A method and system for indoor measurement robot BIM collaborative mapping and quality assessment

Abstract

The application discloses a kind of indoor survey robot BIM collaborative mapping and quality evaluation method and system, including BIM model analysis and measurement task generation, autonomous measurement site planning, path planning and motion sequence generation, autonomous inspection and multi-source data acquisition and reference unification, multi-source data fusion and three-dimensional reconstruction, BIM and measured model component-level alignment matching, component-level deviation calculation and quality evaluation;The application generates task automatically and plans site path autonomously by BIM semantic driving, converts the site planning process traditionally relying on manual experience into a structured automated generation process, reduces manual intervention, improves the consistency of measurement results in different operating scenarios, and reduces the dependence on operator experience.

Description

Technical Field

[0001] This invention relates to the field of digital measurement and intelligent construction technology in building engineering, specifically to an indoor measurement robot BIM collaborative mapping and quality assessment method and system. Background Technology

[0002] With the ongoing digital transformation of the construction industry, Building Information Modeling (BIM) has gradually become the core data carrier throughout the entire lifecycle of engineering construction. Simultaneously, the rapid development of spatial information acquisition technologies such as 3D laser scanning and mobile robot measurement has provided technical support for the rapid acquisition of the actual geometric state of the construction site. Comparing and analyzing measured spatial data with the BIM design model to achieve quantitative assessment of construction deviations has become an important technical approach to improving building quality management.

[0003] Currently, the measurement and quality inspection of indoor building spaces are still mainly carried out manually. Workers rely on their on-site experience to set up scanning stations, and complete the data collection of a single station by manually moving, setting up, and leveling the stations. Then, they complete the point cloud stitching of multiple stations by manually selecting and registering the points, and finally overlay and compare the point cloud data with the BIM model. This approach has several limitations: First, site layout and path planning heavily rely on operator experience, lacking a standardized generation mechanism. Inconsistent measurement results from different operators are common, and issues such as insufficient coverage or repeated scanning in local areas can arise, affecting measurement efficiency and data uniformity. Second, in existing technological systems, BIM models are primarily used as reference benchmarks for post-comparison, failing to play a spatial constraint and guiding role in front-end stages such as measurement task planning, on-site data acquisition, and point cloud registration. This leads to a disconnect between BIM design information and the actual measurement process, failing to fully leverage the spatial prior value of the BIM model. Third, existing comparison methods often employ point-level distance calculations, directly calculating the nearest neighbor distance from the point cloud to the BIM model surface. This method is susceptible to interference from factors such as point cloud density, occlusion, noise, and temporary objects on-site, making it difficult to generate object-oriented quality conclusions for specific engineering components. The engineering interpretability and verifiability of deviation results are insufficient. Fourth, multi-site point cloud registration lacks reliable initial pose constraints, heavily relying on manual point selection or feature matching. This results in low automation and a tendency for registration deviations to accumulate, affecting the accuracy of the final comparison results.

[0004] To address the aforementioned issues, the industry has conducted relevant technological research. Some solutions introduce mobile robot platforms to achieve semi-automated measurement, using SLAM technology for autonomous navigation and data acquisition. However, the path planning in these solutions often only considers geometric obstacle information, failing to fully utilize the semantic information and spatial topological relationships of building components. The measurement path lacks an effective correlation with component distribution, and the layout of measurement stations has not yet formed a BIM-based structured optimization mechanism. Furthermore, most solutions still position BIM as a post-processing comparison tool, failing to establish a BIM prior constraint system that runs through the entire measurement process. This makes it difficult to fundamentally solve the problems of heavy reliance on manual labor in the measurement process and insufficient refinement of quality assessment. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for indoor measurement robot BIM collaborative mapping and quality assessment, so as to solve the problems existing in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for BIM collaborative mapping and quality assessment of indoor measurement robots, comprising the following steps: Step S1: BIM model semantic parsing and measurement task pre-generation. Perform component parsing and spatial topology analysis on the input BIM model, divide the indoor semantic space and construct a passable grid map to generate an initial set of measurement tasks. Step S2: BIM-constrained measurement site autonomous planning generates candidate sites within the traversable area. After accessibility filtering, visibility analysis, and component importance weighted scoring, the optimal set of measurement sites is obtained through comprehensive optimization of coverage benefits and path costs. Step S3: BIM topology-guided inspection path planning, constructing a site connectivity network based on indoor spatial topology, solving for the optimal site visit sequence, and generating robot motion path and site observation posture; Step S4: Multi-source data acquisition and spatial benchmark unification. The robot performs multi-source sensor data acquisition along the planned path, establishes the transformation relationship between the BIM coordinate system, the robot base coordinate system and the scanner coordinate system, and realizes the coordinate benchmark unification of multi-source data. Step S5: Multi-source data fusion and 3D measurement model construction. Based on pose information, the initial stitching of point clouds from multiple sites is completed. Fine registration is achieved through global optimization. After denoising and homogenization, an indoor 3D measurement model in a unified coordinate system is generated. Step S6: Component-level alignment and matching between BIM and measured models. Line and surface geometric primitives are extracted from the BIM model and the measured 3D model respectively. Following the matching principle of surface first and then line, a one-to-one correspondence between theoretical primitives and measured primitives is established through type constraints, geometric consistency constraints and spatial overlap constraints. Step S7: Component-level deviation calculation and construction quality assessment. Based on the matched line and surface primitives, calculate the spatial deviation index of various building components, determine the qualification of components according to the graded engineering quality threshold, and generate structured quality assessment results.

[0007] Furthermore, step S1 specifically includes: The BIM model is analyzed layer by layer to separate the geometric and semantic information of each building component, and the component types are classified and labeled. Perform topological analysis on the spatial adjacency relationships of each component to construct a network of component connections and a diagram of room connectivity. Based on the geometric information and topological relationships of the components, the interior space is divided into room units, corridor units and functional area units, thus completing the semantic division of the interior space; Generate a spatial rasterized representation model within the passable area and establish an observable association mapping between raster units and building components; Various building components and their observable areas are transformed into measurement targets, and an initial set of measurement tasks is generated by combining the importance of the components with the spatial coverage requirements.

[0008] Furthermore, the optimal set of measurement stations in step S2 is solved using the following objective function: In the formula, For the selected set of sites; To detect the coverage of the target object; This represents the total path length. The cost of turning, detouring, and repeated observations; , , These are the weighting coefficients; During the candidate site generation phase, the sampling density is adjusted according to the density of components in the spatial region. Dense sampling is used in densely populated areas, while sparse sampling is used in open areas. The visibility analysis calculates the visibility range and occlusion ratio of a single site to surrounding components using ray projection. The importance weight of components is set according to the structural and functional attributes of the components, with load-bearing structural components having a higher weight than decorative components. The generated optimal measurement site set includes the spatial coordinates of the sites, the topological adjacency relationships between the sites, and the initial access sequence information.

[0009] Furthermore, step S3 specifically includes: Based on the indoor space topology connectivity graph and the passable grid map, each measurement station is used as a graph node and the passable paths between stations are used as graph edges to construct a station connectivity network model. The optimal site access sequence is solved based on the site connectivity network. The path between sites is generated by combining the indoor obstacle distribution and wall boundary constraints. The path optimization objectives include minimizing the total path length and minimizing the number of turns. The generated path is spatially feasible, and the path is checked point by point to see if it crosses an obstacle area. Impassable path segments are locally modified and replanned. Configure observation attitude parameters for each measurement station, including horizontal yaw angle and pitch angle, so that the core observation components are within the central field of view of the scanning equipment.

[0010] Furthermore, the spatial reference unification of multi-source data in step S4 is achieved through the following coordinate transformation relationship: In the formula, , , The coordinates of the point in the BIM coordinate system; , , The coordinates of the point in the scanner coordinate system; The extrinsic transformation matrix from the scanner coordinate system to the robot base coordinate system is obtained through device calibration. The pose transformation matrix from the robot's base coordinate system to the BIM coordinate system is obtained through building structure feature registration and pose calculation. The sequence of measurement stations and spatial topology generated in step S2 are used to provide initial pose constraints for multi-station point cloud registration in step S5, thus narrowing the solution range for fine registration.

[0011] Furthermore, step S5 specifically includes: Initial stitching of multi-site point cloud data is performed based on the initial pose information of each site to complete coarse alignment; A global optimization method is used to perform fine registration of the point clouds of all stations, and the pose parameters of each station are optimized by utilizing the overlapping areas of adjacent stations to eliminate pose accumulation errors. Point cloud data is filtered and denoised, and density is equalized to remove discrete outliers and to homogenize areas with density differences. Structural consistency optimization is performed on planar components such as walls and floors, and linear features such as edges and corners, to weaken the impact of point cloud noise on geometric features while retaining construction deviations. Data from all sites is integrated to generate an indoor 3D measured point cloud model in a unified coordinate system.

[0012] Furthermore, the unified expression form of the line-surface geometric primitives in step S6 is as follows: In the formula, It is a set of BIM theoretical primitives. Let m be the theoretical surface primitive. Let k be the basic element of the theoretical line; For the set of measured primitives, For the nth measured surface primitive, Let t be the basic element of the measured line; The matching degree of surface primitives is determined using the following comprehensive scoring function: In the formula, This represents the vertical offset from the measured surface to the theoretical surface. The angle between the normals of the two surfaces; The spatial overlap rate of the two sides; The fitting residual of the measured plane; to These are the weighting coefficients for each indicator; When the comprehensive score meets the preset threshold and the spatial overlap rate meets the requirements, the correspondence between the theoretical surface primitive and the measured surface primitive is established.

[0013] Furthermore, in step S6, the matching of line elements is carried out based on the adjacency relationship of the already matched surface elements, and the directional deviation of the line elements is calculated by the following formula: In the formula, Let the direction vector be the theoretical line element. The direction vector of the measured line element; The distance between line elements is calculated using the following formula: In the formula, These are sampling points on the measured line element. The total number of sampling points. The spatial distance function from a point to a theoretical line element; By combining directional deviation, line distance, endpoint overlap, and the relationship between the corresponding surface elements, the matching consistency between theoretical line elements and measured line elements is determined.

[0014] Furthermore, the process of determining the engineering quality threshold in step S7 is as follows: The first step is to extract the corresponding deviation limits for various components such as walls, floors, doors and windows, beams and columns, based on the allowable deviation values ​​of the components. The second step is to classify the components according to their structural importance. Load-bearing walls, structural columns, main beams and other key components are set as the first level, and thresholds that are stricter than the standard benchmarks are used. Non-load-bearing partitions, decorative components and other general components are set as the second level, and standard benchmark thresholds are used. The third step is to make adaptation adjustments based on the current construction stage. The precision requirements of the corresponding construction procedures are matched for the main structure stage, rough decoration stage, and fine decoration stage, forming staged thresholds. The fourth step involves customized adjustments based on the project's specific accuracy requirements, ultimately forming a two-level quality judgment threshold system that includes qualified and excellent thresholds. At the same time, an out-of-tolerance warning threshold is set to mark components that are approaching the limit.

[0015] Furthermore, the indicators for calculating component-level deviations in step S7 include wall position offset, wall flatness, floor elevation deviation, bay depth, door and window opening dimensions, and squareness of internal and external corners; After the deviation calculation is completed, the deviation results of various components are statistically analyzed to generate a spatial error distribution map, and the out-of-tolerance components are spatially located and labeled. For components that are missed or obstructed, resulting in mismatch, mark them as low-confidence results and generate supplementary measurement prompts to guide the location and range of supplementary measurements.

[0016] An indoor measurement robot BIM collaborative mapping and quality assessment system includes a BIM semantic parsing and task planning module, an autonomous site planning module, a path planning module, a multi-source data acquisition and benchmark unification module, a 3D reconstruction module, a component-level matching analysis module, and a quality assessment module. The output of the BIM semantic parsing and task planning module is connected to the input of the autonomous site planning module. It is used to parse the component information and spatial topology of the BIM model and generate a passable grid map and initial measurement tasks. The output of the autonomous site planning module is connected to the input of the path planning module. It is used to generate the optimal set of measurement sites under BIM constraints and output the spatial distribution of sites and the mapping relationship between components. The output of the path planning module is connected to the input of the multi-source data acquisition and benchmark unification module, which is used to construct the site connectivity network and solve the optimal access path, and generate robot motion sequence and site observation posture parameters. The output of the multi-source data acquisition and benchmark unification module is connected to the input of the three-dimensional reconstruction module. It is used to control the robot to perform multi-source data acquisition, establish multi-coordinate system transformation relationships, and output the original measurement dataset under a unified benchmark. The output of the 3D reconstruction module is connected to the input of the component-level matching analysis module, and is used to register, denoise and fuse point cloud data to generate an indoor 3D measured model. The output of the component-level matching analysis module is connected to the input of the quality assessment module. It is used to extract the line and surface geometric primitives of the BIM and the measured model and complete the component-level matching and alignment, and output the component-level deviation data. The quality assessment module is used to determine the compliance of construction quality based on graded quality thresholds and output structured quality assessment results and report documents.

[0017] Furthermore, the autonomous site planning module incorporates a candidate site generation unit, an accessibility filtering unit, a visibility analysis unit, a weight modeling unit, and an optimization filtering unit. The candidate site generation unit is used to generate a set of candidate sites for differentiated sampling based on the density of regional components; The accessibility filtering unit is used to eliminate candidate sites in impassable areas; The visibility analysis unit is used to calculate the observation coverage of each candidate site on surrounding components; The weighted modeling unit is used to assign a comprehensive observation score to candidate sites based on the importance of components; The optimization filtering unit is used to select the optimal site layout through a comprehensive optimization model that considers coverage benefits and path costs.

[0018] Compared with the prior art, the beneficial effects of the present invention are: By using BIM semantic-driven automatic task generation and autonomous site path planning, the traditional point placement and planning process that relies on human experience is transformed into a structured automated generation process, reducing human intervention, improving the consistency of measurement results in different work scenarios, and reducing reliance on operator experience.

[0019] BIM-constrained site layout takes into account both component coverage integrity and path access efficiency, optimizing the number of sites while ensuring coverage. At the same time, the site sequence and topology relationship generated by site planning can provide reliable initial pose constraints for multi-site point cloud registration, avoiding the manual point selection and registration process, and improving the automation and processing efficiency of point cloud post-processing.

[0020] By extracting and matching line and surface geometric primitives, the traditional point-level distance comparison is upgraded to component-level object-oriented deviation calculation, reducing interference from factors such as point cloud noise, occlusion, and temporary objects. The deviation results directly correspond to specific engineering components such as walls, doors, windows, and floors, and can clearly distinguish different types of errors such as overall offset, local flatness, and edge deviation, which facilitates problem location, rectification review, and quality traceability.

[0021] By moving BIM from a single post-processing comparison benchmark to a spatial prior constraint that runs through the entire process of task planning, data acquisition, registration and alignment, and quality assessment, the measurement objects at the front and back ends remain consistent, forming a closed-loop feedback mechanism. For areas with low reliability or missed measurements, it can guide the location and scope of supplementary measurements, thereby improving the completeness and reliability of the measurement results. Attached Figure Description

[0022] Figure 1 This is the overall flowchart of the present invention; Figure 2 This is a schematic diagram of the BIM-based prior constraint-based site planning algorithm of the present invention. Figure 3 This is a flowchart illustrating the component-level alignment and matching process between BIM and the measured model according to the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1-3 This invention provides a method for BIM collaborative mapping and quality assessment of indoor measurement robots, comprising the following steps: Step 1: BIM Model Analysis and Measurement Task Generation The purpose of this step is to convert the BIM model, which contains design geometric and semantic attributes, into a structured measurement task that can be performed by robots, establishing a semantic expression system for the indoor space, and providing unified basic constraint information for subsequent site planning, path generation, and quality assessment. The specific implementation process is as follows: 1.1 Component Analysis and Attribute Separation The input BIM model is processed layer by layer, traversing the building component entities and decomposing the information of each component into two independent dimensions: geometric information and semantic information. For geometric information processing, parameters such as the three-dimensional spatial coordinates, contour dimensions, and spatial orientation of each component are extracted. The surface of the component is then discretized into a mesh representation, forming a geometric surface model that can be used for visual field calculations. For semantic information processing, components are classified and labeled according to their attribute information, clarifying the engineering type of the component, including but not limited to load-bearing walls, non-load-bearing partition walls, structural columns, beams, floor slabs, roof slabs, doors, windows, and openings. Simultaneously, the design parameters, floor, and room associated attributes of the component are recorded, providing a basis for subsequent importance weighting and quality index calculation.

[0025] 1.2 Spatial Relationship Construction Based on the geometric location and semantic attributes of each component, spatial topological relationship analysis is performed to construct a two-level connectivity network. The first level is a component-level adjacency relationship network, recording the spatial connections, adjacencies, and inclusion relationships between components such as walls and beams / columns, walls and doors / windows, and walls and floors, clarifying the geometric constraints between components. The second level is a room-level connectivity network, using passageways such as doorways and corridors as connection media to organize the connectivity paths between various room units, forming a topological connectivity graph of the interior space, which is used for spatial constraints in subsequent path planning.

[0026] 1.3 Semantic Division of Indoor Space By combining the geometric boundaries and topological relationships of building components, semantic zoning of the building's interior space is achieved. Based on spatial function and accessibility, the interior space is divided into different types such as room units, corridor units, stairwell units, and functional area units, clearly defining the spatial boundaries, entrance / exit locations, and connectivity relationships of each unit. This semantic zoning transforms the abstract three-dimensional space into a set of spatial units with engineering semantics, facilitating the zoning organization of measurement tasks and the structured planning of pathways.

[0027] 1.4 Rasterized Spatial Modeling A spatial rasterized representation model is generated within the accessible indoor area. The raster resolution is set according to measurement accuracy requirements, discretizing the continuous space into raster units of uniform size. Based on the spatial positional relationship between each raster unit and surrounding building components, a component observability association model is established, describing the visible range, observation angle, and occlusion of various building components from that raster location, providing basic data for subsequent candidate site scoring calculations. Simultaneously, inaccessible obstacle areas such as walls, columns, and fixed equipment are marked on the raster map, forming a binary representation of the accessible area.

[0028] 1.5 Measurement Task Generation Various building components and their corresponding observable areas are transformed into measurement targets. Based on factors such as the structural importance of the components, quality inspection requirements, and spatial distribution density, the coverage priority and accuracy requirements for each measurement target are set. Measurement targets are grouped and organized according to spatial zoning and component categories to generate an initial set of measurement tasks, completing the mapping transformation from the BIM design model to robot-executable measurement tasks.

[0029] The final output of this step includes: BIM semantic space model, component semantic and geometric information database, indoor space topology connectivity graph, accessible grid map, and initial measurement task set.

[0030] Step 2: Autonomous Measurement Site Planning This step, based on the BIM semantic spatial model, indoor topology connectivity map, and accessible grid map output in step 1, generates the optimal set of measurement stations within the accessible indoor area under the constraints of BIM semantics and spatial structure. This enables the measurement equipment to achieve efficient coverage of the indoor space and target components under limited station conditions. The specific implementation process is as follows: 2.1 Candidate Site Generation Within the accessible grid map area, an initial candidate site set is generated using a differentiated density sampling strategy based on the component distribution characteristics of the spatial region. For areas with dense components and complex structures, such as room corners, around door and window openings, and beam-column joints, a denser sampling method is used to reduce the sampling grid interval and increase the candidate site density. For open spaces with simple components, such as the center of an empty room or the middle section of a long corridor, a sparse sampling method is used to increase the sampling grid interval and reduce redundant candidate sites. Through differentiated sampling, the total number of candidate sites is controlled while ensuring coverage, reducing the complexity of subsequent optimization calculations.

[0031] 2.2 Accessibility Filtering For each candidate site, spatial accessibility analysis is performed. Based on a traversable grid map, a path accessibility detection algorithm is used to determine whether the candidate site is located within a traversable area and whether there is a traversable connecting path between it and other sites. Candidate sites located in obstacle areas, enclosed by walls, or inaccessible by robot movement are eliminated, and all spatially accessible candidate sites are retained to form a filtered candidate site set.

[0032] 2.3 Viewpoint Analysis For each filtered candidate site, a visibility analysis is performed, using ray projection to simulate the observation capability of the measuring equipment from that site to surrounding components. Using the candidate site as the origin, a detection ray is emitted across the entire horizontal angle range. The intersection points of the ray with the surfaces of building components are recorded, and the area covered by the ray for each component is calculated. The visible range and occlusion ratio of various components such as walls, doors, windows, beams, and columns are calculated for that site. For areas with occlusion, the source and extent of the occlusion are recorded, providing a basis for subsequent site scoring and blind spot filling.

[0033] 2.4 Weight Modeling A component importance weighting mechanism is introduced, assigning differentiated weight coefficients to different types of components based on their structural functional attributes and quality inspection requirements. Higher weight values ​​are assigned to load-bearing walls, structural columns, main beams, and other key structural components, as well as critical inspection areas such as corners and opening edges. Normal weight values ​​are assigned to general components such as non-load-bearing partitions and decorative surfaces. Combining the site component coverage area obtained from visibility analysis, a comprehensive observation score is calculated for each candidate site; the larger the area of ​​high-weight components covered by a site, the higher the comprehensive score.

[0034] 2.5 Optimize Filtering A model that comprehensively optimizes coverage benefits and path costs is used to select the optimal set of sites from the candidate site set. The objective function for optimization is: In the formula, The final set of selected sites; The overall coverage of the selected site for all detected objects is obtained by multiplying the coverage area ratio of each component by its corresponding weight and summing the results. The total path length for traversing all selected sites; The additional costs associated with turning, detours, and repeated observations in the path; , , These are the weighting coefficients for coverage, path length, and additional cost, respectively, which can be adjusted according to the project's priority of efficiency or accuracy.

[0035] The optimization solution can be achieved by combining greedy selection with local optimization: First, select stations from high to low based on their comprehensive scores, updating the overall coverage with each station added until the preset coverage requirements are met; then, optimize the spatial uniformity of the selected stations, adjusting their positions to avoid excessive clustering, and adding stations in blind spots to eliminate coverage gaps; finally, an optimal set of measurement stations with reasonable spatial distribution and complete and efficient coverage is formed. Simultaneously, establish the association mapping relationship between each measurement station and the observed components, clarifying the specific component range that each station is responsible for observing.

[0036] The final output of this step includes: the optimal set of measurement sites, site spatial distribution parameters, site-component association mapping relationship, and initial pose information of the sites.

[0037] Step 3: Path planning and motion sequence generation This step, based on the measurement site set and spatial distribution information output in step 2, and combined with the indoor topology connectivity graph from step 1, models the connection relationships between the sites to generate the optimal movement path for the robot in the indoor environment, achieving efficient and orderly access between sites. The specific implementation process is as follows: 3.1 Graph Structure Construction A site connectivity network model is constructed based on an indoor topological connectivity graph and a passable grid map. Each measurement station is treated as a node in the graph, and the passable paths between stations are treated as edges. The weight of each edge is determined by a combination of factors, including path length, traversal difficulty, and number of turns. For stations located in different rooms, connections are established through doorways, corridors, and other connecting passages to ensure that the graph structure matches the actual spatial topology.

[0038] 3.2 Path Optimization Solution Based on the constructed site connectivity network, the optimal site access sequence and connection path are solved. The objectives of path optimization include minimizing the total path length, maximizing site access efficiency, and minimizing the number of turns. A graph search algorithm is used to solve for the globally optimal access order. At the same time, combined with the indoor obstacle distribution and wall boundary constraints, specific travel paths between sites are generated to ensure that the paths are traversable in physical space and conform to the robot's kinematic constraints.

[0039] 3.3 Feasibility Verification After the path is generated, its spatial feasibility is verified. The path is discretized into a continuous sequence of path points, and each point is checked for obstacles, walls, or fixed structures. For path segments with collisions or that are unreachable, local adjustments and replanning are performed to ensure the overall feasibility of the path.

[0040] 3.4 Attitude Planning Add robot observation attitude parameters, including horizontal yaw and pitch angles, to each measurement station. Based on the distribution of the observed components at each station, determine the optimal observation orientation for the robot at that station, ensuring that the core observed components are within the central field of view of the scanning equipment, thus improving the quality of the observation data. For stations requiring multi-directional scanning, set multiple observation attitude angles to ensure complete coverage of surrounding components.

[0041] The final output of this step includes: the robot's global walking path, the sequence of site visits, and the observed posture parameters for each site.

[0042] Step 4: Autonomous Inspection and Multi-Source Data Acquisition and Benchmark Unification In this step, the robot autonomously moves within the indoor environment and completes multi-source data acquisition based on the motion path sequence output in step 3 and the station observation posture parameters. Simultaneously, it establishes transformation relationships between multiple coordinate systems to achieve spatial benchmark unification for all acquired data. The specific implementation process is as follows: 4.1 Autonomous Navigation Execution The robot autonomously navigates and moves according to a planned path sequence. During movement, it estimates its own pose in real time through a simultaneous localization and mapping (SLAM) system and performs positioning calibration at each measurement station to ensure the accuracy of its stopping position and attitude. When moving between stations, the robot travels smoothly along the planned path, and the obstacle avoidance system monitors temporary obstacles on the path in real time, making minor adjustments to the local path when necessary.

[0043] 4.2 Station Stops and Data Collection After arriving at the pre-set measurement station, the robot precisely stops and adjusts the scanning equipment angle according to the planned observation posture to perform multi-source data acquisition. The acquired data types include: LiDAR point cloud data for acquiring 3D spatial geometric information; depth image data for supplementing local structural details; visible light image data for texture recording and semantic recognition; and inertial attitude data for motion state estimation and pose-assisted calculation. After completing scanning and acquisition at all angles at each station, the robot moves to the next station according to the path sequence until data acquisition at all stations is completed.

[0044] 4.3 Unification of Spatial Benchmarks for Multi-Source Data To ensure the accuracy of subsequent point cloud fusion and BIM comparison, a unified spatial coordinate benchmark needs to be established, clarifying the transformation relationships between the BIM coordinate system, the robot base coordinate system, and the scanner coordinate system. The expression for transforming any point from the scanner coordinate system to the BIM coordinate system is: In the formula, , , These are the three-dimensional coordinates of a point in the BIM coordinate system. , , These are the three-dimensional coordinates of the point in the scanner's coordinate system. The extrinsic transformation matrix from the scanner coordinate system to the robot base coordinate system is obtained in advance through the equipment calibration process; The pose transformation matrix from the robot's base coordinate system to the BIM coordinate system can be obtained by matching and solving stable building structural features such as walls, doorways, beams, and columns.

[0045] In practice, initial registration is performed using globally stable structural features such as wall planes, corridor boundaries, and doorway outlines to determine the initial values ​​of the transformation matrix. Then, fine-tuning is done using local features to ensure the accuracy of the coordinate transformation. Simultaneously, the initial pose and topological relationship of the stations generated in step 2 can be directly used as initial constraints for pose transformation, narrowing the solution range and improving registration efficiency and stability.

[0046] 4.4 Data Association Storage All collected multi-source data undergoes time synchronization processing and is aligned according to a unified timestamp. Then, through the aforementioned coordinate transformation relationship, all data is bound to a unified BIM coordinate system reference framework. The collected data is associated and stored with the corresponding robot pose information and site information to form a structured raw measurement dataset, providing a data foundation for subsequent 3D reconstruction.

[0047] The final output of this step includes: raw multi-site point cloud data, multi-view image data, robot pose trajectory data, and unified coordinate system transformation parameters.

[0048] Step 5: Multi-source data fusion and 3D reconstruction This step fuses the point cloud data, image data, and pose trajectory data output from step 4. Through point cloud registration and data denoising algorithms, a unified coordinate system-based 3D indoor measurement model is constructed, realizing the digital representation of the building's interior space. The specific implementation process is as follows: 5.1 Initial Assembly Based on the initial pose information of each station obtained in step 4, the point cloud data of different stations are initially stitched together to achieve coarse alignment. Since the initial pose of each station in the BIM coordinate system has been provided during the station planning stage, the coarse alignment process does not require manual point selection and can be completed directly based on the initial pose values, which greatly improves the efficiency and consistency of coarse registration.

[0049] 5.2 Refined Optimization Based on the initial stitching, a global optimization method is used to perform fine registration of the point clouds at all stations. Utilizing the overlapping regions between adjacent stations, the pose parameters of each station are optimized through an iterative nearest-point algorithm or a primitive matching algorithm, eliminating cumulative pose errors and improving the overall stitching accuracy of the point clouds. The topological adjacency relationships between stations can be used to constrain the solution direction of the registration, avoiding erroneous registration results.

[0050] 5.3 Noise Treatment After point cloud registration, the point cloud data is filtered and denoised. Statistical filtering algorithms are used to remove discrete outliers and eliminate invalid data caused by environmental noise and scanning errors. At the same time, density equalization is performed on the point cloud, downsampling areas with excessive density and interpolating areas with insufficient density to maintain a uniform point cloud density and improve the overall quality of the model.

[0051] 5.4 Structural Consistency Optimization Structural consistency optimization is performed on point clouds of different component regions. For planar components such as walls and floors, the flatness of the point cloud is optimized through planar fitting; for linear features such as edges and corners, the regularity of the features is optimized through line fitting. While preserving the actual construction deviations, the impact of point cloud noise on the geometric features of components is reduced, providing a high-quality data foundation for subsequent primitive extraction.

[0052] 5.5 Model Fusion All processed site data are merged to form a complete indoor 3D measured point cloud model under a unified coordinate system, covering all measured indoor spaces and components.

[0053] The final output of this step is: a measured 3D point cloud model of the building's interior.

[0054] Step 6: Component-level alignment and matching between BIM and the actual measured model This step spatially aligns the BIM design model with the measured 3D model, and achieves precise component-level correspondence through the extraction and matching of line and surface geometric primitives, providing an object-oriented basis for subsequent deviation calculations. The specific implementation process is as follows: 6.1 Unified Representation of Linear and Surface Geometric Elements Both the BIM design model and the measured point cloud model are represented as a unified set of geometric primitives, which are divided into two categories: surface primitives and line primitives. The unified representation is as follows: In the formula, This is the set of BIM theoretical primitives, in which This represents the m-th theoretical plane element, corresponding to planar components such as walls, floors, and roofs in BIM, and includes attributes such as object number, component type, normal vector, center point coordinates, plane equation constants, and spatial boundary range. This represents the k-th theoretical line primitive, corresponding to linear features in BIM such as door opening edges, window opening edges, inside and outside corner lines, and component intersection lines. It includes attributes such as line type, reference point on the line, direction vector, line segment length, and spatial range.

[0055] For the set of measured primitives, where This represents the nth measured surface primitive, which is extracted from the measured point cloud through plane fitting. Let t represent the measured line primitive, obtained from the measured point cloud through linear feature extraction and fitting. The attribute structure of the measured primitive corresponds one-to-one with the theoretical primitive, facilitating subsequent matching and comparison.

[0056] Through the unified expression of line and surface primitives, the original disordered point cloud data is organized into geometric objects that can be directly associated with BIM components. Subsequent deviation calculations no longer rely on random nearest neighbor points, but instead revolve around specific engineering components and features, and the deviation results have clear engineering meanings.

[0057] 6.2 Surface Primitive Matching The matching of theoretical primitives and measured primitives follows the principles of surface first, then line, type constraint, and geometric consistency, and the matching of surface primitives is completed first.

[0058] The surface primitive matching is constrained by object type, normal angle, point-to-surface distance, center position difference, and projection range overlap. A comprehensive scoring function is constructed to quantitatively evaluate the matching degree between each set of theoretical and measured surfaces. The comprehensive scoring function is as follows: In the formula, This represents the vertical offset distance between the measured surface and the theoretical surface. The angle between the normals of the two faces; The overlap ratio of the spatial projection ranges of the two surfaces; The fitting residual of the measured plane reflects the flatness of the measured surface; to The weighting coefficients for each indicator can be adjusted according to the matching accuracy requirements.

[0059] The lower the score, the higher the matching degree between the theoretical surface and the measured surface. When the comprehensive score meets the preset matching threshold and the spatial overlap rate meets the requirements, a one-to-one correspondence between the theoretical surface primitive and the measured surface primitive is established. For theoretical surfaces that do not have a corresponding measured surface, they are marked as missed or obstructed, and their location and reason are recorded to provide guidance for subsequent supplementary measurements.

[0060] 6.3 Line Primitive Matching After completing the surface primitive matching, line primitive matching is performed. Line primitives exist attached to surface primitives; for example, the edge of a doorway is attached to a wall, and the inside and outside corner lines are attached to adjacent walls. Therefore, line primitive matching can be performed within the adjacency range of already matched surfaces, narrowing the matching range and improving matching accuracy. The constraints for line primitive matching include: line type consistency, direction deviation, distance between lines, overlap of endpoint ranges, and correspondence with the corresponding surface primitives.

[0061] The formula for calculating the directional deviation between the theoretical line and the measured line is as follows: In the formula, Let the direction vector be the theoretical line element. The direction vector of the measured line element; The distance between line elements is calculated using the following formula: In the formula, These are sampling points on the measured line element. The total number of sampling points. Let be the spatial distance function from a point to a theoretical line primitive.

[0062] By considering factors such as directional deviation, distance between lines, overlap of endpoints, and relationship with the surface to which they belong, the matching consistency between theoretical line elements and measured line elements is determined, and a one-to-one correspondence is established.

[0063] By using a surface-first-line matching logic and multiple constraint control, the accuracy of component matching can be effectively improved, erroneous matching between similar components can be avoided, and the reliability of subsequent deviation calculation can be ensured.

[0064] The final output of this step includes: BIM-measured alignment model, set of theoretical-measured primitive matching pairs, and component-level spatial correspondence.

[0065] Step 7: Component-level deviation calculation and quality assessment This step calculates the spatial deviation index of various building components based on the matched line-surface primitive pairs, and determines their compliance according to the graded engineering quality threshold, thus forming a structured construction quality assessment result. The specific implementation process is as follows: 7.1 Calculation of component-level deviation index Based on matched surface and line primitives, various typical construction quality deviation indices are calculated, including: (1) Wall position offset: determined by the vertical offset distance of the matching wall. It is obtained directly and reflects the overall positional deviation of the wall. (2) Wall surface flatness: derived from the fitting residuals of the measured surface elements. This reflects the flatness of a specific area of ​​the wall surface; (3) Floor slab elevation deviation: It is obtained from the elevation difference between the theoretical surface and the actual surface of the floor slab, and reflects the elevation deviation of the floor slab; (4) Width and depth dimensions: These are calculated from the measured distance between two sets of opposite walls in the room, reflecting the deviation of the room's net dimensions; (5) Door and window opening dimensions: These are calculated from the measured line element distances of the top, bottom, left, and right sides of the opening, reflecting the dimensional deviation of the opening; (6) Squareness of internal and external corners: Calculated from the included normal angle of the measured surface elements of adjacent walls, reflecting the angular deviation of internal and external corners. All deviation indicators correspond to specific building components and can be classified and statistically analyzed according to component type, room area, floor, etc.

[0066] 7.2 Determination of Engineering Quality Thresholds Engineering quality thresholds are the core basis for determining the qualification of components. This invention adopts a hierarchical dynamic determination mechanism to ensure the scientific nature and engineering adaptability of the thresholds. The specific determination process consists of four steps: The first step is to extract the standard benchmark. Based on the current national and industry construction quality acceptance standards, including the "Unified Standard for Acceptance of Construction Quality of Building Engineering", "Code for Acceptance of Construction Quality of Concrete Structure Engineering", and "Standard for Acceptance of Construction Quality of Building Decoration and Renovation Engineering", the allowable deviation values ​​corresponding to various components and quality indicators are extracted as the benchmark reference for the threshold.

[0067] The second step is the classification of component importance. Based on the structural function and the degree of impact on quality, components are divided into two levels: key control components and general components. Components that have a significant impact on structural safety and core functions, such as load-bearing walls, structural columns, main beams, and openings in core areas, are classified as key control components, and thresholds stricter than those specified in the code are applied. Non-load-bearing partitions, decorative components, etc., are classified as general components, and thresholds specified in the code are applied.

[0068] The third step is construction stage adaptation. Thresholds are adjusted based on the current measurement and corresponding construction stage: the main structure stage corresponds to the structural construction accuracy requirements, the rough finishing stage corresponds to the base layer construction accuracy requirements, and the fine finishing stage corresponds to the surface finishing accuracy requirements, ensuring that the thresholds match the quality control objectives of the current process.

[0069] The fourth step is customized modification. Based on the specific accuracy requirements proposed by the construction and design units, the thresholds are customized to meet the project's unique quality control needs. Through these four steps, a two-tiered quality judgment threshold system is ultimately formed: the qualified threshold is the minimum standard for component quality; components with deviations exceeding the qualified threshold are judged as unqualified. The excellent threshold is the standard for judging component quality as excellent; components with deviations less than the excellent threshold are judged as excellent. Simultaneously, an out-of-tolerance warning threshold is set; components with deviations approaching the qualified threshold are marked with an early warning to facilitate early intervention and rectification. 7.3 Qualification Judgment and Statistical Analysis Based on the established graded quality thresholds, each tested component is assessed for conformity, and the deviation value and conformity status of each component are marked. The deviation results of all components are classified and statistically analyzed, and statistical indicators such as average deviation, maximum deviation, deviation distribution pattern, pass rate, and excellent rate are calculated for different component types.

[0070] 7.4 Spatial Visualization and Anomaly Marking The deviation results are spatially visualized, and the degree of deviation of different components is displayed in the 3D model through color mapping, forming a spatial error distribution map. Non-conforming components that exceed tolerances are highlighted, clearly indicating their spatial location, deviation value, and deviation type, facilitating on-site location and rectification verification. Components that cannot be matched due to obstruction or missed measurement are marked as low-confidence results, and supplementary measurement prompts are generated, clearly specifying the recommended location and scope for supplementary measurement, forming a closed-loop feedback.

[0071] 7.5 Quality Report Generation This system integrates all deviation statistics, conformity assessment results, spatial error distribution maps, and details of abnormal components to generate a structured quality assessment report. The report includes an overview of component deviations, regional quality statistics, details of components exceeding tolerances, and rectification suggestions. It can be exported in a standard document format, providing data support for construction acceptance, quality management, and rectification review.

[0072] The final output of this step includes: a component-level deviation data table, a spatial error distribution map, a quality compliance determination result, and a construction quality assessment report.

[0073] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for BIM collaborative mapping and quality assessment of indoor measurement robots, characterized in that: Includes the following steps: Step S1: BIM model semantic parsing and measurement task pre-generation. Perform component parsing and spatial topology analysis on the input BIM model, divide the indoor semantic space and construct a passable grid map to generate an initial set of measurement tasks. Step S2: BIM-constrained measurement site autonomous planning generates candidate sites within the traversable area. After accessibility filtering, visibility analysis, and component importance weighted scoring, the optimal set of measurement sites is obtained through comprehensive optimization of coverage benefits and path costs. Step S3: BIM topology-guided inspection path planning, constructing a site connectivity network based on indoor spatial topology, solving for the optimal site visit sequence, and generating robot motion path and site observation posture; Step S4: Multi-source data acquisition and spatial benchmark unification. The robot performs multi-source sensor data acquisition along the planned path, establishes the transformation relationship between the BIM coordinate system, the robot base coordinate system and the scanner coordinate system, and realizes the coordinate benchmark unification of multi-source data. Step S5: Multi-source data fusion and 3D measurement model construction. Based on pose information, the initial stitching of point clouds from multiple sites is completed. Fine registration is achieved through global optimization. After denoising and homogenization, an indoor 3D measurement model in a unified coordinate system is generated. Step S6: Component-level alignment and matching between BIM and measured models. Line and surface geometric primitives are extracted from the BIM model and the measured 3D model respectively. Following the matching principle of surface first and then line, a one-to-one correspondence between theoretical primitives and measured primitives is established through type constraints, geometric consistency constraints and spatial overlap constraints. Step S7: Component-level deviation calculation and construction quality assessment. Based on the matched line and surface primitives, calculate the spatial deviation index of various building components, determine the qualification of components according to the graded engineering quality threshold, and generate structured quality assessment results. The optimal set of measurement stations in step S2 is solved using the following objective function: In the formula, For the selected set of sites; To detect the coverage of the target object; This represents the total path length. The cost of turning, detouring, and repeated observations; , , These are the weighting coefficients; During the candidate site generation phase, the sampling density is adjusted according to the density of components in the spatial region. Dense sampling is used in densely componented areas, while sparse sampling is used in open areas. Visibility analysis calculates the visible range and occlusion ratio of a single site to surrounding components using ray projection. The importance weight of components is set according to the structural and functional attributes of the components, with structural load-bearing components having a higher weight than decorative components; The generated optimal set of measurement sites includes the spatial coordinates of the sites, the topological adjacency relationships between the sites, and the initial access sequence information; Step S3 specifically includes: Based on the indoor space topology connectivity graph and the passable grid map, each measurement station is used as a graph node and the passable paths between stations are used as graph edges to construct a station connectivity network model. The optimal site access sequence is solved based on the site connectivity network. The path between sites is generated by combining the indoor obstacle distribution and wall boundary constraints. The path optimization objectives include minimizing the total path length and minimizing the number of turns. The generated path is spatially feasible, and the path is checked point by point to see if it crosses an obstacle area. Impassable path segments are locally modified and replanned. Configure observation attitude parameters for each measurement station, including horizontal yaw angle and pitch angle, so that the core observation components are within the central field of view of the scanning equipment.

2. The method for indoor measurement robot BIM collaborative mapping and quality assessment according to claim 1, characterized in that: Step S1 specifically includes: The BIM model is analyzed layer by layer to separate the geometric and semantic information of each building component, and the component types are classified and labeled. Perform topological analysis on the spatial adjacency relationships of each component to construct a network of component connections and a diagram of room connectivity. Based on the geometric information and topological relationships of the components, the interior space is divided into room units, corridor units and functional area units, thus completing the semantic division of the interior space; Generate a spatial rasterized representation model within the passable area and establish an observable association mapping between raster units and building components; Various building components and their observable areas are transformed into measurement targets, and an initial set of measurement tasks is generated by combining the importance of the components with the spatial coverage requirements.

3. The indoor measurement robot BIM collaborative mapping and quality assessment method according to claim 1, characterized in that: In step S4, the spatial reference of the multi-source data is unified through the following coordinate transformation relationship: In the formula, , , The coordinates of the point in the BIM coordinate system; , , The coordinates of the point in the scanner coordinate system; The extrinsic transformation matrix from the scanner coordinate system to the robot base coordinate system is obtained through device calibration. The pose transformation matrix from the robot's base coordinate system to the BIM coordinate system is obtained through building structure feature registration and pose calculation. The sequence of measurement stations and spatial topology generated in step S2 are used to provide initial pose constraints for multi-station point cloud registration in step S5, thus narrowing the solution range for fine registration.

4. The method for indoor measurement robot BIM collaborative mapping and quality assessment according to claim 1, characterized in that: Step S5 specifically includes: Initial stitching of multi-site point cloud data is performed based on the initial pose information of each site to complete coarse alignment; A global optimization method is used to perform fine registration of the point clouds of all stations, and the pose parameters of each station are optimized by utilizing the overlapping areas of adjacent stations to eliminate pose accumulation errors. Point cloud data is filtered and denoised, and density is equalized to remove discrete outliers and to homogenize areas with density differences. Structural consistency optimization is performed on planar components such as walls and floors, and linear features such as edges and corners, to weaken the impact of point cloud noise on geometric features while retaining construction deviations. Data from all sites is integrated to generate an indoor 3D measured point cloud model in a unified coordinate system.

5. The indoor measurement robot BIM collaborative mapping and quality assessment method according to claim 1, characterized in that: The unified expression form of the line-surface geometric primitives in step S6 is as follows: In the formula, It is a set of BIM theoretical primitives. Let m be the theoretical surface primitive. Let k be the basic element of the theoretical line; For the set of measured primitives, For the nth measured surface primitive, Let t be the basic element of the measured line; The matching degree of surface primitives is determined using the following comprehensive scoring function: In the formula, This represents the vertical offset from the measured surface to the theoretical surface. The angle between the normals of the two surfaces; The spatial overlap rate of the two sides; The fitting residual of the measured plane; to These are the weighting coefficients for each indicator; When the comprehensive score meets the preset threshold and the spatial overlap rate meets the requirements, the correspondence between the theoretical surface primitive and the measured surface primitive is established.

6. The indoor measurement robot BIM collaborative mapping and quality assessment method according to claim 5, characterized in that: In step S6, the matching of line elements is carried out based on the adjacency relationship of already matched surface elements. The directional deviation of the line elements is calculated using the following formula: In the formula, Let the direction vector be the theoretical line element. The direction vector of the measured line element; The distance between line elements is calculated using the following formula: In the formula, These are sampling points on the measured line element. The total number of sampling points. The spatial distance function from a point to a theoretical line element; By combining directional deviation, line distance, endpoint overlap, and the relationship between the corresponding surface elements, the matching consistency between theoretical line elements and measured line elements is determined.

7. The indoor measurement robot BIM collaborative mapping and quality assessment method according to claim 1, characterized in that: The process of determining the engineering quality threshold in step S7 is as follows: The first step is to extract the corresponding deviation limits for various components such as walls, floors, doors and windows, beams and columns, based on the allowable deviation values ​​of the components. The second step is to classify the components according to their structural importance. Load-bearing walls, structural columns, and main beams, which are the main control components, are set as the first level, and the thresholds are stricter than the standard benchmarks. Non-load-bearing partitions and decorative components are set as the second level, and the standard benchmark thresholds are used. The third step is to make adaptation adjustments based on the current construction stage. The precision requirements of the corresponding construction procedures are matched for the main structure stage, rough decoration stage, and fine decoration stage, forming staged thresholds. The fourth step involves customized adjustments based on the project's specific accuracy requirements, ultimately forming a two-level quality judgment threshold system that includes qualified and excellent thresholds. At the same time, an out-of-tolerance warning threshold is set to mark components that are approaching the limit.

8. The method for indoor measurement robot BIM collaborative mapping and quality assessment according to claim 1, characterized in that: The indicators for calculating component-level deviations in step S7 include wall position offset, wall flatness, floor elevation deviation, bay depth, door and window opening dimensions, and squareness of internal and external corners; After the deviation calculation is completed, the deviation results of various components are statistically analyzed to generate a spatial error distribution map, and the out-of-tolerance components are spatially located and labeled. For components that are missed or obstructed, resulting in mismatch, mark them as low-confidence results and generate supplementary measurement prompts to guide the location and range of supplementary measurements.

9. An indoor measurement robot BIM collaborative mapping and quality assessment system according to claim 1, characterized in that: It includes modules for BIM semantic parsing and task planning, autonomous site planning, path planning, multi-source data acquisition and benchmark unification, 3D reconstruction, component-level matching analysis, and quality assessment. The output of the BIM semantic parsing and task planning module is connected to the input of the autonomous site planning module. It is used to parse the component information and spatial topology of the BIM model and generate a passable grid map and initial measurement tasks. The output of the autonomous site planning module is connected to the input of the path planning module. It is used to generate the optimal set of measurement sites under BIM constraints and output the spatial distribution of sites and the mapping relationship between components. The output of the path planning module is connected to the input of the multi-source data acquisition and benchmark unification module, which is used to construct the site connectivity network and solve the optimal access path, and generate robot motion sequence and site observation posture parameters. The output of the multi-source data acquisition and benchmark unification module is connected to the input of the three-dimensional reconstruction module. It is used to control the robot to perform multi-source data acquisition, establish multi-coordinate system transformation relationships, and output the original measurement dataset under a unified benchmark. The output of the 3D reconstruction module is connected to the input of the component-level matching analysis module, and is used to register, denoise and fuse point cloud data to generate an indoor 3D measured model. The output of the component-level matching analysis module is connected to the input of the quality assessment module. It is used to extract the line and surface geometric primitives of the BIM and the measured model and complete the component-level matching and alignment, and output the component-level deviation data. The quality assessment module is used to determine the compliance of construction quality based on graded quality thresholds and output structured quality assessment results and report documents.

10. The indoor measurement robot BIM collaborative mapping and quality assessment system according to claim 9, characterized in that: The autonomous site planning module includes a candidate site generation unit, an accessibility filtering unit, a visibility analysis unit, a weight modeling unit, and an optimization filtering unit. The candidate site generation unit is used to generate a set of candidate sites for differentiated sampling based on the density of regional components; The accessibility filtering unit is used to eliminate candidate sites in impassable areas; The visibility analysis unit is used to calculate the observation coverage of each candidate site on surrounding components; The weighted modeling unit is used to assign a comprehensive observation score to candidate sites based on the importance of components; The optimization filtering unit is used to select the optimal site layout through a comprehensive optimization model that considers coverage benefits and path costs.

Images