preprocessing module, and the real-scene 3D model generation module generates a roadbed model through the SfM algorithm and the MVS algorithm.

[0010] The measurement platform is used for data acquisition by the real-scene 3D model generation module, and the measurement platform performs feature extraction on the roadbed model to obtain point engineering quantity measurement data, length engineering quantity measurement data and area engineering quantity measurement data.

[0011] The reporting module is used to acquire data from the measurement platform to generate engineering quantity reports.

[0012] Furthermore, the data acquisition module collects roadbed foundation data through handheld GPS, UAV LiDAR, vehicle-mounted rangefinder, ground scanner, and multispectral imaging.

[0013] Furthermore, the data preprocessing module processes the roadbed foundation data sequentially through noise filtering, time-series calibration, anomaly detection, and integrity check to obtain preprocessed data;

[0014] The data preprocessing module performs the following steps for data processing:

[0015] S1.1 The statistical outlier removal algorithm will be used to filter the discrete noise points in the point cloud of the roadbed foundation data to obtain noise-reduced data;

[0016] S1.2 The noise reduction data from step S1.1 is used to register multiple time-series data using the ICP algorithm to process device motion distortion in order to obtain calibration data;

[0017] S1.3 The calibration data obtained in step S1.2 is processed using the isolated forest algorithm to remove outliers, thereby obtaining outlier-removed data;

[0018] S1.4 The abnormal data removed in step S1.3 is compared with the acquisition path and the design path using a length verification algorithm to obtain preprocessed data.

[0019] Furthermore, the real-scene 3D model generation module sequentially calculates the camera pose through SfM, generates dense point clouds through MVS, and forms a roadbed model through meshing;

[0020] The expression function for constructing the roadbed model by the real-scene 3D model generation module is:

[0021] Formula (1)

[0022] in, For the roadbed model, it represents the set of vertices and faces. ;

[0023] For SfM modules; Given a set of multi-view images as input. ; This is the output of the SfM module. , For the camera rotation matrix set, , For each image, there is a camera rotation matrix. Let be the set of camera translation vectors. , This is the camera translation vector for each image; For sparse 3D point clouds, , This refers to the location of traceable feature points in the image in three-dimensional space. This represents the number of sparse points;

[0024] For the MVS module, Dense point cloud as output of the MVS module , , For dense point clouds The total number of points in, and Much larger ;

[0025] For gridded modules, Output a triangular mesh model with topological structure; , , , This is a set of triangular facet indices.

[0026] 5. A highway subgrade measurement system based on a real-world 3D model according to claim 4, characterized in that: the subgrade model is subject to prior constraints based on pavement flatness, lane geometry, subgrade slope gradient, and subgrade base flatness.

[0027] Formula (2)

[0028] in, Loss of photometric uniformity; Geometric priors; To smooth the prior; For roadbed slope constraints; For the planarity constraint of the roadbed base; The weight hyperparameter is used to adjust the quality of the mesh.

[0029] Furthermore, the measurement platform includes a feature extraction module, a point quantity module, a length quantity module, and an area quantity module.

[0030] The feature extraction module extracts features from the roadbed model through point cloud clustering, path tracing, and the U-Net algorithm.

[0031] The point engineering quantity module is used for the feature vector input of the point, and the point engineering quantity data within the region is obtained through the real scene 3D model generation module and the feature extraction module.

[0032] The length engineering quantity module is used for the feature vector input of the engineering quantity length, and the length engineering quantity measurement data is obtained through the real scene 3D model generation module and the feature extraction module.

[0033] The area engineering quantity module is used for the feature vector input of the engineering quantity area, and the area engineering quantity measurement data is obtained through the real scene 3D model generation module and the feature extraction module.

[0034] Furthermore, the expression for feature extraction performed by the feature extraction module is as follows:

[0035] Formula (3)

[0036] in, The output is a set of road elements. ; For roadbed model; For U-Net networks, the parameters are: Input a BEV image and output a semantic probability map for each pixel; Perform Softmax or Argmax operations to obtain a discrete semantic label graph; For the back projection operation, the BEV label map is mapped back to the original point cloud to obtain a point set with semantic labels. ; For the reason The projected bird's-eye view includes channels for elevation, intensity, and density. To perform spatial clustering based on semantic categories, each It is a connected object, where each cluster object Its eigenvectors are defined as follows: ; For the geometric feature calculation module, for each cluster object Extract structured attributes.

[0037] Furthermore, the point-based engineering quantity module obtains the feature vector of a single point through the real-scene 3D model generation module and the feature extraction module, and trains an MLP model based on the historical data of the feature vector of a single point to obtain a point-based MLP model. The point-based MLP model is used to obtain the classification and count prediction of the points, and the classification and count prediction of the points are identified in the roadbed model through an AI model to calculate the total number of points in the roadbed model to obtain length engineering quantity measurement data. The point-based engineering quantity module outputs the length engineering quantity measurement data through the report module.

[0038] Furthermore, the length engineering quantity module extracts the total length, segment length, and type features through the real-scene 3D model generation module and the feature extraction module, and trains an LSTM model based on the labeled data of the total length, segment length, and type features to obtain a length LSTM model. The length LSTM model is used to obtain line type prediction and length estimation, and the line type prediction and length estimation are identified in the roadbed model through an AI model. Length measurement is performed in the roadbed model to obtain the length engineering quantity measurement data. The length engineering quantity module outputs the length engineering quantity measurement data through the report module.

[0039] Furthermore, the area engineering quantity module extracts the total area, sub-region area, and material features through the real-scene 3D model generation module and the feature extraction module. Based on the samples of the total area, sub-region area, and material features, it trains a Mask R-CNN model to obtain an area Mask R-CNN model. The area Mask R-CNN model is used to obtain surface segmentation and area estimation. The surface segmentation and area estimation are then identified on the roadbed model using an AI model. Area calculation is performed on the roadbed model to obtain area engineering quantity measurement data. The area engineering quantity module outputs the area engineering quantity measurement data through the report module.

[0040] The beneficial effects of this invention are:

[0041] With the help of the data acquisition module, roadbed foundation data can be collected using handheld GPS, drone LiDAR, vehicle-mounted rangefinders, ground scanners, and multispectral imaging devices. Simultaneously, the data preprocessing module standardizes the foundation data, providing necessary data support for subsequent real-scene 3D model generation and measurement platforms. The real-scene 3D model generation module provides an intuitive understanding of the highway construction process, realistically reproducing geometric information such as road alignment, cross-section, slopes, culverts, and bridges. The roadbed model integrates information from multiple disciplines, including route, roadbed, pavement, bridges and culverts, traffic safety, and pipelines, ensuring that the quantities of each discipline are neither duplicated nor omitted, which is helpful for construction and monitoring. The system enables communication and review among project managers, cost estimators, and other stakeholders based on the same data source, reducing disputes. The measurement platform, through a newly added receiving management module, deeply integrates receiving scenarios with measurement technology, resolving the core dispute in traditional measurement methods where the "measurement scope is disconnected from the receiving scope." Simultaneously, based on the roadbed model and its characteristics, it can directly obtain point quantity measurement data, length quantity measurement data, and area quantity measurement data from the roadbed model, improving measurement accuracy and efficiency. It also supports dynamic adjustments during construction, applicable to point, linear, and area measurement, optimizing material usage, settlement efficiency, and project cost control, avoiding human error risks, and improving the economy and safety of highway engineering. The reporting module can output project location, design quantity, measured quantity, discrepancy, timestamp, and model snapshot, and directly uploads them to the project management database via API, supporting platform result viewing and improving the management efficiency of highway construction. This invention addresses the needs of roadbed measurement, achieving a counting accuracy of ≥98% for anchor bolts, pile positions, and other point facilities, reducing the omission rate by 25% compared to manual counting. It also reduces the measurement deviation of linear facilities such as roadbed ditches and guardrail foundations to <1%, adapting to dynamic construction adjustment requirements. Furthermore, it improves the measurement accuracy of surface areas such as roadbed filling and excavation, and slope surfaces by 20%, resolving errors caused by terrain undulations. Through point-line-surface collaborative logic, it reduces measurement disputes caused by isolated data in roadbed measurement, shortens the settlement cycle by over 30%, and saves 10-15% on material usage, making it particularly suitable for scenarios involving frequent adjustments to filling and excavation during roadbed construction. Attached Figure Description

[0042] Figure 1 is a structural block diagram of a highway subgrade volume measurement system based on a real-world 3D model according to a preferred embodiment of the present invention.

[0043] Figure 2 is a schematic diagram of the roadbed model structure of a roadbed measurement system based on a real-scene 3D model according to a preferred embodiment of the present invention.

[0044] Figure 3 is a schematic diagram of the roadbed model measurement of the roadbed measurement collaborative measurement system based on a real-scene 3D model according to a preferred embodiment of the present invention.

[0045] In the diagram, 1-Data acquisition module, 2-Data preprocessing module, 3-Real-scene 3D model generation module, 4-Measurement platform, 41-Feature extraction module, 42-Point quantity module, 43-Length quantity module, 44-Area quantity module, 45-Survey management module, and 5-Report module. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0048] Referring to Figure 1, a preferred embodiment of the highway subgrade measurement system based on a real-scene 3D model of the present invention includes a data acquisition module 1, a data preprocessing module 2, a real-scene 3D model generation module 3, a measurement platform 4, and a report module 5.

[0049] Data acquisition module 1 is used for data acquisition from multiple sources to obtain roadbed foundation data, which includes point data, linear path data, and area data.

[0050] Data acquisition module 1 collects roadbed foundation data via handheld GPS, UAV LiDAR, vehicle-mounted rangefinder, ground scanner, and multispectral imaging. In this embodiment, the equipment is configured as follows: the handheld GPS first collects control points to calibrate the UAV LiDAR data; acquisition conditions: outdoor environment with sufficient sunlight and a temperature of 10-30°C, avoiding rainy or foggy weather; the device communicates with data acquisition module 1, and the collected data is automatically uploaded to the task list.

[0051] In this embodiment, the UAV LiDAR flight altitude is set to 50-100m (to suit the roadbed construction site range), and the point cloud density is ≥100 points / ㎡ (to meet the needs of slope surface detail acquisition); the handheld GPS control point acquisition density is one every 50m (to suit the linear extension characteristics of the roadbed); the acquisition conditions need to avoid the dusty period of roadbed construction, and for fill / cut areas, the acquisition frequency is increased to once every 2 days to adapt to dynamic terrain changes.

[0052] The data preprocessing module 2 is used to acquire data from the data acquisition module 1, and the data preprocessing module 2 performs standardization processing on the roadbed foundation data to obtain preprocessed data.

[0053] The data preprocessing module 2 processes the subgrade foundation data sequentially through noise filtering, time-series calibration, anomaly detection, and integrity check to obtain preprocessed data;

[0054] The data processing steps performed by data preprocessing module 2 include:

[0055] S1.1 A statistical outlier removal algorithm will be used to filter discrete noise points in the point cloud of the roadbed foundation data to obtain denoised data; the threshold in this embodiment is set to 3. The 3σ threshold effectively filters out invalid data from discrete noise caused by construction machinery obstruction in roadbed point cloud data.

[0056] S1.2 The noise-reduced data from step S1.1 is used to register multi-time-series data using the ICP algorithm to process device motion distortion and obtain calibration data; the number of iterations in this embodiment is <50.

[0057] S1.3 The calibration data from step S1.2 is processed using the Isolation Forest algorithm to remove outliers, resulting in outlier-removed data; in this embodiment, the contamination rate is set to 0.1.

[0058] S1.4 The abnormal data removed in step S1.3 is compared with the collected path and the design path using a length verification algorithm to obtain preprocessed data; the deviation threshold in this embodiment is <5%. The 5% deviation threshold is adapted to the reasonable deviation between the design path and the actual path in roadbed construction (such as temporary adjustments in the embankment area) to avoid excessive removal of valid data.

[0059] The real-scene 3D model generation module 3 is used to acquire data from the data preprocessing module 2, and the real-scene 3D model generation module 3 generates the roadbed model through the SfM algorithm and the MVS algorithm.

[0060] The real-scene 3D model generation module 3 sequentially calculates the camera pose through SfM, generates dense point clouds through MVS, and forms a model through meshing to construct the roadbed model.

[0061] The expression function for constructing the roadbed model in the real-scene 3D model generation module 3 is as follows:

[0062] Formula (1)

[0063] in, For the roadbed model, it represents the set of vertices and faces. ;

[0064] For SfM modules; Given a set of multi-view images as input. ; This is the output of the SfM module. , For the camera rotation matrix set, , For each image, there is a camera rotation matrix. Let be the set of camera translation vectors. , This is the camera translation vector for each image; For sparse 3D point clouds, , This refers to the location of traceable feature points in the image in three-dimensional space. This represents the number of sparse points;

[0065] For the MVS module, Dense point cloud as output of the MVS module , , For dense point clouds The total number of points in, and Much larger ;

[0066] For gridded modules, Output a triangular mesh model with topological structure; , , , This is a set of triangular facet indices.

[0067] Prior constraints are applied to the roadbed model based on pavement flatness, lane geometry, roadbed slope gradient, and roadbed base flatness:

[0068] Formula (2)

[0069] in, Loss of photometric uniformity; Geometric priors; To smooth the prior; For roadbed slope constraints; For the planarity constraint of the roadbed base; The weight hyperparameter is used to adjust the quality of the mesh.

[0070] In this embodiment, the slope constraint of the roadbed side slope is adapted to the undulation characteristics of the roadbed slope, and the planarity constraint of the roadbed base ensures the uniformity of the measurement benchmark for the filling and excavation areas.

[0071] The measurement platform 4 is used for data acquisition by the real-scene 3D model generation module 3, and it also performs feature extraction on the roadbed model to obtain point quantity measurement data, length quantity measurement data, and area quantity measurement data. In this embodiment, the point quantity measurement data consists of the counts of discrete point facilities such as roadbed anchors and piles; the length quantity measurement data consists of the length measurements of linear facilities such as roadbed side ditches and guardrail foundations; and the area quantity measurement data consists of the calculations of the roadbed fill area, cut area, and slope surface area.

[0072] The measurement platform 4 includes a feature extraction module 41, a point quantity module 42, a length quantity module 43, an area quantity module 44, and a measurement management module 45.

[0073] The feature extraction module 41 extracts features from the roadbed model through point cloud clustering, path tracing, and the U-Net algorithm.

[0074] In this embodiment, the expression for feature extraction performed by the feature extraction module 41 is:

[0075] Formula (3)

[0076] in, The output is a set of road elements. ; For roadbed model; For U-Net networks, the parameters are: Input a BEV image and output a semantic probability map for each pixel; Perform Softmax or Argmax operations to obtain a discrete semantic label graph; For the back projection operation, the BEV label map is mapped back to the original point cloud to obtain a point set with semantic labels. ; For the reason The projected bird's-eye view includes channels for elevation, intensity, and density. To perform spatial clustering based on semantic categories, each It is a connected object, where each cluster object Its eigenvectors are defined as follows: ; For the geometric feature calculation module, for each cluster object Extract structured attributes.

[0077] In this embodiment, the output road element set This includes the characteristics of roadbed point facilities (anchor diameter, pile spacing), linear facility characteristics (side ditch cross-sectional dimensions, roadbed centerline slope), and area characteristics (slope gradient, fill / cut compaction correlation parameters).

[0078] In this embodiment:

[0079] coordinate: Formula (4)

[0080] in, It is a point vector;

[0081] length: Formula (5)

[0082] in, To extract the centerline through path tracing;

[0083] area: Formula (6)

[0084] in, It is a planar projection;

[0085] type: Formula (7)

[0086] in, The mode label;

[0087] Material: Formula (8)

[0088] in, This represents the RGB mean of the point cloud.

[0089] The point engineering quantity module 42 is used for the input of feature vectors of the points, and obtains the point engineering quantity data in the area through the real scene 3D model generation module 3 and feature extraction module 41.

[0090] The point-based engineering quantity module 42 obtains the feature vectors of individual points through the real-scene 3D model generation module 3 and the feature extraction module 41. Based on historical data of the feature vectors of individual points, it trains an MLP model to obtain the point-based MLP model. The point-based MLP model is used to obtain the classification and count prediction of points. The classification and count prediction of points are then identified in the roadbed model using an AI model to calculate the total number of points in the roadbed model, thereby obtaining the length engineering quantity measurement data. The point-based engineering quantity module 42 outputs the length engineering quantity measurement data through the report module 5. The feature vectors of individual points include roadbed anchors, pile locations, etc.

[0091] In this embodiment, the MLP model input is a feature vector of a single point (coordinates [3D], color [RGB], intensity [1D], dimension 7), trained based on historical data to obtain a point-based MLP model. The feature vector of the input point in the point-based MLP model can be directly output from the roadbed model to count the number of points within the output area. The point-based MLP model is a 3-layer MLP with linear activation in the output layer, replacing manual counting on the platform. Through the collaboration between the AI ​​model and the roadbed model, no manual interaction is required; the user can select the area, but the system processes it automatically.

[0092] The Point Quantity Module 42 improves measurement accuracy and efficiency through realistic 3D models, supports dynamic adjustments during construction, and is suitable for roadbed and node facility measurement. Ultimately, it achieves efficient calculation of multiple quantities of engineering work, improving cost control and schedule management of highway projects.

[0093] The length engineering quantity module 43 is used for the feature vector input of the engineering quantity length, and the length engineering quantity measurement data is obtained through the real scene 3D model generation module 3 and the feature extraction module 41.

[0094] The length engineering quantity module 43 extracts the total length, segment length, and type features through the real-scene 3D model generation module 3 and the feature extraction module 41. Based on the labeled data of the total length, segment length, and type features, it trains an LSTM model to obtain a length LSTM model. The length LSTM model is used to obtain alignment prediction and length estimation. The alignment prediction and length estimation are then identified in the roadbed model using an AI model, and length measurements are performed in the roadbed model to obtain length engineering quantity measurement data. The length engineering quantity module 43 outputs the length engineering quantity measurement data through the report module 5. In this embodiment, the total length, segment length, and type features are the total length, segment length, and type features of the roadbed side ditch and guardrail foundation; the labeled data is the labeled data based on the roadbed linear facilities.

[0095] In this embodiment, the specific steps for training an LSTM model based on labeled data to achieve alignment prediction and length estimation are as follows: Prepare a labeled sequence of roadbed ditches and guardrail foundations (>500 samples, including construction dynamic adjustment data) → Train the LSTM (2 layers, 128 hidden units) → Input path points and output predicted length (MSE loss, training time <2h). By combining the AI ​​model with the roadbed model, the length can be measured automatically directly on the roadbed model, replacing platform distance measurement.

[0096] The Length Quantity Module 43 improves measurement accuracy and efficiency through realistic 3D models, supports dynamic adjustments during construction, and is suitable for slope and drainage system measurement. Ultimately, it enables accurate length quantity calculation, optimizes material usage, and improves settlement efficiency.

[0097] The area engineering quantity module 44 is used for the feature vector input of the engineering quantity area, and obtains the area engineering quantity measurement data through the real scene 3D model generation module 3 and the feature extraction module 41.

[0098] The area engineering quantity module 44 extracts the total area, sub-region area, and material features through the real-scene 3D model generation module 3 and the feature extraction module 41. Based on the samples of the total area, sub-region area, and material features, it trains a Mask R-CNN model to obtain the area Mask R-CNN model. The area Mask R-CNN model is used to obtain surface segmentation and area estimation. The surface segmentation and area estimation are identified in the roadbed model by an AI model, and the area is calculated in the roadbed model to obtain the area engineering quantity measurement data. The area engineering quantity module 44 outputs the area engineering quantity measurement data through the report module 5.

[0099] The sample training in this embodiment is for the roadbed surface area, which includes the total area of ​​roadbed fill and cut, slope surface, sub-region area, and material characteristics.

[0100] In this embodiment, the specific steps and principles for implementing surface segmentation and area estimation based on sample-trained Mask R-CNN model are as follows: labeled roadbed fill and cut, slope surface samples (>1000 images, including different slope and compaction scenarios) → training (backbone ResNet, RPN to generate proposals) → output mask and estimated area. This embodiment is based on instance segmentation and regression, replacing the platform's manual area measurement.

[0101] The relationship and difference between area estimation and area calculation: estimation is a preliminary value that the model outputs quickly (for dynamic adjustment), while calculation is an accurate integral (based on the segmentation mask, which improves efficiency compared to platform volume measurement).

[0102] The area quantity module 44 improves measurement accuracy and efficiency through realistic 3D models, supports dynamic adjustments during construction, is applicable to pavement and slope measurement, and ultimately realizes automated area quantity calculation, improving the economy and accuracy of highway engineering.

[0103] By combining AI models with roadbed models, the area can be calculated directly on the 3D model through boundary tracing and integration. The platform supports slope area calculation but is fully automatic.

[0104] The acceptance management module 45 is used to acquire data from the feature extraction module 41 to obtain the features of the roadbed elements, delineate the roadbed acceptance area and associate it with the construction ledger. The acceptance management module 45 sends the feature data within the acceptance area to the point quantity module 42, the length quantity module 43 and the area quantity module 44. The acceptance management module 45 generates the preliminary review results of the roadbed acceptance and sends them to the report module 5.

[0105] In this embodiment, the acceptance management module 45 serves as the core implementation platform for the roadbed acceptance scenario, primarily realizing three major functions: acceptance area delineation, targeted allocation of measurement data, and preliminary review of acceptance results. Firstly, acceptance area delineation supports precise delineation of the constructed roadbed acceptance area (such as fill sections, cut sections, and slope protection sections) in the real-world 3D model through two methods: "point-selecting boundaries" and "batch importing construction sections," and automatically associates it with construction ledger information (including construction date, construction team, and supervisor confirmation records). Secondly, targeted allocation of measurement data uses "regional boundary masking" technology to mark the acceptance area, and then extracts the effective roadbed acceptance data within the area using a feature filtering algorithm. Basic feature characteristics (such as anchor bolt locations, ditch paths, and slope topography data within the area) are directionally transmitted to the point quantity module 42, length quantity module 43, and area quantity module 44. Only the quantity of work within the acceptance area is calculated, excluding unconstructed areas and interfering data. Thirdly, the acceptance results are preliminarily reviewed. The "point-line-area" quantity data output by each measurement module is initially verified, and the deviation between the design quantity and the measured quantity is compared (an early warning is automatically triggered when the deviation exceeds 5%). A "Preliminary Acceptance Review Form for Roadbed" is generated, which includes a screenshot of the acceptance area, preliminary measurement results, and deviation explanations. This is then linked to the supervision end for online confirmation. The input of the acceptance management module 45 is the roadbed feature characteristics and construction ledger data output by the feature extraction module 41. The output is the directionally allocated acceptance area feature data and the "Preliminary Acceptance Review Form for Roadbed", realizing a deep binding between the acceptance scenario and measurement technology, ensuring that the measurement range completely matches the actual construction acceptance area.

[0106] Report module 5 is used to acquire data from measurement platform 4 to generate engineering quantity reports.

[0107] In this embodiment, the report module 5 can output the engineering location, design quantity, actual measurement quantity, difference, timestamp, and model snapshot, and directly upload them to the project management database via API, supporting platform result viewing and improving the management efficiency of highway engineering construction.

[0108] The core framework of this embodiment is data acquisition → 3D modeling → AI automatic identification and measurement. It introduces an AI kernel to achieve fully automated identification, classification and batch processing, rather than manual interaction (such as the platform requiring left-click selection and right-click rotation), thus improving its applicability to dynamic construction scenarios.

[0109] By using real-world 3D models and AI-automated measurement, the accuracy (error rate reduced to within 5%) and efficiency of measurement are improved, measurement time is shortened by more than 50%, dynamic adjustment of construction is supported, and it is applicable to point measurement, such as the number of anchor bolts, linear measurement, guardrail length, and area measurement, such as road surface area measurement.

[0110] Data support: In the simulation test, the accuracy of point counting reached 98%, based on 1000 anchor bolt samples, which is 30% better than manual counting on the platform; the linear length prediction deviation was <1%, with a test path of 10km, which is 20% faster than manual path tracking on the platform; the accuracy of area calculation was improved by 20%, with a test area of ​​5000㎡, which reduces human error in slope measurement compared to the platform.

[0111] Ultimately, this optimizes material usage by 10-15%, shortens the settlement cycle by 30%, and controls project costs, while avoiding human-related risks such as rotation / scaling errors during manual platform operation, thus improving the economy and safety of highway engineering.

[0112] Figure 2 is a partial schematic diagram of the roadbed model in this embodiment. Basic operations: ① Left-click to pan, ② Right-click to rotate, ③ Scroll wheel to zoom, ④ Press and hold Ctrl + left mouse button during measurement to pan the model during measurement.

[0113] Figure 3 shows a schematic diagram of the roadbed model measurement. Clicking the "Measure" function allows you to perform operations such as measuring "distance, angle, height, area, and volume," with text prompts for specific measurement operations. Clicking the "Analyze" function allows you to measure the slope area using the "Slope Surface Area" function. You can select the area to be measured on the model by clicking "Start Measurement" and "Double-click to End."

Claims

1. A collaborative measurement system for highway subgrade volume measurement based on a real-scene 3D model, characterized in that, The system includes a data acquisition module (1), a data preprocessing module (2), a real-scene 3D model generation module (3), a measurement platform (4), and a report module (5). The data acquisition module (1) is used to acquire data from multiple sources to obtain roadbed foundation data, which includes point data, linear path data, and area data. The data preprocessing module (2) is used to acquire data from the data acquisition module (1) and performs standardization processing on the roadbed foundation data to obtain preprocessed data. The real-scene 3D model generation module (3) is used to acquire data from the data preprocessing module (2) and generates a roadbed model using the SfM algorithm and the MVS algorithm. The measurement platform (4) is used to acquire data from the real-scene 3D model generation module (3) and performs feature extraction on the roadbed model to obtain point engineering quantity measurement data, length engineering quantity measurement data, and area engineering quantity measurement data. The report module (5) is used to acquire data from the measurement platform (4) to generate an engineering quantity report.

2. The highway subgrade volume measurement system based on a real-scene 3D model according to claim 1, characterized in that: The data acquisition module (1) acquires roadbed foundation data through handheld GPS, UAV LiDAR, vehicle-mounted rangefinder, ground scanner and multispectral imaging.

3. The highway subgrade volume measurement system based on a real-scene 3D model according to claim 1, characterized in that: The data preprocessing module (2) processes the roadbed basic data sequentially through noise filtering, time-series calibration, anomaly detection, and integrity check to obtain preprocessed data; The data preprocessing module (2) performs the following steps for data processing: S1.1 Filtering discrete noise points in the point cloud of the roadbed foundation data using a statistical outlier removal algorithm to obtain denoised data; S1.2 Registering the denoised data from step S1.1 with multi-time series data using the ICP algorithm to process equipment motion distortion to obtain calibration data; S1.3 Removing outliers from the calibration data from step S1.2 using the isolated forest algorithm to obtain outlier-removed data; S1.4 Comparing the collected path with the designed path using a length verification algorithm on the outlier-removed data from step S1.3 to obtain preprocessed data.

4. The highway subgrade volume measurement system based on a real-scene 3D model according to claim 1, characterized in that: The real-scene 3D model generation module (3) sequentially calculates the camera pose using SfM, generates dense point clouds using MVS, and forms a grid to construct the roadbed model; the expression function for constructing the roadbed model by the real-scene 3D model generation module (3) is: Formula (1) where, For the roadbed model, it represents the set of vertices and faces. ； For SfM modules; Given a set of multi-view images as input. ； This is the output of the SfM module. ， For the camera rotation matrix set, ， For each image, there is a camera rotation matrix. Let be the set of camera translation vectors. ， This is the camera translation vector for each image; For sparse 3D point clouds, ， This refers to the location of traceable feature points in the image in three-dimensional space. This represents the number of sparse points; For the MVS module, Dense point cloud as output of the MVS module ， ， For dense point clouds The total number of points in, and Much larger ； For gridded modules, Output a triangular mesh model with topological structure; ， ， ， This is a set of triangular facet indices.

5. A highway subgrade volume measurement system based on a real-scene 3D model according to claim 4, characterized in that: Prior constraints are applied to the roadbed model based on pavement flatness, lane geometry, roadbed slope gradient, and roadbed base flatness: Formula (2) where, Loss of photometric uniformity; Geometric priors; To smooth the prior; For roadbed slope constraints; For the planarity constraint of the roadbed base; The weight hyperparameter is used to adjust the quality of the mesh.

6. The highway subgrade volume measurement system based on a real-scene 3D model according to claim 1, characterized in that: The measurement platform (4) includes a feature extraction module (41), a point quantity module (42), a length quantity module (43), an area quantity module (44), and a receiving management module (45). The feature extraction module (41) extracts features from the roadbed model through point cloud clustering, path tracing, and the U-Net algorithm. The point quantity module (42) is used for the feature vector input of the point, and obtains the point quantity data within the area through the real-scene 3D model generation module (3) and the feature extraction module (41). The length quantity module (43) is used for the feature vector input of the length of the quantity, and obtains the length data within the area through the real-scene 3D model generation module (3) and the feature extraction module (41). The length engineering quantity measurement data; the area engineering quantity module (44) is used for the feature vector input of the engineering quantity area, and the area engineering quantity measurement data is obtained through the real scene three-dimensional model generation module (3) and the feature extraction module (41); the acceptance management module (45) is used for the data acquisition of the feature extraction module (41) to obtain the subgrade element features, delineate the subgrade acceptance area and associate it with the construction ledger, the acceptance management module (45) sends the feature data in the acceptance area to the point engineering quantity module (42), the length engineering quantity module (43) and the area engineering quantity module (44), and the acceptance management module (45) sends the generated subgrade acceptance preliminary review result to the report module (5).

7. A highway subgrade measurement system based on a real-scene 3D model, as described in claim 6, is characterized in that: The feature extraction module (41) performs feature extraction using the following expression: Formula (3) where, The output is a set of road elements. ； For roadbed model; For U-Net networks, the parameters are: Input a BEV image and output a semantic probability map for each pixel; Perform Softmax or Argmax operations to obtain a discrete semantic label graph; For the back projection operation, the BEV label map is mapped back to the original point cloud to obtain a point set with semantic labels. ； For the reason The projected bird's-eye view includes channels for elevation, intensity, and density. To perform spatial clustering based on semantic categories, each It is a connected object, where each cluster object Its eigenvectors are defined as follows: ； For the geometric feature calculation module, for each cluster object Extract structured attributes.

8. A highway subgrade measurement system based on a real-scene 3D model, as described in claim 6, characterized in that: The point quantity module (42) obtains the feature vector of a single point through the real scene 3D model generation module (3) and the feature extraction module (41), and trains an MLP model based on the historical data of the feature vector of a single point to obtain the point MLP model. The point MLP model is used to obtain the classification and count prediction of the points, and the classification and count prediction of the points are identified in the roadbed model through the AI ​​model to calculate the total number of points in the roadbed model to obtain the length engineering quantity measurement data. The point quantity module (42) outputs the length engineering quantity measurement data through the report module (5).

9. A highway subgrade volume measurement system based on a real-scene 3D model according to claim 6, characterized in that: The length engineering quantity module (43) extracts the total length, segment length and type features through the real scene 3D model generation module (3) and the feature extraction module (41), and trains an LSTM model based on the labeled data of the total length, segment length and type features to obtain a length LSTM model. The length LSTM model is used to obtain line type prediction and length estimation, and the line type prediction and length estimation are identified in the roadbed model through an AI model. The length is measured in the roadbed model to obtain the length engineering quantity measurement data. The length engineering quantity module (43) outputs the length engineering quantity measurement data through the report module (5).

10. A highway subgrade volume measurement system based on a real-scene 3D model according to claim 6, characterized in that: The area engineering quantity module (44) extracts the total area, sub-region area and material features through the real scene 3D model generation module (3) and the feature extraction module (41), and trains the Mask R-CNN model based on the samples of the total area, the sub-region area and the material features to obtain the area Mask R-CNN model. The area Mask R-CNN model is used to obtain the surface segmentation and area estimation. The surface segmentation and the area estimation are identified in the roadbed model through the AI ​​model. The area is calculated in the roadbed model to obtain the area engineering quantity measurement data. The area engineering quantity module (44) outputs the area engineering quantity measurement data through the report module (5).

Images