Building surveying system

Abstract

The application relates to the technical field of digital surveying and mapping and construction quality control of building engineering, and discloses a house surveying and mapping system for building engineering, which comprises a data acquisition module that synchronously generates panoramic image signals and inertial measurement unit signals; a preprocessing module that receives the panoramic image signals and the inertial measurement unit signals and generates a pose calibration signal through visual-inertial odometer fusion; a fusion modeling module that receives a purified point cloud signal and the pose calibration signal and generates building component parameterized data based on a deep learning component recognition algorithm; a result output module that receives the building component parameterized data and generates a construction deviation analysis report based on a preset tolerance threshold; and an intelligent control module that generates a collection quality feedback signal to the data acquisition module through a real-time quality evaluation algorithm to adjust a scanning path, and sends a dynamic sampling rate control signal to the preprocessing module to optimize a calculation load. The application solves the problems of low house surveying and mapping efficiency, poor modeling precision and weak on-site real-time feedback.

Description

Technical Field

[0001] The present invention relates to the technical field of digital surveying and mapping of construction projects and construction quality control, and in particular to a house surveying and mapping system for construction projects. Background Art

[0002] Current construction project surveying and mapping relies primarily on traditional equipment such as total stations and theodolites, requiring manual point-by-point measurement and data recording. This process is inefficient, susceptible to environmental interference, and exhibits significant human error. While 3D laser scanning technology has significantly increased data acquisition speed in recent years, point cloud data processing still relies on specialized software for denoising, registration, and segmentation, which can take hours. Furthermore, scanner accuracy plummets in strong light and dusty environments, and the ability to capture obscured areas is limited. Existing automated modeling solutions often utilize offline processing, making it impossible to generate a visual model in real time on-site, preventing immediate correction of construction deviations. Furthermore, the conversion of point clouds into building information models is highly manual: technicians must manually segment components and annotate parameters, resulting in a lengthy modeling cycle and the introduction of secondary errors. Construction deviation analysis is often deferred to the acceptance phase, with manual reports compiled by comparing design drawings with measured data. This delays problem rectification and increases rework costs.

[0003] Patent authorization announcement number CN117371961B discloses a mobile-based method and system for field data mapping and collection of farmhouse data, reducing the tediousness of traditional pen-and-paper record-keeping. The industry has attempted to introduce robotic platforms equipped with sensors to improve automation, but insufficient multi-sensor coordination leads to data asynchrony in time and space, requiring complex post-processing and alignment. The robustness of the visual and laser data fusion algorithm is poor, and registration errors often occur in complex indoor scenes. Research on deep learning component recognition has focused on laboratory environments. Interference factors such as component incompleteness, occlusion, and material reflectivity in actual construction sites significantly reduce recognition accuracy. Existing systems lack a closed-loop control mechanism for the entire process: repeated scans are required when data collection quality is uncontrollable, and rigid computing resource allocation leads to processing delays on the mobile side, making it difficult to support on-site decision-making. Summary of the Invention

[0004] In view of the above shortcomings of the prior art, the purpose of the present invention is to provide a house surveying and mapping system for construction engineering, which is used to solve the problems of low house surveying and mapping efficiency, poor modeling accuracy, and weak on-site real-time feedback.

[0005] The building surveying and mapping system for construction engineering of the present invention comprises:

[0006] A data acquisition module, which synchronously generates panoramic image signals and inertial measurement unit signals;

[0007] The preprocessing module receives panoramic image signals and inertial measurement unit signals, forms a purified point cloud signal through a real-time denoising algorithm, and generates a pose calibration signal through visual inertial odometry fusion;

[0008] The fusion modeling module receives the purified point cloud signal and the pose calibration signal, maps the panoramic image signal to the point cloud space to form a colored point cloud signal through a multimodal registration algorithm, and generates parametric data of building components based on a deep learning component recognition algorithm;

[0009] The result output module receives the parametric data of building components, generates a three-dimensional building model signal through the BIM engine, and generates a construction deviation analysis report based on the preset tolerance threshold;

[0010] Intelligent control module, the intelligent control module is connected to the data acquisition module, preprocessing module and fusion modeling module respectively, and generates an acquisition quality feedback signal to the data acquisition module through a real-time quality assessment algorithm to adjust the scanning path, and at the same time sends a dynamic sampling rate control signal to the preprocessing module to optimize the computing load.

[0011] In one embodiment of the present invention, the data acquisition module includes a lidar unit and a panoramic camera unit, which are fixed to the same rotating base through a rigid connector; the lidar unit generates a point cloud raw data stream in a preset scanning mode, and the panoramic camera unit triggers a global shutter within the synchronization period of the lidar emission pulse to generate a panoramic image signal with timestamp alignment; the inertial measurement unit signal is generated by a three-axis gyroscope and accelerometer embedded in the rotating base, and the angular velocity and linear acceleration data are bound to the point cloud raw data stream into a time-space synchronization signal package through a pulse counter.

[0012] In one embodiment of the present invention, the pre-processing module executes a real-time denoising algorithm comprising three processing stages; the first stage uses an inertial measurement unit signal based on a motion distortion compensation model to perform pose interpolation correction on the original point cloud data stream; the second stage uses an adaptive voxel filtering algorithm to dynamically adjust the voxel grid size according to the point cloud density distribution to filter out environmental suspended particle noise; the third stage uses a geometric feature enhancement algorithm to extract the point cloud curvature mutation area and strengthen the edge point weights, and finally outputs a purified point cloud signal that retains key structural features.

[0013] In one embodiment of the present invention, the visual-inertial odometry fusion process adopts a sliding window optimization framework to jointly optimize the inertial measurement unit signal and the panoramic image feature point observation values ​​in the continuous time domain; by establishing a joint cost function of the reprojection error and the inertial measurement error, the camera pose change is iteratively solved; when a closed-loop feature point is detected, the global pose graph optimization is triggered, and a pose calibration signal is generated to eliminate the accumulated error and the spatial coordinate system transformation matrix is ​​updated in real time.

[0014] In one embodiment of the present invention, the multimodal registration algorithm includes a dual mechanism of spherical projection transformation and feature matching; first, the panoramic image signal is projected into a virtual spherical coordinate system, and the initial corresponding point pairs are established through the geometric mapping relationship between the spherical coordinates and the Cartesian coordinates of the point cloud; secondly, the panoramic image texture features and the point cloud normal vector features are extracted, and bidirectional matching is performed based on the similarity of feature descriptors. After eliminating the mismatched points, the optimal rigid body transformation parameters are solved to achieve high-precision color mapping of the panoramic image signal to the point cloud space and generate a color point cloud signal.

[0015] In one embodiment of the present invention, a deep learning component recognition algorithm adopts a multi-scale feature fusion network architecture; the network input layer receives spatial context information derived from color point cloud signals and posture calibration signals, the first branch extracts local geometric features through three-dimensional sparse convolution, the second branch constructs component topological relationships through a graph convolutional neural network, and the third branch uses an attention mechanism to weightedly fuse multi-scale features; the output layer regresses component type probability and bounding box parameters through a fully connected layer to generate parametric data of building components containing component geometric dimensions and spatial orientation.

[0016] In one embodiment of the present invention, the BIM engine drives the instantiation of parametric components and the reconstruction of topological relationships; calls a predefined component family library template based on the type identifier in the parametric data of the building component, injects the geometric size parameters into the template to generate a parametric component instance; and automatically constructs a topological connection diagram of the wall adjacent relationship, floor support relationship, and pipeline penetration relationship by analyzing the spatial orientation data of the component instance, and finally assembles it into a three-dimensional building model signal that meets the industrial foundation class standard.

[0017] In one embodiment of the present invention, the real-time quality assessment algorithm includes dual indicators of coverage analysis and confidence assessment; the coverage analysis generates an area integrity coefficient by calculating the ratio of the point cloud density of the scanned area to a preset density threshold; the confidence assessment generates an identification reliability coefficient by monitoring the classification probability entropy value and the bounding box regression residual in the deep learning component recognition algorithm; the above coefficients are combined to generate an acquisition quality feedback signal, which includes a scanning path correction vector and a device posture adjustment instruction set.

[0018] In one embodiment of the present invention, a dynamic sampling rate control signal is generated by a computational complexity prediction model; the model receives a density distribution histogram of the purified point cloud signal and the optimized number of iterations of the pose calibration signal, and predicts the computing resources required for subsequent frame data processing; when the predicted value exceeds the real-time processing capability threshold of the preprocessing module, a sampling rate control instruction is generated to reduce the number of lidar scan lines or limit the panoramic image resolution, ensuring that the data processing delay remains within an interactive time range.

[0019] In one embodiment of the present invention, the result output module integrates a lightweight rendering engine to convert the three-dimensional building model signal into a multi-level-of-detail model data stream; automatically selects the model detail level according to the performance of the user terminal device, and outputs the interactive three-dimensional model to the mobile terminal through a progressive transmission protocol; and simultaneously generates a structured construction deviation analysis report, which includes a three-dimensional coordinate index of the deviation position, a vector diagram of the deviation amount, and a text of recommended corrective measures.

[0020] Beneficial effects: The house surveying and mapping system for construction projects provided by the present invention generates time-space synchronized panoramic images and measurement signals by integrating lidar, panoramic cameras and inertial units, and forms purified data through posture calibration and point cloud denoising; uses a multimodal registration algorithm to map image textures to point cloud space to generate a color point cloud, identifies building components based on deep learning and outputs parametric data; generates three-dimensional models and construction deviation reports in real time through a building model engine, and the intelligent control module dynamically adjusts the acquisition path and computing load based on quality assessment, thereby realizing closed-loop optimization of the entire surveying and modeling process. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 This is a system architecture diagram of the house surveying and mapping system for construction engineering of the present invention. DETAILED DESCRIPTION

[0023] The following describes the embodiments of the present invention through specific examples. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other unless they conflict.

[0024] See Figure 1 As shown, the house surveying and mapping system for construction engineering of the present invention includes a data acquisition module, which synchronously generates a panoramic image signal and an inertial measurement unit signal;

[0025] The preprocessing module receives panoramic image signals and inertial measurement unit signals, forms a purified point cloud signal through a real-time denoising algorithm, and generates a pose calibration signal through visual inertial odometry fusion;

[0026] a fusion modeling module receiving the purified point cloud signal and the pose calibration signal, mapping the panoramic image signal to the point cloud space to form a color point cloud signal through a multi-modal registration algorithm, and generating building component parameterized data based on a deep learning component recognition algorithm;

[0027] a result output module receiving the building component parameterized data, generating a three-dimensional building model signal through a BIM engine, and generating a construction deviation analysis report based on a preset tolerance threshold;

[0028] an intelligent control module connected with the data acquisition module, the preprocessing module and the fusion modeling module, generating an acquisition quality feedback signal to the data acquisition module through a real-time quality evaluation algorithm to adjust the scanning path, and sending a dynamic sampling rate control signal to the preprocessing module to optimize the calculation load.

[0029] Figure 1 As shown, the core architecture is composed of five functional modules to form a closed-loop workflow. The data acquisition module, as the physical perception layer, synchronously generates two types of original signals through an integrated sensor array: one is the panoramic image signal captured by the optical imaging unit, which covers a horizontal 360-degree and vertical 180-degree field of view in a spherical projection manner, and records the texture information of the house structure completely; the other is the six-degree-of-freedom motion signal generated by the micro-electromechanical inertial measurement unit, which includes time series data streams of three-axis angular velocity and three-axis linear acceleration. The preprocessing module, as the real-time processing center, performs two types of key operations in parallel after receiving the above two signals: first, for the original point cloud data stream, a real-time denoising algorithm based on statistical outlier rejection and motion distortion compensation is adopted to filter out noise points introduced by environmental dust, flying insects and device vibration, and generate a purified point cloud signal that retains the characteristics of the building structure; second, through tightly coupled visual-inertial odometry technology, the feature point trajectories of the panoramic image sequence are spatio-temporally aligned with the inertial measurement unit signal, a joint optimization objective function of re-projection error and inertial pre-integral error is constructed, and the device pose change is solved and the pose calibration signal is output, which includes the rotation matrix and translation vector of the device in the global coordinate system.

[0030] Furthermore, the fusion modeling module undertakes the functions of multi-source data fusion and semantic analysis: after receiving the purified point cloud signal and pose calibration signal from the preprocessing module, its multimodal registration algorithm first uses the spatial transformation parameters of the pose calibration signal to map the panoramic image signal from the two-dimensional pixel coordinate system to the three-dimensional point cloud space, and by solving the optimal rigid body transformation of the image texture feature points and the point cloud geometric feature points, it realizes the precise binding of color attributes and geometric structures, and generates a color point cloud signal with real texture; then, the component recognition algorithm based on the deep convolutional neural network performs voxel segmentation on the color point cloud, extracts local geometric features through the three-dimensional convolution layer, regresses the component type probability distribution through the fully connected layer, and outputs parametric data of building components including parameters such as wall thickness, beam and column cross-sectional dimensions, and door and window positioning coordinates. The result output module delivers results: after receiving the parametric data of building components, it calls the application program interface of the building information model engine, instantiates the parametric family library template according to the component type identifier, automatically assembles topological relationships such as beam-column nodes and wall panel connections, and generates a three-dimensional building model signal that meets the industrial foundation standards; at the same time, it performs spatial overlay analysis on the model and the preset design model, calculates the Euclidean distance deviation of key control points, and generates a structured construction deviation analysis report containing the three-dimensional coordinates of the deviation position, the deviation vector direction, and rectification suggestions when the deviation exceeds the tolerance threshold.

[0031] In one embodiment of the present invention, the intelligent control module serves as the system optimization hub, monitoring data quality throughout the entire process through a real-time quality assessment algorithm. It receives the component identification confidence index output by the fusion modeling module and, combined with the point cloud coverage density statistics from the preprocessing module, calculates the integrity coefficient and reliability coefficient of the current scanning area. If the coefficient is lower than a preset threshold, it generates an acquisition quality feedback signal to the data acquisition module. This signal includes a scan path correction vector (such as increasing the pitch angle scan) and an instruction to adjust the device's motion speed. Simultaneously, it monitors the computational delay of the preprocessing module. When the point cloud frame processing time exceeds the real-time threshold, it sends a dynamic sampling rate control signal to the preprocessing module, instructing it to adaptively reduce the point cloud sampling density or limit the image resolution based on the computational load, ensuring that the system maintains an interactive response speed. Signals are transmitted between modules via a high-speed serial bus, forming a complete technical chain from data acquisition, processing, modeling, to closed-loop optimization.

[0032] The data acquisition module's hardware configuration is further refined: it comprises two core components: a lidar unit and a panoramic camera unit. These are rigidly fixed to a common aluminum alloy rotating base via a high-rigidity carbon fiber connector, ensuring optical axis deviation within 0.1 mm and a constant relative position. The lidar unit utilizes the time-of-flight ranging principle, emitting near-infrared pulsed laser light in a preset spiral raster scanning pattern. Reflected photons are received and generated as a point cloud raw data stream containing distance values ​​and reflection intensities. The scan rate can be dynamically adjusted between 100,000 and 2 million points per second. The panoramic camera unit comprises a circular array of multiple fisheye lenses, each with its optical axis at a predetermined angle to the base's rotation axis. An image stitching algorithm is employed to generate a seamless panoramic image signal. Its exposure triggering mechanism is strictly synchronized with the lidar pulse emission cycle: When the lidar transmit circuit generates a pulse rising edge, it triggers the panoramic camera's global shutter controller, ensuring that each laser ranging point and the corresponding image pixel have the same timestamp.

[0033] The IMU signal is generated by miniaturized sensors embedded in the rotating base: a three-axis MEMS gyroscope measures the base's angular velocity at a kilohertz sampling rate, while a three-axis MEMS accelerometer measures linear acceleration. The raw data from these two sources is temperature-compensated to eliminate drift errors before being fed into a high-precision data fusion chip. To ensure spatiotemporal synchronization of the multi-source signals, the system employs a hardware-level synchronization strategy: With each scan pulse generated by the lidar, a pulse counter simultaneously records the current clock cycle count and writes this count into the corresponding point cloud data packet header. The IMU data packet header also includes the same clock cycle count, and the panoramic image frame header includes the cycle count when the pulse counter was triggered. The data acquisition module's central processor, based on the clock cycle count tagged in the header, bundles the simultaneously acquired point cloud raw data streams, panoramic image signals, and IMU signals into a spatiotemporally synchronized signal packet. This packet is then transmitted to the preprocessing module via a Gigabit Ethernet interface. This design eliminates motion blur and registration errors caused by traditional time-sharing acquisition, paving the way for subsequent fusion processing.

[0034] Furthermore, the preprocessing module's real-time denoising algorithm comprises three cascaded processing stages. The first stage performs motion distortion compensation: Utilizing the angular velocity and linear acceleration data from the inertial measurement unit (IMU) signal, combined with the lidar's internal scan timing model (i.e., the precise emission time of each laser point), a continuous-time function of the device's motion trajectory is constructed. Based on this function, pose interpolation correction is performed on the raw point cloud data stream, transforming each laser point from the instantaneous measurement coordinate system to a unified scan start coordinate system to eliminate point cloud distortion caused by device translation or rotation. The second stage employs an adaptive voxel filtering algorithm: First, a spatial density distribution histogram of the input point cloud is calculated to identify high-density areas (such as walls) and low-density areas (such as door and window openings). The voxel grid size is then dynamically adjusted based on the density value: a small grid size is used in high-density areas to preserve detail, while a large grid size is used in low-density areas to accelerate processing. Within each voxel grid, the centroid coordinates of all points are calculated and used to replace the original point set for data reduction. Furthermore, an outlier threshold is set based on the standard deviation of the distances between neighboring points to filter out suspended particle noise points (such as dust reflections) that exceed the threshold. The third phase involves geometric feature enhancement: principal component analysis is used to calculate the surface curvature of local neighborhoods within the point cloud, identifying areas of sudden curvature changes (such as wall edges and beam-column intersections). These identified feature points are assigned high weights and prioritized for subsequent point cloud sampling. Furthermore, an edge extraction algorithm based on normal vector consistency is employed to enhance the probability of retaining structural boundary points. The resulting purified point cloud signal must meet two key criteria: structural integrity, ensuring a point cloud missing rate of less than 5% for continuous wall surfaces; and feature fidelity, requiring curvature errors at beam-column intersections to be less than 0.01 radians.

[0035] The visual-inertial odometry fusion process utilizes a keyframe and sliding window optimization strategy: FAST corner features are extracted from the panoramic image sequence and matched across frames using the BRIEF descriptor. Simultaneously, the inertial measurement unit (IMU) signal is pre-integrated to estimate the relative motion between adjacent keyframes. A cost function is established that incorporates reprojection error and inertial pre-integration error. The reprojection error calculates the Euclidean distance between the feature point's actual observed position in the current frame and its projected position based on the pose estimate; the inertial pre-integration error calculates the difference between the pre-integration result and the value predicted by the pose transformation model. The pose parameters within a sliding window (typically containing 5-10 keyframes) are iteratively optimized using the Levenberg-Marquardt algorithm. When closed-loop feature points in historical keyframes are detected, a global pose graph optimization is triggered. Using all keyframe poses as nodes and the visual observation constraints and inertial constraints as edges, a globally consistent pose calibration signal is obtained and the spatial coordinate system transformation matrix is ​​updated. This process runs in real time, ensuring a pose drift error of less than 0.1% per 100 meters.

[0036] Specifically, the visual-inertial odometry fusion process uses a sliding window optimization framework to achieve high-precision pose estimation. During system initialization, a sliding window consisting of 5 to 10 consecutive keyframes is established. Each keyframe consists of a feature point set consisting of hundreds of FAST corner features extracted from the panoramic image signal and their BRIEF descriptors. The inertial measurement unit signal is pre-integrated: the relative rotation quaternion between adjacent keyframes is calculated based on gyroscope angular velocity data, and the displacement increment is calculated using accelerometer data combined with the estimated gravity direction, forming the pre-integrated observation value. A nonlinear optimization problem is constructed within the sliding window, with the cost function consisting of a visual reprojection error term and an inertial measurement error term. The visual reprojection error term calculates the Euclidean distance between the actual pixel coordinates of the feature points in the current keyframe and their projected coordinates based on the camera pose estimate; the inertial measurement error term calculates the Mahalanobis distance between the pre-integrated observation value and the value predicted by the pose transformation model. The pose state vector (including the rotation matrix and translation vector) of all keyframes within the window is iteratively optimized using the Levenberg-Marquardt algorithm. After each iteration, the error function is linearized and the Hessian matrix is ​​updated. When the system detects a closed-loop event, it uses a bag-of-words model to retrieve feature points from similar scenes in historical keyframes. If the number of matching feature points exceeds a threshold, the closed-loop is considered established. This triggers global pose graph optimization, using the optimized keyframe poses from the sliding window as nodes, the visual reprojection constraints and the inertial pre-integration constraints as edges, and adding spatial position constraint edges for closed-loop matching point pairs to construct a sparse pose graph. The Gauss-Newton method is used to solve for the global optimal pose, ultimately outputting a pose calibration signal that eliminates cumulative errors. This signal updates the spatial coordinate system transformation matrix in real time, ensuring that the pose drift error within a 100-meter operating distance is less than one thousandth.

[0037] Specifically, the multimodal registration algorithm achieves precise color-geometry fusion through a dual mechanism of spherical projection transformation and feature matching. In the spherical projection transformation stage, the panoramic image signal is projected from a two-dimensional flattened image onto a virtual spherical coordinate system, where the center of the sphere corresponds to the camera's optical center, and the longitude and latitude of the sphere correspond to the image pixel coordinates. The camera extrinsic parameter matrix in the pose calibration signal is used to calculate the projected coordinates of each 3D point in the point cloud onto the virtual sphere, establishing a mapping relationship between the point cloud's Cartesian coordinate system and the spherical coordinate system. In the feature matching stage, a scale-invariant feature transform descriptor is extracted from the panoramic image, and normal vector histogram features are calculated on the point cloud. Initially mapped point pairs are subjected to bidirectional matching verification: forward matching searches for the most similar normal vector feature in the point cloud neighborhood based on image features; reverse matching searches for the most similar visual feature in the image neighborhood based on point cloud features. Only point pairs with consistent bidirectional matching are retained, and mismatched points are discarded. An objective function is constructed based on the retained point pairs: minimizing the spatial distance residual between image feature points and corresponding point cloud feature points, thereby solving for the optimal rigid body transformation parameters, which include a rotation matrix and a translation vector. The above process is iterated until the residual converges, ultimately assigning the red, green, and blue color values ​​of the panoramic image to the corresponding point cloud, generating a color point cloud signal with realistic texture. To address illumination variations, the algorithm incorporates a color constancy correction module: it statistically analyzes the correlation between the point cloud's reflection intensity and image brightness, dynamically adjusting the color mapping curve parameters to ensure that color consistency errors under cloudy and bright conditions are below the human eye's discernibility threshold. The result output module uses a building model engine to implement parametric modeling and construction deviation analysis. The parametric component instantiation phase involves calling a predefined component family library based on the type identifier in the building component's parametric data. For example, a rectangular cross-section parametric template is called for a "load-bearing wall." The regressed geometric dimension vectors are then injected into the templates: thickness and height parameters for wall templates, cross-section parameters for beam and column templates, and opening size parameters for door and window templates. This automatically generates 3D component geometry that meets the parametric constraints. During the topology reconstruction phase, the spatial orientation data of components is analyzed and the automatic wall connection algorithm is executed. This algorithm detects overlapping areas between adjacent walls, removes duplicate geometry, and creates coplanar connections. The floor support relationship construction algorithm calculates the contact area between the floor slab boundary and the underlying wall or beam column. If the area exceeds a threshold, a support topology connection is established. The pipeline penetration processing algorithm automatically creates circular openings and adds casing geometry when pipes intersect with walls. The final assembly is a building information model that complies with industrial foundation standards. During the construction deviation analysis phase, the generated 3D building model is spatially aligned with the pre-set design model. Key control points (such as column centers and wall corners) are selected to calculate coordinate deviation vectors. If the deviation exceeds the tolerance threshold, the deviation area is highlighted in red in the 3D model, and a structured report is generated, including the deviation component number, deviation type (position offset, dimensional error), 3D spatial error vector (ΔX, ΔY, ΔZ), and recommended corrective measures based on construction specifications.A lightweight rendering engine converts the model into a multi-level-of-detail data stream: Level 1 preserves the complete geometric topology, Level 2 simplifies surface tessellation, and Level 3 retains only the bounding box. The rendering level is automatically selected based on the graphics processing capabilities of the user terminal, supporting real-time interaction at 30 frames per second on mobile devices.

[0038] The house surveying and mapping system for construction projects of the present invention generates panoramic images and measurement signals synchronized in time and space by integrating lidar, panoramic cameras and inertial units, and forms purified data through posture calibration and point cloud denoising. It uses a multimodal registration algorithm to map image textures to point cloud space to generate a colored point cloud, and identifies building components based on deep learning and outputs parametric data. It generates three-dimensional models and construction deviation reports in real time through a building model engine, and the intelligent control module dynamically adjusts the acquisition path and computing load based on quality assessment, thereby achieving closed-loop optimization of the entire surveying and modeling process.

[0039] Therefore, the house surveying and mapping system for construction projects of the present invention can solve the problems of low house surveying and mapping efficiency, poor modeling accuracy, and weak on-site real-time feedback.

[0040] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the present invention. Anyone skilled in the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by one of ordinary skill in the art without departing from the spirit and technical principles disclosed herein are intended to be covered by the claims of the present invention.

Claims

1. House surveying and mapping system for construction engineering, characterized by: include: A data acquisition module, which synchronously generates panoramic image signals and inertial measurement unit signals; the data acquisition module includes a lidar unit and a panoramic camera unit, both of which are fixed to the same rotating base via a rigid connector; the lidar unit generates a point cloud raw data stream using a preset scanning pattern, and the panoramic camera unit triggers a global shutter within the synchronization period of the lidar emission pulse to generate a panoramic image signal with aligned timestamps; the inertial measurement unit signal is generated by a three-axis gyroscope and accelerometer embedded in the rotating base, and the angular velocity and linear acceleration data are bound to the point cloud raw data stream into a spatiotemporally synchronized signal packet via a pulse counter; A preprocessing module receives the panoramic image signal and the inertial measurement unit signal, forms a purified point cloud signal through a real-time denoising algorithm, and generates a pose calibration signal through visual-inertial odometry fusion. The visual-inertial odometry fusion process adopts a sliding window optimization framework to jointly optimize the inertial measurement unit signal and the panoramic image feature point observation values ​​in the continuous time domain. The camera pose change is iteratively solved by establishing a joint cost function of the reprojection error and the inertial measurement error. When a closed-loop feature point is detected, a global pose graph optimization is triggered to generate a pose calibration signal that eliminates the accumulated error and updates the spatial coordinate system transformation matrix in real time. a fusion modeling module, which receives the purified point cloud signal and the pose calibration signal, maps the panoramic image signal to the point cloud space to form a color point cloud signal through a multimodal registration algorithm, and generates parametric data of building components based on a deep learning component recognition algorithm; a result output module, the result output module receiving the parametric data of the building components, generating a three-dimensional building model signal through a BIM engine drive, and generating a construction deviation analysis report based on a preset tolerance threshold; The intelligent control module is connected to the data acquisition module, the preprocessing module and the fusion modeling module respectively. It generates an acquisition quality feedback signal to the data acquisition module through a real-time quality assessment algorithm to adjust the scanning path, and sends a dynamic sampling rate control signal to the preprocessing module to optimize the computing load.

2. The building surveying and mapping system for construction engineering according to claim 1, characterized in that: The preprocessing module executes a real-time denoising algorithm that includes three processing stages. The first stage uses the inertial measurement unit signal based on the motion distortion compensation model to perform pose interpolation correction on the original point cloud data stream. The second stage uses an adaptive voxel filtering algorithm to dynamically adjust the voxel grid size according to the point cloud density distribution to filter out environmental suspended particle noise. The third stage uses a geometric feature enhancement algorithm to extract the point cloud curvature mutation area and strengthen the edge point weights, and finally outputs a purified point cloud signal that retains key structural features.

3. The building surveying and mapping system for construction engineering according to claim 1, characterized in that: The multimodal registration algorithm includes a dual mechanism of spherical projection transformation and feature matching. First, the panoramic image signal is projected into a virtual spherical coordinate system, and the initial corresponding point pairs are established through the geometric mapping relationship between the spherical coordinates and the Cartesian coordinates of the point cloud. Secondly, the panoramic image texture features and the point cloud normal vector features are extracted, and bidirectional matching is performed based on the similarity of feature descriptors. After eliminating the mismatched points, the optimal rigid body transformation parameters are solved to achieve high-precision color mapping of the panoramic image signal to the point cloud space and generate a colored point cloud signal.

4. The building surveying and mapping system for construction engineering according to claim 1, characterized in that: The deep learning component recognition algorithm adopts a multi-scale feature fusion network architecture; the network input layer receives spatial context information derived from color point cloud signals and posture calibration signals, the first branch extracts local geometric features through three-dimensional sparse convolution, the second branch constructs component topological relationships through a graph convolutional neural network, and the third branch uses an attention mechanism to weightedly fuse multi-scale features; the output layer regresses component type probabilities and bounding box parameters through a fully connected layer to generate parametric data of building components containing component geometric dimensions and spatial orientations.

5. The building surveying and mapping system for construction engineering according to claim 1, characterized in that: The BIM engine drive includes parametric component instantiation and topological relationship reconstruction; According to the type identifier in the parametric data of the building component, the predefined component family library template is called, and the geometric size parameters are injected into the template to generate a parametric component instance; by analyzing the spatial orientation data of the component instance, a topological connection diagram of the wall adjacent relationship, floor support relationship and pipeline penetration relationship is automatically constructed, and finally assembled into a three-dimensional building model signal that meets the industrial foundation class standards.

6. The building surveying and mapping system for construction engineering according to claim 1, characterized in that: The real-time quality assessment algorithm includes two indicators: coverage analysis and confidence assessment. Coverage analysis generates a regional integrity coefficient by calculating the ratio of the point cloud density of the scanned area to a preset density threshold. Confidence assessment generates an identification reliability coefficient by monitoring the classification probability entropy value and bounding box regression residual in the deep learning component recognition algorithm. The above coefficients are combined to generate an acquisition quality feedback signal, which includes a scanning path correction vector and a device posture adjustment instruction set.

7. The building surveying and mapping system for construction engineering according to claim 1, characterized in that: The dynamic sampling rate control signal is generated by a computational complexity prediction model; the model receives the density distribution histogram of the purified point cloud signal and the number of optimized iterations of the pose calibration signal, and predicts the computing resources required for subsequent frame data processing; when the predicted value exceeds the real-time processing capability threshold of the preprocessing module, it generates a sampling rate control instruction to reduce the number of lidar scan lines or limit the resolution of the panoramic image, ensuring that the data processing delay remains within the interactive time range.

8. The building surveying and mapping system for construction engineering according to claim 1, characterized in that: The result output module integrates a lightweight rendering engine to convert the three-dimensional building model signal into a multi-level-of-detail model data stream; automatically selects the model detail level according to the performance of the user terminal device, and outputs the interactive three-dimensional model to the mobile terminal through a progressive transmission protocol; and simultaneously generates a structured construction deviation analysis report, which includes a three-dimensional coordinate index of the deviation location, a vector diagram of the deviation amount, and a text of recommended corrective measures.

Images