Room position dimension deviation automatic detection method based on handheld point cloud acquisition

Abstract

The invention discloses a room position dimension deviation automatic detection method based on handheld point cloud acquisition, relates to the technical field of three-dimensional space information processing, and aims to solve the problems of low data acquisition efficiency, disjunction in processing and lack of real-time feedback in the prior art. According to the method, on the basis of a handheld scanning terminal, a mobile application program and a cloud server, through BIM-driven augmented reality scanning path planning, field data real-time quality monitoring and end-cloud collaborative point cloud intelligent processing, automatic recognition and evaluation of room position size deviation are achieved, and intelligent judgment and visual delivery of results are carried out. By adopting the technical scheme, the whole-process automation and intelligentization of the detection process can be realized, the operation efficiency is remarkably improved, and the data quality and the reliability of an analysis result are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional space information processing, and in particular to an automatic detection method for room position size deviation based on handheld point cloud acquisition. Background Art

[0002] In the full life cycle management of construction and subsequent operation and maintenance of building projects, accurate verification of the consistency between the built entity and the design blueprint is a core link in ensuring project quality, controlling construction costs, and avoiding safety risks. Specifically, the measurement of a series of parameters such as the clear height, span, and depth of the room in the interior space, as well as the precise positioning, geometric dimensions, surface flatness, and spatial verticality of key components such as walls, columns, beams, doors and windows, constitute the basic content of building quality inspection. For a long time, technicians in this field have mainly relied on traditional contact or single-point measurement tools, such as tape measures, laser rangefinders, rulers, and total stations widely used in the field of surveying and mapping. These technical means provided a basic solution for the measurement and determination of engineering quality in their specific historical period. Their core working principle is to approximate the geometric form of components through discrete point measurements. However, the inherent limitations of this method are becoming increasingly prominent. Not only is its operating efficiency low and requires high labor intensity from operators, but the more critical problem lies in the point-like sparseness of its measurement results, which makes it difficult to fully and objectively reflect the continuous surface geometric characteristics of components such as walls and floor slabs. Moreover, the measurement process and results are extremely susceptible to interference from human operational errors.

[0003] To overcome the data incompleteness limitations of traditional surveying methods, ground-based 3D laser scanning technology has emerged. By emitting laser beams from fixed stations and receiving reflected signals, this technology can quickly acquire millions of 3D coordinate points across a target scene, generating high-density point cloud data. This technological paradigm significantly improves the comprehensiveness of data acquisition, replacing points with surfaces, providing an unprecedentedly refined digital representation of the geometric form of building components. This in turn provides a reliable data foundation for analyzing surface features such as flatness and verticality. However, the inherent operational nature of ground-based scanning technology presents new challenges when dealing with the dynamic and complex environments of construction sites. The equipment is typically bulky, making transfer and deployment cumbersome. Furthermore, ensuring data integrity often requires scanning from multiple stations within the same space, followed by complex point cloud data registration. This significantly prolongs both field and field data processing cycles, making it difficult to meet the rapid response and immediate feedback required on-site.

[0004] In recent years, with the rapid development and maturity of Simultaneous Localization and Mapping (SLAM) technology, handheld 3D laser scanning devices that integrate this technology have received widespread attention in the industry due to their unparalleled portability and flexibility. Operators can quickly complete the collection of point cloud data of the environment by freely moving the handheld device in space. In theory, this perfectly meets the needs of fast and mobile inspections on construction sites and effectively solves the problem of low efficiency of ground-based station scanners. However, in-depth application of the technology has revealed a more hidden and profound contradiction: the current application of handheld scanning technology has, to a large extent, caused a serious disconnect between the efficiency of front-end data collection and the efficiency of back-end data processing and application. Specifically, existing application solutions generally position handheld devices as a simple "data collection tool", and the raw point cloud data they generate is essentially a set of geometric coordinates that lacks semantic information and has a low degree of structure. To transform this raw data into meaningful quality assurance conclusions, subsequent data processing, including point cloud noise reduction, segmentation, alignment with the underlying Building Information Model (BIM) or CAD drawings, and subsequent complex operations such as component identification, dimension extraction, and deviation comparison, still requires manual or semi-manual processing by specialized technicians on high-performance personal computers (PCs) using specialized software. This fragmented workflow results in the time saved by handheld devices in front-end acquisition being multiplied by back-end "data back-office" processing. More importantly, this non-integrated process lacks an effective real-time feedback loop. Upon completing a scanning task, field personnel have no immediate information on whether the data quality meets the requirements for subsequent analysis, such as whether there are blind spots or insufficient point cloud density in key feature areas. These flaws are often only discovered when the data is imported into a PC for processing, leading to costly re-tests and significantly undermining the claimed efficiency of handheld scanning devices in practice. The reason is that existing technologies fail to deeply couple the portability advantages of handheld devices with the intelligence and automation of data processing, and fail to use the rich semantic information contained in BIM models as prior knowledge to guide and drive the processing and analysis of point cloud data, resulting in an insurmountable gap between the two key stages of "data collection" and "information extraction" in the entire inspection process.

[0005] Therefore, how to break through the technical bottleneck of "fast front end and slow back end" in current handheld scanning applications, deeply integrate the convenience of handheld scanning with the inherent intelligence of building information models, and build a full-process closed-loop method from on-site data collection, real-time quality monitoring, cloud-based intelligent processing to automated deviation analysis and report generation, so as to overcome the separation between data collection and data application in existing technologies and realize the immediacy, accuracy and intelligence of building quality inspection, has become a key challenge and technical problem that needs to be solved urgently by technical personnel in this field. Summary of the Invention

[0006] The present invention aims to overcome existing shortcomings, such as the severe disconnect between the data acquisition efficiency of handheld scanning devices and the efficiency of back-end data processing and application, a technological gap between data acquisition and information extraction, and a lack of effective real-time feedback and closed-loop management. To this end, the present invention provides a method for automatically detecting room position and dimensional deviations based on handheld point cloud acquisition. This method, through the construction of a fully automated system integrating augmented reality guidance, real-time quality monitoring, end-to-end collaborative intelligent processing, and deep-driven building information modeling (BIM), aims to achieve instant, accurate, and intelligent detection and assessment of position and dimensional deviations in building interior spaces and components.

[0007] To achieve the above-mentioned purpose of the invention, the present invention provides a method for automatic detection of room position size deviation based on handheld point cloud acquisition, and the method relies on a system architecture consisting of a handheld scanning terminal, a supporting mobile application (App) and a cloud server. The handheld scanning terminal integrates a three-dimensional lidar sensor, an inertial measurement unit (IMU) and a visible light camera. The central wavelength of the three-dimensional lidar sensor is 905nm, the ranging range is not less than 60 meters, the ranging accuracy is better than ±10mm, and the point cloud acquisition rate is not less than 320,000 points / second. The inertial measurement unit is a six-axis MEMS sensor, which includes a three-axis gyroscope and a three-axis accelerometer. The cloud server is equipped with a high-performance computing unit and a distributed database. The method specifically includes the following steps: First, the task loading and augmented reality scanning path planning steps are performed. The operator launches the mobile application on the handheld terminal, connects to the cloud server via the Secure Sockets Layer (SSL) protocol, and loads the building information model of the target project. The data format of the building information model is the Industry Foundation Class (IFC) standard format. The system automatically performs BIM model parsing and automatic division of inspection task units. Specifically, the system's embedded BIM parsing engine traverses the IFC data structure, identifies and extracts entity information of IfcBuildingStorey (floor), IfcGrid (axis grid), IfcSpace (space), and custom property sets (PropertySet) related to the division of construction flow sections. Based on the above information, the system automatically decomposes the entire project model into a series of independent inspection task units with unique identifiers (UUIDs). The data structure of each inspection task unit includes: the task unit UUID, the floor and area identifier, the globally unique identifier (GUID) of the associated IfcSpace object, the 3D geometric bounding box (Bounding Box) of the space, and a list of GUIDs of all BIM components in the space (such as walls, slabs, columns, doors, and windows).

[0008] After the inspection task units are divided, the system automatically plans the optimal scanning path for the selected task unit. The goal of this path planning algorithm is to generate the shortest trajectory covering all critical surfaces within the space while meeting a preset point cloud density (unit: points / square meter). Specifically, the system first voxelizes the 3D geometric space of the task unit with a voxel resolution of 5 cm. Based on the BIM model, voxels occupied by physical components such as walls, floors, and columns within the space are marked as obstacles. Subsequently, a modified Rapidly Expanding Random Tree Star (RRT*) algorithm is used to search for paths in non-obstacle areas. The algorithm's cost function considers not only path length but also the visibility of path points on BIM component surfaces and the expected scan coverage. The algorithm ultimately outputs a 3D spatial path consisting of a series of six-degree-of-freedom (6-DoF) poses. The mobile application projects this 3D path in real time onto the handheld terminal's camera preview screen, rendering a continuous 3D arrow model as augmented reality (AR) visual guidance to instruct the operator to move along this optimized path.

[0009] Next, on-site data collection and real-time quality monitoring steps are performed. The operator, holding the scanning terminal, moves within the room being measured, following the AR path guidance. The handheld terminal's built-in simultaneous localization and mapping (SLAM) system tightly couples the point cloud data from the 3D LiDAR sensor with the readings from the inertial measurement unit to calculate the device's six-degree-of-freedom pose in the world coordinate system in real time and simultaneously generate raw point cloud data. During the scanning process, the mobile application, in a separate computational thread, performs real-time registration of the point cloud data with the BIM model and quality analysis. Specifically, the system uses the device pose output by the SLAM system as the initial transformation matrix, downsampling a portion of the point cloud data generated each second to a 5cm resolution. This data is then lightweight-registered with a preloaded, similarly voxelized BIM model using a single iteration of the generalized iterative closest point (G-ICP) algorithm. After registration, the system evaluates the coverage and density of the collected point cloud on the BIM component surface in real time. The implementation method is as follows: the triangular mesh model (Mesh) of each BIM component is parameterized and expanded into a two-dimensional UV texture coordinate space. For each newly collected and registered point cloud point, its nearest projection point on the BIM component surface is calculated, and the UV coordinates corresponding to the projection point are recorded in a two-dimensional gridded cumulative amount matrix. The mobile application renders the BIM model surface in real time into three colors according to the count value in the cumulative amount matrix: when the grid count value corresponding to a certain area is greater than the preset density threshold ρ target When the count value is between 0 and ρ, it is rendered in green; when the count value is between 0 and ρ target When the count value is between 0 and 1, it is rendered in yellow; when the count value is 0, it is rendered in gray. Based on this visual feedback, the operator performs additional scans on the yellow area until most of the area turns green, thus ensuring the integrity and uniformity of the original data collection.

[0010] Next, the end-cloud collaborative point cloud preprocessing step is performed. After a scan is completed for an inspection task unit, the operator confirms submission on the mobile application. The handheld terminal locally performs feature-sensitive adaptive downsampling of the point cloud data. This downsampling process aims to minimize the data volume without losing key geometric features. The specific algorithm is curvature-sensitive adaptive voxel grid downsampling. First, an octree index is constructed for the collected complete raw point cloud data. Second, the covariance matrix of the three-dimensional coordinates of the point cloud subset contained in each leaf node in the octree is calculated. Third, the covariance matrix is ​​subjected to eigenvalue decomposition to obtain three eigenvalues ​​λ1 ≥ λ2 ≥ λ3. Then, a normalized curvature metric C = λ3 / (λ1 + λ2 + λ3) is calculated based on the eigenvalues. This metric C approaches 0 when the point cloud distribution is linear or planar, and is higher when the distribution is corner-shaped or scattered. Finally, the retention strategy for the point cloud within the leaf node is determined based on metric C: retain all points within the leaf node or only its centroid. The probability of this decision is positively correlated with the C value, ensuring that point clouds in areas of high curvature, such as corners, opening edges, and structural ridges, are retained with a high probability, while point clouds on flat surfaces are significantly downsampled. This method compresses point cloud data to 15% to 20% of its original size.

[0011] The downsampled, feature-enhanced point cloud data, along with the UUID of the detection task unit to which it belongs, is packaged and uploaded to the cloud server via an HTTPS-based API. Upon receiving the data, the cloud server first performs deep noise reduction. This process utilizes a pre-trained point cloud noise classifier model based on a graph convolutional network. This model can identify and remove complex discrete noise points caused by material reflectivity, airborne particles, and other factors, achieving superior results compared to traditional statistical outlier removal (SOR) or radius filtering algorithms.

[0012] After noise reduction is complete, the server performs the core BIM-driven point cloud individualization and semantic segmentation. This step leverages the BIM model as strong prior knowledge to guide the point cloud segmentation process. Specifically, based on the received inspection task unit UUID, the server queries the associated BIM database for a list of GUIDs for all BIM components within that task unit, along with their precise geometric definitions and spatial locations. For each BIM component in the list, the server extracts the wall's boundary polygon and extrusion vector from the BIM definition to generate its precise digital geometric surface. Then, within the received point cloud data, a search space is defined with the BIM wall geometry as the center and a 30cm extension. Within this restricted search space, a Directed Random Sampling Consensus (Directed RANSAC) plane fitting algorithm is executed. The algorithm's sampling process is constrained within the search space, ensuring robustness and efficiency of the fitting process. All points within the fitted plane constitute the point cloud subset belonging to that wall. The server separates this point cloud subset from the main point cloud and assigns it a semantic label, which is directly derived from the BIM component GUID. This process is repeated for all BIM components within the task unit until the entire point cloud data is accurately segmented into individual component point clouds with semantic labels that correspond one-to-one to each BIM component.

[0013] Subsequently, the automated judgment step of the multi-dimensional data quality compliance is executed. After completing the individual segmentation, the system performs a quantitative quality assessment on each component point cloud subset to determine whether it meets the accuracy requirements of the subsequent deviation analysis. The evaluation system includes three core indicators: coverage completeness S cov , data validity S den and feature completeness S feat Coverage completeness S cov The calculation formula is: the area covered by the projection of the segmented component point cloud onto the corresponding BIM component geometric surface, divided by the total surface area of ​​the BIM component. den The calculation formula is: the average point density of the component point cloud divided by the preset target density threshold ρ target , the upper limit of the result is 1.0. Feature completeness S feat The calculation method is as follows: First, on the geometric model of the BIM component, use the 3D Harris corner detection algorithm to extract all its geometric corner points and sharp edge segments as the key feature point set; then, for each key feature point, check whether there are at least 10 points from the corresponding component point cloud subset within a spherical neighborhood with a radius of 5 cm in 3D space. Feature completeness S feat That is, the percentage of the key feature points that are successfully verified to the total number of key feature points. Finally, a comprehensive quality score Q scoreCalculated by weighted average: Q score = 0.5 * S cov + 0.3 * S den + 0.2 * S feat .

[0014] The system will calculate the Q score Compare with a preset qualification threshold. score If Q is greater than or equal to the threshold, the data quality is determined to be qualified and the process continues. score If the data quality falls below this threshold, the system is deemed unqualified. At this point, the system generates a command and sends a re-survey request to the on-site operator via a push notification on the mobile app. This request clearly specifies the GUID of the component whose data quality failed and the reason for failure (for example, "coverage completeness is only 70%" or "feature point missing in the northeast corner"), providing precise guidance for re-surveying.

[0015] After the data quality is judged to be qualified, the system starts the overall spatial position deviation detection and component-level geometric dimension deviation detection in parallel.

[0016] For overall spatial position deviation detection, the system treats all successfully segmented and labeled component point clouds as a rigid whole and performs global optimal registration with the corresponding component geometry set in the BIM model. This registration uses the more robust Levenberg-Marquardt iterative closest point algorithm (LM-ICP) to obtain an optimal 4×4 homogeneous transformation matrix T that transforms the entire point cloud from the current measured position to the BIM design position. deviation The matrix T deviation This matrix represents the overall spatial deviation of the room in its built state relative to the design blueprint. The system decomposes this matrix into translation vectors (ΔX, ΔY, ΔZ) and rotation Euler angles (ΔRx, ΔRy, ΔRz), which quantify the deviation of the room's overall position and posture.

[0017] For component-level geometric dimension deviation detection, the system executes specific analysis algorithms for different types of components. Taking door and window openings as an example, the system first determines the principal plane of its individualized point cloud subset through principal component analysis (PCA) and projects the point cloud onto this plane. In the two-dimensional projection, the Hough Transform algorithm is used to detect the four straight lines that form the opening boundary. By calculating the intersection of these four straight lines, the coordinates of the four corner points of the opening are accurately obtained. The measured width and height of the opening are then calculated. The system then queries the Width and Height parameters of the opening's IfcOpeningElement object from the BIM database as the design values. Subtracting the two values ​​yields the dimensional deviation.

[0018] Taking wall components as an example, the system performs two geometric tolerance analyses on its individualized point cloud subsets. The first is a flatness check: the system uses the least squares method to perform an optimal plane fit on the wall point cloud, obtaining a reference plane equation Ax + By + Cz + D = 0. The orthogonal distance from each point in the point cloud subset to this reference plane is then calculated. The maximum absolute value of all distances is the wall's flatness deviation. The second is a perpendicularity check: the system extracts the unit normal vector n = (A, B, C) of the fitted reference plane and calculates the angle θ = arccos(n·v) between this normal vector and the Z-axis unit vector v = (0, 0, 1), representing the absolute vertical direction. This angle θ is the wall's perpendicularity deviation.

[0019] Finally, the logical composite judgment and multi-dimensional data delivery steps of the results are executed. After the deviation values ​​of all inspection items are calculated, the system starts a configurable specification rule engine. This engine is pre-loaded with a structured database that stores the national construction engineering construction quality acceptance specifications or project-defined tolerance limits. Each record contains the component type, the inspection item name, and the upper and lower limits of the allowed tolerance. For example, a record is (component type: interior wall, inspection item: flatness, tolerance limit: 4mm). The system automatically compares each calculated deviation value with the corresponding entry in the rule engine.

[0020] If the deviation values ​​for all test items are within their respective tolerance limits, the product is deemed "passed." If any one or more items are out of tolerance, the product is deemed "failed." The system marks and records any failed items. The process ultimately summarizes all pass and fail results.

[0021] The system automatically generates a structured inspection report and delivers it via the mobile app and web portal. This report presents the results in a multi-dimensional, visual manner. Within the mobile app on a handheld terminal, the system overlays the inspection results directly onto the 3D BIM model, using color coding (e.g., green for acceptable, red for non-tolerance) and numerical labels to visually display the deviation status of each component. The system also generates a detailed inspection report in standard PDF or Excel format with a single click. This report includes a project overview, inspection scope, a summary table of deviation data for all inspection items, a detailed list of non-tolerance items and their specific deviation values, and a visual analysis chart of the deviations for key non-conforming components (e.g., walls with non-tolerance flatness). The flatness deviation visualization chart is a heat map in which each point in the wall point cloud is assigned a color value based on its distance from the fitted reference plane. This color spectrum, from blue (concave) to green (flat), to red (convex), clearly illustrates the distribution and severity of uneven areas on the wall. This concludes the inspection process.

[0022] Compared with the prior art, the present invention has the following beneficial effects: (1) The entire inspection process is automated and intelligent, significantly improving operational efficiency. By introducing BIM-driven automated task planning, AR path guidance, intelligent processing of end-cloud collaboration, and rule-based automatic evaluation, the present invention integrates the traditionally separate data collection, processing, and analysis links. This allows the entire process from on-site scanning to generating detailed reports to be completed in a very short time, achieving “instant measurement and analysis, instant reporting”, thereby shortening the traditional inspection cycle from hours or even days to minutes.

[0023] (2) Ensure the quality and reliability of data collection and analysis results. The real-time quality monitoring mechanism proposed in this paper ensures the integrity and validity of the original data from the source by visually feedback the coverage and density of the point cloud during the scanning process, thus avoiding ineffective rework caused by data quality issues. In addition, the BIM-driven semantic segmentation method, which uses the BIM model as strong prior knowledge, greatly improves the accuracy and robustness of building component recognition in the point cloud, laying a solid data foundation for subsequent accurate deviation calculation.

[0024] (3) It deepens the application value of Building Information Modeling (BIM) in engineering quality inspection. This invention no longer regards the BIM model as a simple geometric reference background, but instead elevates it to an intelligent engine that drives and guides the entire inspection process. From the automatic decomposition of tasks and the planning of scanning paths, to the semantic segmentation of point cloud data, and then to the automatic extraction and comparison of design values, the inherent logic and rich information of BIM run through the entire process, achieving a deep integration of data collection, processing and design intent.

[0025] (4) An efficient closed-loop quality management system has been established. The instantaneous analysis results and multi-dimensional visual reports provided by the present invention enable construction management personnel to accurately identify and locate quality problems at the first possible moment and on the spot. Real-time warnings for non-conforming items and precise retesting guidance, combined with detailed deviation analysis reports, provide a timely and reliable basis for subsequent corrective decision-making, forming a rapid, closed-loop quality control process of "measurement-analysis-feedback-correction", effectively avoiding cost increases and construction delays caused by delayed discovery of quality problems.

[0026] (5) Lowering the threshold for using advanced detection technology. Due to the high degree of automation of the entire process, the present invention significantly reduces the professional skills required of operators. Operators only need to complete simple scanning actions according to AR guidance, and all complex data processing and analysis work is automatically completed by the background system, making high-precision three-dimensional laser scanning detection technology popularized and applied in a wider range of engineering scenarios.

[0027] In order to make the above-mentioned objects, features and advantages of the present invention more clearly understood, embodiments of the present invention are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0029] Figure 1 It is a schematic diagram of the architecture of the system of the present invention; Figure 2 is an overall flow chart of the method of the present invention; Figure 3 is a flow chart of task loading and augmented reality scanning path planning according to the present invention; Figure 4 is a flow chart of on-site data collection and real-time quality monitoring according to the present invention; Figure 5 This is a flow chart of the point cloud preprocessing of the end-cloud collaboration described in the present invention; Figure 6 It is a flow chart of the automated judgment of the quality compliance of multi-dimensional data according to the present invention; Figure 7 Flowchart of deviation detection and result delivery according to the present invention. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0031] Reference Figure 1 , which demonstrates the system architecture underlying the present invention. The system consists of three core components: a handheld scanning terminal, a supporting mobile application running on the terminal, and a cloud server. The handheld scanning terminal, such as the GeoScanner Pro S, integrates a series of highly integrated sensors and processing units. Specifically, its core sensing component is a 3D lidar sensor with an eye-safe central wavelength of 905nm, which is Class 1 laser safety. It has an effective ranging range of up to 60 meters and an accuracy of better than ±10mm at a distance of 10 meters. It also captures a point cloud at a rate of up to 320,000 points per second, with a 360-degree horizontal field of view and a 270-degree vertical field of view. To achieve accurate attitude estimation, the terminal also integrates a high-performance six-axis inertial measurement unit (IMU), specifically a Bosch BMI160 microelectromechanical system (MEMS) sensor. This unit synchronously outputs angular velocity data from a three-axis gyroscope and linear acceleration data from a three-axis accelerometer at a frequency of 200Hz. The device also features a 12-megapixel visible light camera with a 120-degree wide-angle field of view to capture texture information from the surrounding environment and support augmented reality. The device's local processing is provided by a high-performance ARM-based system-on-chip (SoC), such as the Qualcomm Snapdragon 8 Gen 2 processor, and is supplemented by 12GB of RAM to ensure smooth operation of on-device algorithms.

[0032] The accompanying mobile app, installed and running on the handheld scanning terminal, serves as the primary interface for human-computer interaction and the dispatch center for on-device processing tasks. The app maintains stable and reliable data communication with the cloud server via a Secure Sockets Layer (SSL)-encrypted channel.

[0033] The cloud server is deployed in a high-performance data center. Its hardware configuration includes a graphics processing unit (GPU) cluster for massively parallel computing and a distributed database system for persistent storage of massive amounts of data. For example, a database based on PostgreSQL, extended with a PostGIS plug-in for efficient processing of geospatial data, handles the computationally intensive data processing and analysis tasks of this invention.

[0034] Combine Figures 2 to 7As shown, the specific implementation steps of the method for automatic detection of room position size deviation based on handheld point cloud acquisition provided by the present invention will be described in detail below.

[0035] like Figure 3 As shown, the first step of the process is task loading and augmented reality scanning path planning. The operator first starts the supporting mobile application on the handheld scanning terminal. The application initiates a connection request to the cloud server through the aforementioned SSL encrypted channel and performs user identity authentication. After the verification is passed, the operator selects and loads the building information model (BIM) of the target project to be detected from the project list on the server. The BIM model processed by the present invention adopts the Industry Foundation Class (IFC) standard format, specifically the IFC4 version. After the model is loaded, the system automatically starts an embedded BIM parsing engine, which can be built based on an open source IFC processing library (such as IfcOpenShell). The engine is responsible for traversing the hierarchical structure of the entire IFC data file, accurately identifying and extracting information on key entities such as IfcBuildingStorey (floor), IfcGrid (axis grid), IfcSpace (space), and custom property sets (PropertySet) that record construction flow segment division information.

[0036] Based on this parsed structured information, the system automatically divides inspection task units. This process logically and automatically decomposes a large building project model into a series of independent, manageable inspection task units. Each inspection task unit is created as a separate record in the database and assigned a universally unique identifier (UUID). The data structure of this record is designed to include the following fields: the task unit's own UUID, the floor and area identifier (e.g., "Building A - 4th Floor - East Zone"), the globally unique identifier (GUID) of the core IfcSpace object to which it is associated, the specific coordinate range of the 3D geometric bounding box (Bounding Box) of the space in the project coordinate system, and a list of the GUIDs of all BIM components within the space (e.g., IfcWall walls, IfcSlab floors, IfcColumn columns, IfcDoor doors, IfcWindow windows, etc.). This division breaks down complex inspection tasks into atomic tasks based on individual rooms or spaces, greatly simplifying subsequent data management and processing.

[0037] After a detection task unit is selected, for example, the operator clicks on the task unit representing "Room 401" on the application interface, the system automatically plans the optimal scanning path for the unit. The core goal of the path planning algorithm is to generate a walking trajectory with the shortest moving distance for the operator, while ensuring that the average density of the point cloud on the surface of all key components is not lower than the preset threshold (for example, 5000 points / square meter). Specifically, the planning algorithm first voxelizes the three-dimensional geometric space of the selected task unit, and the size resolution of the voxel is set to 5 cm. Subsequently, based on the BIM model, the system marks the voxels occupied by all solid components (walls, floor slabs, columns, etc.) in the space as inaccessible obstacle pixels. On this basis, the system uses an improved rapidly expanding random tree star algorithm (RRT*) to search for paths in the free space composed of all non-obstacle pixels. The improvement here is reflected in the design of its cost function. The traditional RRT* algorithm mainly takes the path length as the optimization target, while the cost function in the present invention c(x) is defined as a weighted sum: c(x)=αL(x)+β(1-V s (x))+γ(1-C r (x)) in, L(x) is the path length, V s (x) is the visibility score of each point on the path to the key surface of the BIM component, C r (x) It is the coverage of the BIM surface by the point cloud generated based on the expected path points. α,β,γ The introduction of the following parameters (for example, they can be set to 0.4, 0.3, and 0.3, respectively) allows the algorithm, while seeking the shortest path, to proactively gravitate toward locations that can better "see" and scan all building surfaces. The algorithm's ultimate output is a three-dimensional spatial path consisting of a series of sequentially arranged six-degree-of-freedom (6-DoF) pose points. After receiving this path data, the mobile application uses a graphics rendering engine to project it in real time onto the handheld terminal's camera preview screen. This is manifested as a continuous, eye-catching three-dimensional arrow model, creating an intuitive augmented reality (AR) visual guide that instructs the operator to move and scan along this optimized path.

[0038] like Figure 4As shown, the second step of the process is on-site data collection and real-time quality monitoring. The operator holds the scanning terminal and moves smoothly in the room to be tested according to the guidance of the AR arrow on the screen. During this process, the terminal's built-in simultaneous positioning and mapping (SLAM) system starts working. The present invention adopts a SLAM algorithm based on a tightly coupled lidar-inertial navigation odometer (LIO-SAM). The algorithm tightly couples and fuses the raw point cloud data generated by the three-dimensional lidar sensor at a rate of 320,000 points per second with the angular velocity and acceleration data output by the inertial measurement unit at a frequency of 200Hz in a factor graph optimization framework. This method can solve the six-degree-of-freedom pose (three-dimensional position x, y, z and three-dimensional attitude) of the scanning terminal in the world coordinate system (usually with the scanning starting point as the origin) in real time and with high precision. roll, pitch, yaw ), and simultaneously build the original point cloud map of the environment.

[0039] At the same time, to ensure data quality at the source, the mobile application performs real-time registration and quality analysis of the point cloud data and the BIM model in parallel in a separate computing thread. This process is designed to be extremely lightweight to ensure that it does not affect the real-time performance of the SLAM system. Specifically, the system uses the device pose just output by the SLAM system as a high-precision initial transformation matrix to downsample the newly generated portion of the point cloud data (approximately 320,000 points) every second to a voxel grid. The downsampling resolution is set to 5 cm to reduce the amount of computation. Subsequently, by executing an iterative generalized iterative closest point (G-ICP) algorithm, this batch of downsampled point clouds is quickly and lightweightly aligned with the pre-loaded and similarly voxelized BIM model.

[0040] After the registration is completed, the system immediately conducts a real-time evaluation of the coverage and density of the collected point cloud on the surface of the BIM component. Its implementation method is innovative: the system first parametrically unfolds the triangular mesh model (Mesh) of each BIM component in the background and maps it to a two-dimensional UV texture coordinate space. Then, the system maintains a two-dimensional rasterized cumulant matrix corresponding to the UV space. For each newly collected point cloud point that is registered to the component surface, the system calculates its nearest projection point on the component surface, and maps the UV coordinates corresponding to the projection point to the corresponding grid of the cumulant matrix, so that the count value of the grid is increased by one. The mobile application dynamically renders the BIM model surface into three different colors based on the real-time count value in this cumulant matrix and presents it on the screen: when the grid count value corresponding to a certain area is greater than the preset density threshold ρ target (e.g., an equivalent value of 5000 points per square meter), the area is rendered green on the screen, indicating that the data collection is sufficient; when the count value is between 0 and ρ targetWhen the count is between 0 and 0, the image is rendered yellow, indicating insufficient data. When the count is 0, the image is rendered gray, indicating that the area has not been scanned. Based on this intuitive visual feedback, operators can conduct additional scans in the yellow areas until the majority of the BIM model on the screen appears green. This mechanism ensures the integrity and uniformity of the original data collection, fundamentally avoiding later rework caused by data quality defects.

[0041] like Figure 5 As shown in Figure 3, the third step in the process involves device-cloud collaborative point cloud preprocessing. Once the scanning of an inspection task unit (e.g., "Room 401") is deemed complete based on the aforementioned quality control mechanism, the operator clicks the "Complete and Submit" button on the mobile app. The handheld device then performs a feature-sensitive adaptive downsampling process on the collected complete raw point cloud data (which may contain tens of millions of points). This step aims to minimize the data volume while preserving key geometric features (such as corners, opening edges, and structural edges) for rapid upload. This downsampling algorithm is a curvature-based adaptive voxel grid downsampling method. First, the complete raw point cloud data is constructed into an octree data structure for spatial indexing. Second, for each leaf node in the octree, the covariance matrix of the 3D coordinates of the point cloud subset is calculated. Finally, the covariance matrix is ​​subjected to eigenvalue decomposition, yielding three eigenvalues: λ1 ≥ λ2 ≥ λ3. Then, a normalized curvature metric value C = λ3 / (λ1+λ2+λ3) is calculated based on these three eigenvalues. Theoretically, when the point cloud is linearly distributed (such as an edge line) or planarly distributed within a leaf node, λ3 will be very small and the C value will approach 0; when the point cloud is distributed as three-dimensional corner points or randomly scattered points, the three eigenvalues ​​are closer and the C value is higher. Finally, the system determines the retention strategy of the point cloud within the leaf node based on this curvature metric value C, which is specifically a probability function positively correlated with the C value, for example, P(retain) = k * C + P base , where k is the scale factor, P base This strategy ensures that point clouds in areas of high curvature are fully preserved with a high probability, while flat surfaces are significantly downsampled, typically based on their centroids. This method typically compresses point cloud data to 15% to 20% of its original size while preserving the vast majority of geometric detail.

[0042] The downsampled, feature-enhanced point cloud data, along with the UUID of the detection task unit to which it belongs, is packaged and securely uploaded to the cloud server via a RESTful API based on the HTTPS protocol. Upon receiving the data packet, the cloud server first feeds it into a deep noise reduction module. This module utilizes a pre-trained graph convolutional network (GCN)-based point cloud noise classifier model, such as one based on the DGCNN architecture. Trained on a large dataset of synthetic point clouds containing simulated noise, this model accurately learns and identifies patterns of complex discrete noise points caused by surface material reflectivity, airborne particles, and sensor electrical noise, effectively removing these noise points. Its noise reduction performance significantly outperforms traditional statistical outlier removal (SOR) or radius-based filtering algorithms, and is particularly robust when processing point clouds with non-uniform density.

[0043] After denoising, the server performs one of the most critical steps in the entire process: BIM-driven point cloud individualization and semantic segmentation. This step aims to accurately segment the unordered point cloud data into independent, semantically labeled subsets corresponding to BIM components. Based on the received inspection task unit UUID, the server retrieves a list of GUIDs and their precise geometric definitions (such as bounding polygons, extrusion vectors, and spatial positions) for all BIM components within the task unit from the associated BIM database. For each BIM component in the list, the server performs the following automated operations: First, it extracts the precise digital geometric surface model of the wall from the BIM definition. Then, within the received, denoised point cloud data, it delineates a virtual 3D search space centered on the BIM wall geometric model and extending uniformly outward by 30 cm. Within this defined search space, the server performs a directed random sampling consensus (Directed RANSAC) plane fitting algorithm. Since the sampling process is strictly constrained within a small range containing the target component, the interference of irrelevant point clouds in the distance is greatly eliminated, ensuring the extremely high efficiency and robustness of the plane fitting process.

[0044] All point cloud points within the optimally fitted plane (i.e., inliers) are considered to belong to the wall's point cloud subset. The server separates this point cloud subset from the main point cloud and assigns it a semantic label derived directly from the BIM component's GUID. This process is repeated sequentially for all BIM components within the task unit (walls, slabs, columns, door openings, etc.) until the entire room's point cloud data is completely and accurately segmented into a series of individual component point clouds with unique semantic identities, each corresponding to a specific BIM component.

[0045] like Figure 6As shown in Figure 1, the fourth step of the process is the automated judgment of the quality compliance of multi-dimensional data. After completing the individual segmentation, in order to ensure the accuracy and reliability of the subsequent deviation analysis, the system will conduct a quantitative quality assessment on the point cloud subset of each component. The assessment system includes three core indicators: coverage completeness S cov , data validity S den and feature completeness S feat Coverage completeness S cov The calculation method is: project the segmented component point cloud onto the corresponding BIM component geometric surface, calculate the covered area, and then divide it by the total surface area of ​​the BIM component itself to get a percentage. den The calculation method is: calculate the average point density of the component point cloud and divide it by the preset target density threshold ρ target , the result is normalized and capped at 1.0. Feature completeness S feat The calculation method of is more sophisticated: the system first uses the 3D Harris corner detection algorithm on the precise geometric model of the BIM component to extract all its geometric corner points and sharp edge segments, and uses the sampling point set of these points and segments as the key feature point set. Then, for each key feature point, the system checks whether there are at least 10 points from the corresponding component point cloud subset within a spherical neighborhood with a radius of 5 cm in its 3D space. Feature completeness S feat The final value is defined as the percentage of key feature points that are successfully verified (i.e., the number of points in the neighborhood meets the standard) to the total number of key feature points.

[0046] Finally, a comprehensive quality score Q score The formula is obtained by taking the weighted average of the above three indicators: Q score = 0.5 * S cov + 0.3 * S den + 0.2 * S feat The setting of weight coefficients reflects the importance attached to different quality dimensions, among which coverage completeness is considered to be the most important. score Compare with a preset passing threshold (e.g., 85 points). score If Q is greater than or equal to the threshold, the data quality of the component is determined to be qualified and the process continues. scoreIf the data quality falls below this threshold, the system is deemed unqualified. At this point, the system automatically generates a command and sends a re-measurement request to the on-site operator's scanning terminal via the mobile app's push notification service. This request clearly specifies the GUID of the component whose data quality failed and the specific reason for the failure (for example, "coverage completeness is only 70%, data is missing near the east wall corner," or "feature points on the upper edge of the door opening are missing"), providing precise, closed-loop guidance for re-measurement.

[0047] like Figure 7 As shown in the figure, after the data quality of all relevant components is judged to be qualified, the system will start two core analysis tasks in parallel: overall spatial position deviation detection and component-level geometric dimension deviation detection.

[0048] For overall spatial position deviation detection, the system treats all successfully segmented and labeled component point clouds as a rigid whole and performs a global optimal registration with the corresponding component geometry set in the BIM model. This registration uses the more robust and convergent Levenberg-Marquardt iterative closest point algorithm (LM-ICP) and sets strict convergence criteria (for example, the change in the transformation matrix between two iterations is less than 1e-6 or the maximum number of iterations is 100). The algorithm ultimately obtains an optimal 4×4 homogeneous transformation matrix T that accurately transforms the entire point cloud from the current measured position to the BIM design position. deviation This matrix itself contains information about the overall spatial deviation of the room in its built state relative to the design blueprint. The system further decomposes the matrix into translation vectors (ΔX, ΔY, ΔZ) and rotation Euler angles (ΔRx, ΔRy, ΔRz). These values ​​are used as the final quantified results of the room's overall position and posture deviation.

[0049] For component-level geometric dimensional deviation detection, the system invokes specific analysis algorithms for different component types. Taking door and window openings as an example, after extracting a subset of the individual point clouds, the system first determines the principal plane through principal component analysis (PCA) and projects all point cloud points onto this plane, thereby reducing the dimensionality of the three-dimensional problem to two dimensions. On this two-dimensional projected point set, the system employs the Hough Transform algorithm, which, due to its insensitivity to noise and missing data, robustly detects the four straight lines that form the opening's boundary. By calculating the intersection of these four straight lines, the system accurately determines the coordinates of the opening's four corner points. Furthermore, by calculating the distances between the corner points, the measured width and height of the opening are determined. The system then retrieves the design values ​​for the Width and Height parameters of the corresponding IfcOpeningElement object from the BIM database. Subtracting the measured values ​​from the design values ​​yields the dimensional deviation of the opening.

[0050] Taking a wall component as an example, the system performs two key geometric tolerance analyses on its individualized point cloud subsets. The first is a flatness check: the system uses the least squares method to perform an optimal plane fit on the wall point cloud, obtaining a reference plane equation Ax+By+Cz+D=0 that represents the overall trend of the wall surface. The system then calculates the orthogonal distance from each point in the point cloud subset to this reference plane, compares the absolute values ​​of all distances, and takes the maximum value, which is defined as the flatness deviation of the wall surface. The second is a perpendicularity check: the system extracts the unit normal vector n = (A, B, C) of the fitted reference plane and calculates the angle θ = arccos(n·v) between this normal vector and the Z-axis unit vector v = (0,0,1), representing the absolute vertical direction. This angle θ is the wall's perpendicularity deviation and can be directly converted into millimeters per meter of height.

[0051] The last step of the process is the logical compound judgment of the results and the delivery of multi-dimensional data. After the deviation values ​​of all inspection items are calculated, the system will start a flexibly configurable specification rule engine. The engine is pre-loaded with a structured database, which stores the relevant national construction engineering construction quality acceptance standards (such as the "Unified Standard for Construction Quality Acceptance of Construction Engineering" GB50300) or the tolerance limits customized by the project party. Each record in the database contains the component type, the name of the inspection item, and the upper and lower limits of the allowable tolerance. For example, a typical record is (component type: indoor plastered wall, inspection item: surface flatness, tolerance limit: 4mm). The system will automatically compare each calculated deviation value with the corresponding entry in the rule engine.

[0052] If the deviation values ​​for all test items fall within their respective tolerance limits, the final result of the inspection task unit is considered "passed." If any one or more items exceed the tolerance, the result is considered "failed." The system highlights all failed items and records them in detail. The process ultimately summarizes all pass and fail results.

[0053] Based on this summary, the system automatically generates a structured inspection report, delivered via a mobile app and a supporting web portal. This report presents inspection results in a multi-dimensional, highly visual manner. Within the mobile app, the system displays inspection results directly overlaid on the 3D BIM model, using intuitive color coding (e.g., green for qualified components, red for non-compliant components) and digital labels attached to components to visually display the deviation status of each component. The system also supports one-click generation of a detailed inspection report in standard archival format, either PDF or Excel. This rigorous and comprehensive report includes a project overview, the inspection scope (corresponding to the task unit UUID), a summary of deviation data for all inspection items, a detailed list of non-compliant items and their specific deviation values, and, for key non-compliant components (such as walls with severe non-compliant flatness), accompanying deviation visualizations. For example, a visualization analysis of wall flatness deviations shows each point in the wall point cloud assigned a specific color value based on its orthogonal distance to the fitted reference plane. For example, a continuous color spectrum, ranging from blue (representing concave areas) to green (representing flat areas) to red (representing convex areas), clearly and quantitatively demonstrates the precise distribution and severity of wall unevenness. This completes a complete, automated process for detecting room location and dimensional deviations.

[0054] In order to further illustrate the technical effects of the present invention, a specific embodiment and a comparative example are given below.

[0055] Example 1 This example tests the interior space of room "R-401" on the 4th floor of Building A of a commercial complex. The room has a design dimension of 5.0m x 4.0m and a height of 3.0m. It includes four interior walls, a floor, a ceiling, a doorway measuring 2.1m x 0.9m, and a window opening measuring 1.5m x 1.2m.

[0056] The operator used the method described in the present invention. After loading the project BIM model on the mobile application, the system automatically identified "R-401" as an independent inspection task unit with a UUID of {a1b2c3d4-e5f6-7890-1234-567890abcdef}. The AR guidance path automatically generated by the system is 25.3 meters long and contains 12 key posture points. The operator followed the AR guidance and completed the data collection in 2 minutes and 15 seconds. During this period, the real-time quality monitoring interface showed that the coverage of all walls, floors, and ceilings met the green standard. The scan generated a total of 15.24 million original point clouds. After clicking Submit, the terminal took 28 seconds to complete the curvature adaptive downsampling locally, generated a feature-enhanced point cloud of 2.81 million points (data compression rate of 18.4%), and uploaded it to the cloud server within 15 seconds via the 5G network.

[0057] After the cloud server receives the data, it takes 35 seconds to complete GCN noise reduction, BIM-driven semantic segmentation and automatic data quality assessment. score All scores were above 85, indicating passing. The system then initiated a deviation analysis in parallel, completing all calculations in 42 seconds. The results showed that the room had a positive translation of 8mm along the X-axis and a rotation of 0.05 degrees about the Z-axis. The north wall had a flatness of 3.2mm and a verticality of 0.08 degrees. The door opening measured a width of 902mm (+2mm deviation) and a height of 2095mm (-5mm deviation). All deviations were within the project's pre-defined specification limits.

[0058] Finally, the system generated the report in 5 seconds. The total time from scanning to obtaining the detailed PDF report was (135s + 28s + 15s + 35s + 42s + 5s) = 260 seconds, or approximately 4.3 minutes. The report was pushed to the project manager's phone in real time via the app and updated simultaneously on the web.

[0059] Comparative Example 1 This comparative example uses traditional handheld scanning combined with post-processing software to detect the same "R-401" room as Example 1.

[0060] The operator used the same handheld scanner, but without the AR guidance and real-time quality monitoring features. Relying on experience, the operator walked around the room scanning, which took about three minutes. After the scan was complete, the raw point cloud data, containing over 15 million points, was exported to the workstation computer via a data cable, which took about five minutes.

[0061] Data processing engineers open the data on their workstations using third-party point cloud processing software (such as CloudCompare). Manual denoising and rough cropping take approximately 15 minutes. Then, by manually selecting points, they segment the point clouds for walls, floors, ceilings, door openings, and window openings one by one. This process depends on the engineer's skill level and takes approximately 45 minutes, and the segmentation boundaries are subject to subjective errors. After segmentation, each component's point cloud is manually aligned with the corresponding STL model exported from the BIM software. ICP registration is performed on each component, taking approximately 20 minutes.

[0062] After registration, the software's built-in tools were used to analyze deviations. Wall flatness analysis took about 5 minutes, and door and window opening dimensions took about 10 minutes. Finally, screenshots of all results, copying the data, and manually writing the inspection report took about 30 minutes.

[0063] During the data analysis phase, it was discovered that the point cloud at the northwest corner of the room was sparse, a sparse area that had not been noticed during the scan. This resulted in low confidence in the flatness and verticality analysis results for this wall. To ensure quality, the operator had to return to the site for a retest of this area, incurring at least 30 minutes of additional communication, travel, and work time.

[0064] Without considering rework, the total time from starting the scan to obtaining the report is (3min+5min+15min+45min+20min+15min+30min) = 133 minutes, or about 2.2 hours.

[0065] The key performance indicators of Example 1 and Comparative Example 1 are quantitatively compared in conjunction with Table 1: Table 1 Comparison of key performance indicators Through the detailed description and data comparison of the above embodiments and comparative examples, it can be clearly seen that the automatic detection method of room position size deviation based on handheld point cloud acquisition described in the present invention, by constructing a BIM-driven, end-cloud collaborative fully automated process, shows great advantages over the existing technology in detection efficiency, data quality, result reliability and ease of operation.

[0066] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for automatically detecting room position and size deviations based on handheld point cloud acquisition, relying on a system consisting of a handheld scanning terminal integrated with a 3D lidar sensor, an inertial measurement unit, and a visible light camera, a mobile application running on the scanning terminal, and a cloud server, characterized by: The method includes: Task loading and augmented reality scanning path planning: load the building information model of the target project through the mobile application, automatically parse and decompose the building information model into inspection task units with unique identifiers, and automatically plan a spatial three-dimensional path for the selected inspection task unit. The spatial three-dimensional path is visually guided on the display interface of the scanning terminal through augmented reality; On-site data collection and real-time quality monitoring: Operators use handheld scanning terminals and follow augmented reality visual guidance to move. By tightly coupling the point cloud data of the 3D lidar sensor with the readings of the inertial measurement unit, the equipment posture is solved in real time and the original point cloud data is generated synchronously. At the same time, the collected point cloud data is lightweight registered with the building information model in real time, and the coverage quality of the point cloud on the surface of the building information model components is evaluated and visualized in real time; End-to-end collaborative point cloud preprocessing: Feature-sensitive adaptive downsampling is performed on the collected raw point cloud data locally at the scanning terminal. The downsampled point cloud data is uploaded to the cloud server, where it is subjected to noise reduction. Using the building information model as prior knowledge, the point cloud data is automatically segmented into semantically labeled individual component point clouds that correspond one-to-one to components in the building information model. Automatically judge the quality compliance of multi-dimensional data. For each individual component point cloud, a comprehensive quality score is calculated based on the three dimensions of coverage completeness, data validity, and feature completeness. The score is then compared with the preset qualified threshold. If the score fails, a re-measurement instruction is generated. Deviation detection and result delivery: For point clouds of qualified individual components, we perform overall spatial position deviation detection and component-level geometric dimension deviation detection, perform logical compound judgment on the detected deviation values ​​and preset specification rules, and generate a structured detection report including a visual analysis diagram.

2. The method according to claim 1, characterized in that In the task loading and augmented reality scanning path planning, the process of automatically parsing and decomposing the building information model into detection task units specifically includes: Traversing the industrial foundation class data structure of the building information model, identifying and extracting entity information of floors, axis grids, spaces, and custom attribute sets related to construction flow segment division; Based on the entity information, the project model is automatically decomposed into a series of independent inspection task units, and a data structure containing the following information is created for each inspection task unit: a unique identifier of the task unit, the floor and area identifier to which it belongs, a globally unique identifier of the spatial object to which it is associated, a three-dimensional geometric bounding box of the space, and a list of globally unique identifiers of all building information model components within the space.

3. The method according to claim 1 or 2, characterized in that In the task loading and augmented reality scanning path planning, the process of automatically planning the spatial three-dimensional path specifically includes: voxelize the three-dimensional geometric space of the detection task unit, and mark the voxels occupied by the component entities as obstacles according to the building information model; In the non-obstacle area, an improved fast-expanding random tree star algorithm is used for path search, wherein the cost function of the algorithm is defined as a weighted sum of the path length, the visibility of the path points on the surface of the building information model component, and the expected scanning coverage; The final output is a three-dimensional spatial path consisting of a series of six-degree-of-freedom pose points arranged in sequence.

4. The method according to claim 1, wherein In the field data collection and real-time quality monitoring, the process of real-time evaluation and visualization of point cloud coverage quality specifically includes: In a separate computational thread, the newly acquired point cloud data is lightweight-aligned with the pre-loaded building information model using a single iteration of the generalized iterative closest point algorithm, using the device pose output by the real-time positioning and mapping system as the initial transformation matrix. The triangular mesh model of each building information model component is parametrically expanded into a two-dimensional UV coordinate space, and a two-dimensional rasterized cumulant matrix corresponding to the UV space is maintained; For each newly collected and registered point cloud point, calculate its nearest projection point on the surface of the building information model component, and record the UV coordinate corresponding to the projection point in the cumulant matrix, so that the count value of the corresponding grid is increased; According to the count values ​​in the cumulant matrix, the building information model surface is rendered in real time into different colors for visual display: when the grid count value corresponding to a certain area is greater than a preset density threshold, it is rendered in a first color; when the count value is between 0 and the preset density threshold, it is rendered in a second color; when the count value is 0, it is rendered in a third color.

5. The method according to claim 1, wherein During the point cloud preprocessing process of the end-cloud collaboration, the feature-sensitive adaptive downsampling process performed locally on the scanning terminal specifically includes: Construct an octree index from the collected complete original point cloud data; For each point cloud subset contained in the octree leaf node, calculate the covariance matrix of its three-dimensional coordinates; Performing eigenvalue decomposition on the covariance matrix to obtain three eigenvalues ​​λ1, λ2, and λ3, where λ1≥λ2≥λ3; Calculate a normalized curvature metric value C=λ3 / (λ1+λ2+λ3) according to the eigenvalue; The retention strategy of the point cloud in the leaf node is determined according to the size of the curvature metric value C. This strategy makes the leaf node with higher curvature metric value have higher probability of retaining the point cloud in its interior completely, while the leaf node with lower curvature metric value has higher probability of downsampling the point cloud in its interior with its centroid point as the representative.

6. The method according to claim 1 or 5, characterized in that In the point cloud preprocessing process of the end-cloud collaboration, the process of automatically segmenting the point cloud data into individual component point clouds by using the building information model as prior knowledge on the cloud server specifically includes: According to the received unique identifier of the inspection task unit, a list of globally unique identifiers of all building information model components in the unit and their precise geometric definitions and spatial location information are retrieved from the database; For each BIM component in the list, do the following: Extracting its digital geometric surface from its building information model definition, and expanding outwards from the geometric surface as the center to a preset distance, thereby defining a three-dimensional search space; In the three-dimensional search space, a directed random sampling consensus algorithm is executed to fit a point cloud subset that matches the geometric characteristics of the component; The fitted point cloud subset is separated from the main point cloud and assigned a globally unique identifier derived from the building information model component as a semantic label.

7. The method according to claim 1, characterized in that The automated judgment process of the multi-dimensional data quality compliance specifically includes: For each individual component point cloud, calculate its coverage completeness, data validity and feature completeness respectively; The coverage completeness is calculated by projecting the segmented component point cloud onto the geometric surface of the corresponding building information model component, divided by the total surface area of ​​the building information model component; The data validity is calculated by dividing the average point density of the component point cloud by a preset target density threshold; the feature completeness is calculated by extracting a set of key feature points from the geometric model of the building information model component and checking whether there are sufficient number of points from the corresponding component point cloud subset in the neighborhood of each key feature point in three-dimensional space; The comprehensive quality score is calculated by taking a weighted average of the coverage completeness, data validity and feature completeness; The comprehensive quality score is compared with a preset qualified threshold. If it is lower than the threshold, it is judged as unqualified, and a retest instruction including the unqualified component identification and the specific reason for the unqualified component is generated and pushed to the operator through the mobile application.

8. The method according to claim 1, characterized in that In the deviation detection and result delivery, the process of performing the overall spatial position deviation detection specifically includes: All successfully segmented and labeled individual component point clouds are treated as a rigid whole and globally optimally aligned with the corresponding component geometry set in the building information model; The global optimal registration uses the Levenberg-Marquardt iterative closest point algorithm to obtain an optimal 4×4 homogeneous transformation matrix that transforms the entire point cloud from the measured position to the design position of the building information model; The translation vector and the rotation Euler angle are decomposed from the homogeneous transformation matrix as the quantized results of the overall position and posture deviation of the room.

9. The method according to claim 1, characterized in that In the deviation detection and result delivery, the process of performing component-level geometric dimension deviation detection includes checking the flatness and verticality of wall components, where: The flatness check specifically includes: using the least squares method to perform optimal plane fitting on the individualized point cloud subset of the wall component to obtain a reference plane equation, and calculating the orthogonal distance from each point in the point cloud subset to the reference plane, and the maximum value of the absolute values ​​of all distances is used as the flatness deviation of the wall surface; The verticality check is specifically performed by extracting the unit normal vector of the reference plane, and calculating the angle between the normal vector and the Z-axis unit vector representing the absolute vertical direction, and using the angle as the verticality deviation of the wall.

10. The method according to claim 1, characterized in that During the deviation detection and result delivery process, the process of generating a structured test report includes: Overlaying the inspection results on the three-dimensional model of the building information model, using color coding and numerical labels to visually display the qualified or out-of-tolerance status of each component; In addition, for wall components with excessive flatness deviation, a flatness deviation visualization analysis diagram is generated. This analysis diagram is a heat map, in which each point in the wall point cloud is assigned a color value according to its distance to the fitted reference plane, clearly showing the distribution and severity of the uneven areas on the wall through a continuous color spectrum.

Images