Construction building inspection method and device

Abstract

The application relates to the technical field of construction building inspection, and provides a construction building inspection method and device. The method comprises the following steps: obtaining a two-dimensional plane map of a construction building, planning a path for the construction building in combination with the two-dimensional plane map, controlling an inspection robot to inspect the construction building, obtaining three-dimensional point cloud data and a site structure image of the construction building, adopting an RANSAC algorithm and an OPTICS clustering algorithm, fitting the three-dimensional point cloud data based on a preset segmentation model, and obtaining a building information model of the construction building; inputting the site structure image into a preset quality monitoring network, calculating construction condition monitoring data of the construction building, and generating a construction progress inspection report of the construction building based on the building information model and the construction condition monitoring data. The method forms a closed-loop verification mechanism for inspecting the construction building by utilizing dynamic path planning, multi-algorithm fusion processing and construction data and model, and improves the inspection efficiency of the construction building.

Description

Technical Field

[0001] This invention relates to the field of construction inspection technology, and in particular to a construction inspection method and apparatus. Background Technology

[0002] In recent years, inspection robots have been widely used in industries such as petrochemicals, power grids, and coal mines. Leveraging their superior environmental adaptability and high-precision data acquisition capabilities, they can replace or assist manual inspections in dangerous and complex scenarios that are difficult for traditional personnel to cover. Traditional manual inspections, due to high safety requirements requiring protective equipment, are limited by the working environment and personnel physical strength, resulting in a small operating range and difficulty in quickly covering large areas. Manual inspections are easily affected by subjective factors, leading to low inspection effectiveness, and rely on paper records with low digitization. In contrast, robot inspections can adapt to various complex and dangerous scenarios, such as high temperature, high pressure, and low visibility, and have the ability to work continuously for extended periods. They have a wider inspection range and can enter narrow or dangerous areas. Furthermore, robots are equipped with high-precision sensors that can record inspection data in real time and automatically identify anomalies using AI algorithms. The data can be directly uploaded to the cloud for real-time access and management.

[0003] Currently, most common engineering inspection robots navigate based on completed buildings and their corresponding high-precision maps. However, the use of inspection robots in unfinished construction sites remains limited due to the lack of high-precision map scanning capabilities. Most current engineering inspection robots rely on remote control via camera images or geometric information provided by BIM (Building Information Modeling) for positioning and navigation. The drawback is that the 2D raster maps or 3D point cloud maps generated by BIM models are sparse maps with limited semantic information. While the key points provided by sparse maps can assist robot positioning, their semantic information does not support obstacle avoidance and navigation in complex environments. Furthermore, existing inspection robots cannot dynamically switch optimal path algorithms based on different scenarios. Current technologies mostly employ point cloud to BIM updates or BIM to point cloud updates, failing to achieve dynamic closed-loop management during construction. Especially in construction scenarios, due to dynamic environmental changes and structural incompleteness, existing technologies struggle to achieve accurate path planning, 3D modeling, and closed-loop management of quality monitoring, resulting in insufficient accuracy and timeliness in construction progress monitoring.

[0004] Therefore, this study proposes a construction inspection method based on dynamic path planning, multi-algorithm fusion processing, and a closed-loop verification mechanism using construction data and models. This method aims to enable rapid and accurate inspection of construction buildings, which is of great significance for ensuring the effectiveness of construction building inspections. Summary of the Invention

[0005] This invention provides a construction inspection method and apparatus to address the problem that existing inspection methods are insufficient in terms of accuracy and timeliness of construction progress monitoring due to the dynamic changes in the environment and incomplete structures in construction scenarios. This invention lays an important foundation for accurate construction inspection.

[0006] In a first aspect, the present invention provides a construction inspection method, comprising:

[0007] Obtain a two-dimensional plan view of the building under construction;

[0008] Based on the environmental type of different locations in the actual scene, and combined with the two-dimensional planar map, path planning is performed on the construction building to control the inspection robot to inspect the construction building and obtain the three-dimensional point cloud data and on-site structural image of the construction building.

[0009] The RANSAC and OPTICS clustering algorithms are used to fit the three-dimensional point cloud data based on a preset segmentation model to obtain the building information model of the construction building.

[0010] The on-site structural image is input into a preset quality monitoring network to calculate the construction status monitoring data of the building under construction.

[0011] Optionally, obtain a two-dimensional plan view of the building under construction;

[0012] Based on the environmental type of different locations in the actual scene, and combined with the two-dimensional planar map, path planning is performed on the construction building to control the inspection robot to inspect the construction building and obtain the three-dimensional point cloud data and on-site structural image of the construction building.

[0013] The RANSAC and OPTICS clustering algorithms are used to fit the three-dimensional point cloud data based on a preset segmentation model to obtain the building information model of the construction building.

[0014] The on-site structural image is input into a preset quality monitoring network to calculate the construction status monitoring data of the building under construction.

[0015] Based on the building information model and the construction status monitoring data, a construction progress inspection report for the construction building is generated.

[0016] Optionally, obtain a two-dimensional plan view of the building under construction, including:

[0017] Obtain 2D point cloud data of the building under construction, and obtain all 2D plan map templates of buildings in the BIM database;

[0018] Based on template matching algorithm, sliding window method and Hausdorff distance, the template with the highest similarity to the features of the two-dimensional point cloud data is extracted from all the two-dimensional planar map templates of the building to obtain the two-dimensional planar map of the building under construction.

[0019] Optionally, based on template matching algorithms, sliding window methods, and Hausdorff distance, the template with the highest similarity to the two-dimensional point cloud data features is extracted from all the aforementioned two-dimensional building plan map templates to obtain the two-dimensional plan map of the construction building, including:

[0020] Generate a planar structure diagram corresponding to the two-dimensional point cloud data;

[0021] Key feature data are extracted from the planar structure diagram;

[0022] Based on template matching algorithm, sliding window method and Hausdorff distance, the template with the highest similarity to the key feature data is extracted from all the two-dimensional plan map templates of the building to obtain the two-dimensional plan map of the building under construction.

[0023] Optionally, generating a planar structure diagram corresponding to the two-dimensional point cloud data includes:

[0024] The two-dimensional point cloud data is subjected to denoising and simplification processing to obtain the processed two-dimensional point cloud data;

[0025] Based on the processed two-dimensional point cloud data, a planar structure diagram is generated.

[0026] Optionally, based on template matching algorithms, sliding window methods, and Hausdorff distance, the template with the highest similarity to the key feature data features is extracted from all the aforementioned 2D building plan map templates to obtain the 2D plan map of the construction building, including:

[0027] Using template matching algorithm and sliding window method, based on Hausdorff distance, the feature similarity between the key feature data and the corresponding positions in all the building two-dimensional planar map templates is calculated.

[0028] The maximum value is extracted from the feature similarity;

[0029] The building 2D plan map template corresponding to the maximum value is determined as the 2D plan map of the building under construction.

[0030] Optionally, based on the environmental type at different locations in the actual scene, and in conjunction with the two-dimensional planar map, path planning is performed on the construction building to control the inspection robot to inspect the construction building, obtaining three-dimensional point cloud data and on-site structural images of the construction building, including:

[0031] The inspection robot is guided to perform inspections using the two-dimensional planar map.

[0032] During the inspection process, based on the environmental type of different locations in the actual scene, a dynamic path planning method is used to generate the inspection path of the inspection robot at the corresponding location.

[0033] Based on the inspection path, the inspection robot is controlled to inspect the construction building, thereby obtaining the three-dimensional point cloud data and on-site structural images of the construction building.

[0034] Optionally, the RANSAC and OPTICS clustering algorithms are used to fit the 3D point cloud data based on a preset segmentation model to obtain the building information model of the construction structure, including:

[0035] The three-dimensional point cloud data is preprocessed to obtain preprocessed three-dimensional point cloud data;

[0036] The preprocessed 3D point cloud data is input into a preset segmentation model to obtain point cloud data for different regions;

[0037] The RANSAC and OPTICS clustering algorithms were used to fit the point cloud data of the different regions to obtain the building information model of the construction building.

[0038] Optionally, the three-dimensional point cloud data is preprocessed to obtain preprocessed three-dimensional point cloud data, including:

[0039] The 3D point cloud data is denoised to obtain denoised 3D point cloud data.

[0040] The denoised 3D point cloud data is downsampled to obtain preprocessed 3D point cloud data.

[0041] Optionally, the RANSAC algorithm and the OPTICS clustering algorithm are used to fit the point cloud data of the different regions to obtain the building information model of the construction building, including:

[0042] The RANSAC algorithm is used to fit the wall point cloud and column point cloud in the point cloud data of the different regions to obtain the planar wall model and column wall model.

[0043] The OPTICS clustering algorithm is used to fit other point cloud data in the point cloud data of the different regions to obtain the component models and model labels corresponding to the other point cloud data.

[0044] Based on the model tags and corresponding component models, the planar wall model, and the column wall model, a building information model of the construction building is generated.

[0045] Secondly, the present invention provides a construction inspection device, comprising:

[0046] The acquisition module is used to acquire a two-dimensional plan view of the construction site;

[0047] The inspection module is used to plan the path of the construction building according to the environmental type of different locations in the actual scene and in combination with the two-dimensional planar map, so as to control the inspection robot to inspect the construction building and obtain the three-dimensional point cloud data and on-site structural image of the construction building.

[0048] The fitting module is used to fit the three-dimensional point cloud data based on a preset segmentation model using the RANSAC algorithm and the OPTICS clustering algorithm to obtain the building information model of the construction building.

[0049] The calculation module is used to input the on-site structural image into a preset quality monitoring network and calculate the construction status monitoring data of the construction building.

[0050] The generation module is used to generate a construction progress inspection report for the building under construction based on the building information model and the construction status monitoring data.

[0051] This invention provides a method and apparatus for inspecting construction buildings. The method includes: acquiring a two-dimensional planar map of the construction building; performing path planning for the construction building based on the environmental types of different locations in the actual scene and in conjunction with the two-dimensional planar map; controlling an inspection robot to inspect the construction building; obtaining three-dimensional point cloud data and on-site structural images of the construction building; using the RANSAC algorithm and the OPTICS clustering algorithm, and based on a preset segmentation model, fitting the three-dimensional point cloud data to obtain a building information model of the construction building; and inputting the on-site structural images into a preset quality monitoring network to calculate the construction status monitoring data of the construction building. This invention forms a method for inspecting construction buildings that can utilize dynamic path planning, multi-algorithm fusion processing, and a closed-loop verification mechanism of construction data and models. It solves the problem that existing inspection methods are difficult to achieve accurate path planning, three-dimensional modeling, and quality monitoring due to the dynamic changes in the environment and incomplete structures in construction building scenes, resulting in insufficient accuracy and timeliness of construction progress monitoring. This lays an important foundation for accurate inspection of construction buildings. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart illustrating the steps of a construction inspection method according to the present invention.

[0054] Figure 2 This is a flowchart illustrating the steps of a second embodiment of the construction inspection method of the present invention.

[0055] Figure 3 This is a structural block diagram of an embodiment of a construction inspection device according to the present invention. Detailed Implementation

[0056] This invention provides a construction inspection method and apparatus to address the problem that existing inspection methods are insufficient in accuracy and timeliness of construction progress monitoring due to the dynamic changes in the environment and incomplete structures in construction scenarios. This lays an important foundation for accurate construction inspection.

[0057] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0058] Example 1

[0059] Please see Figure 1 , Figure 1 This is a flowchart illustrating the steps of a construction inspection method according to an embodiment of the present invention. The method includes:

[0060] Step S101: Obtain a two-dimensional plan map of the building under construction;

[0061] In one optional embodiment, obtaining a two-dimensional plan view of the construction site includes:

[0062] Acquire 2D point cloud data of the building under construction, and acquire all 2D plan map templates of buildings in the BIM database (Building Information Modeling database);

[0063] Based on template matching algorithm, sliding window method and Hausdorff distance, the template with the highest similarity to the features of the two-dimensional point cloud data is extracted from all the two-dimensional planar map templates of the building to obtain the two-dimensional planar map of the building under construction.

[0064] In this embodiment of the invention, two-dimensional point cloud data of the construction building and all two-dimensional planar map templates of buildings in the BIM database are acquired. The two-dimensional point cloud data is denoised and simplified to obtain processed two-dimensional point cloud data. Based on the processed two-dimensional point cloud data, a planar structure diagram is generated. Key feature data is extracted from the planar structure diagram. Using a template matching algorithm and a sliding window method, based on Hausdorff distance, the feature similarity between the key feature data and the corresponding positions in all two-dimensional planar map templates of buildings is calculated. The maximum value among the feature similarities is extracted, and the two-dimensional planar map template of the building corresponding to the maximum value is determined as the two-dimensional planar map of the construction building.

[0065] Step S102: Based on the environmental type of different locations in the actual scene, and combined with the two-dimensional planar map, a path is planned for the construction building to control the inspection robot to inspect the construction building and obtain the three-dimensional point cloud data and on-site structural image of the construction building.

[0066] This invention uses a two-dimensional planar map to guide an inspection robot to perform inspections. During the inspection process, a dynamic path planning method is used to generate an inspection path for the robot at the corresponding location based on the environmental type of different locations in the actual scene. Based on the inspection path, the robot is controlled to inspect the construction building and obtain the three-dimensional point cloud data and on-site structural image of the construction building.

[0067] Step S103: Using the RANSAC algorithm (Random Sampling Consensus Algorithm) and the OPTICS clustering algorithm (Density-based clustering algorithm), and based on a preset segmentation model, the three-dimensional point cloud data is fitted to obtain the building information model of the construction building.

[0068] In an optional embodiment, the RANSAC algorithm and the OPTICS clustering algorithm are used to fit the three-dimensional point cloud data based on a preset segmentation model to obtain the building information model of the construction building, including:

[0069] The three-dimensional point cloud data is preprocessed to obtain preprocessed three-dimensional point cloud data;

[0070] The preprocessed 3D point cloud data is input into a preset segmentation model to obtain point cloud data for different regions;

[0071] The RANSAC and OPTICS clustering algorithms were used to fit the point cloud data of the different regions to obtain the building information model of the construction building.

[0072] This invention denoises 3D point cloud data to obtain denoised 3D point cloud data. It then downsamples the denoised 3D point cloud data to obtain preprocessed 3D point cloud data. The preprocessed 3D point cloud data is input into a preset segmentation model to obtain point cloud data for different regions. The RANSAC algorithm is used to fit wall and column point clouds in the point cloud data of different regions to obtain planar wall and column wall models. The OPTICS clustering algorithm is used to fit other point cloud data in the point cloud data of different regions to obtain component models and model labels corresponding to the other point cloud data. Based on the model labels and the corresponding component models, planar wall models, and column wall models, a building information model of the construction building is generated.

[0073] Step S104: Input the on-site structural image into a preset quality monitoring network to calculate the construction status monitoring data of the building under construction;

[0074] In this embodiment of the invention, the on-site structural image is input into a preset quality monitoring network, and construction defects or progress deviations are detected by image recognition technology to obtain construction status monitoring data of the construction building. The construction status monitoring data includes the current original image and defect detection target boxes (i.e., marking defects in the on-site structural image).

[0075] Step S105: Based on the building information model and the construction status monitoring data, generate a construction progress inspection report for the construction building;

[0076] This invention integrates building information modeling and construction monitoring data to generate a construction progress inspection report, achieving a two-dimensional closed-loop verification of construction progress and quality.

[0077] The construction building inspection method provided by this invention forms a set of methods for inspecting construction buildings using dynamic path planning, multi-algorithm fusion processing, and a closed-loop verification mechanism of construction data and models. It solves the problem that existing inspection methods are difficult to achieve accurate path planning, 3D modeling, and quality monitoring due to the dynamic changes in the environment and incomplete structures in the construction building scene, resulting in insufficient accuracy and timeliness of construction progress monitoring. This lays an important foundation for accurate inspection of construction buildings.

[0078] Example 2

[0079] Please see Figure 2 , Figure 2 This is a flowchart illustrating the steps of a second embodiment of a construction inspection method according to the present invention. The steps include:

[0080] S201, acquire the two-dimensional point cloud data of the building under construction, and acquire all two-dimensional plan map templates of buildings in the BIM database;

[0081] This invention uses LiDAR to collect two-dimensional point cloud data of the construction site, obtain the planar outline information of the building structure in real time, and retrieve all pre-stored two-dimensional planar map templates of the building from the BIM database. These templates contain standardized layout information of components such as walls, doors and windows.

[0082] S202, Based on template matching algorithm, sliding window method and Hausdorff distance, extract the template with the highest similarity to the features of the two-dimensional point cloud data from all the two-dimensional planar map templates of the building to obtain the two-dimensional planar map of the construction building;

[0083] In an optional embodiment, based on template matching algorithms, sliding window methods, and Hausdorff distance, the template with the highest similarity to the two-dimensional point cloud data features is extracted from all the building two-dimensional plan map templates to obtain the two-dimensional plan map of the construction building, including:

[0084] Generate a planar structure diagram corresponding to the two-dimensional point cloud data;

[0085] Key feature data are extracted from the planar structure diagram;

[0086] Based on template matching algorithm, sliding window method and Hausdorff distance, the template with the highest similarity to the key feature data is extracted from all the two-dimensional plan map templates of the building to obtain the two-dimensional plan map of the building under construction.

[0087] The original two-dimensional point cloud data acquired in the construction scenario in this invention typically includes vibration noise caused by equipment vibration and redundant points generated by repeated scanning. First, a statistical filtering algorithm is used to calculate the average distance between each point and its neighboring points. When the average distance between a point and its neighbors exceeds a preset threshold, it is identified as a noise point and removed. Then, a non-uniform sampling algorithm is used to simplify the denoised point cloud. A higher sampling rate is set at wall corners to preserve geometric details, while the sampling rate is reduced in open areas to decrease the data volume, resulting in processed two-dimensional point cloud data. Finally, using existing technologies, such as surveying software (e.g., CASS software), the processed two-dimensional point cloud data is transformed to form a planar structural map with clearly defined geometric boundaries.

[0088] Key feature data, such as the location and size of walls, doors, and windows, can be extracted from a planar structural diagram using existing image recognition technology.

[0089] A template matching algorithm is used to initially screen 2D building plan map templates. A sliding window method is then used to perform multi-scale scanning of key feature areas, effectively capturing potential local structural deformation features in the construction site. The geometric similarity between each candidate template and the measured key feature data is calculated based on Hausdorff distance. By comparing the similarity values ​​of all templates, the template with the highest similarity value is selected as the optimal matching result, which is then determined as the 2D plan map of the construction site.

[0090] S203, guide the inspection robot to perform inspections using the two-dimensional planar map;

[0091] This invention uses a two-dimensional planar map as a global navigation framework to provide initial positioning and path guidance for the inspection robot.

[0092] The inspection system comprises an inspection robot and its corresponding command platform. The inspection robot, acting as the main inspection unit, performs inspections within the construction site, while the command platform provides remote monitoring and develops inspection plans. The inspection robot consists of three modules: a perception module, a control module, and an execution module. Specifically, the perception module includes a LiDAR, an RGB-D camera, a panoramic camera, environmental data sensors, and a voice module, as well as an inertial measurement unit for robot localization and navigation. The control module is composed of an embedded computing platform or an industrial control unit with a graphics processing unit, equipped with various global and local path planning algorithms, navigation and decision-making algorithms, and an image processing computing unit. It also includes a communication interface for data transmission with the command platform. The execution module includes a drive system consisting of serial bus servos and mechanical legs. The command platform comprises a front-end display interface and a back-end data processing interface. The front-end display interface includes the robot's camera view, a 2D grid map and a 3D point cloud map, inspection plan selection, robot motion parameters and sensor parameters, and manual / automatic motion control switching. The back-end data processing includes sensor data storage and retrieval, automatic inspection report generation, and BIM information processing.

[0093] After the inspection task begins, an inspection robot is placed at any location on the ground within the inspected building. The robot activates its LiDAR scanner, which scans 360 degrees to obtain environmental data within a 20-meter range. The robot's pose, originally described in the chassis coordinate system, is converted to the initial pose in the LiDAR coordinate system to ensure that the environmental information observed by the radar is consistent with the robot's movement and localization. Then, the robot's ROS map coordinates are converted to OpenCV coordinates to provide position information and orientation angle. During the inspection robot's movement, OpenCV image operations are used to handle map matching. The 2D planar map and the robot's scanned map are combined through translation and rotation to obtain the transformation combination with the highest matching degree. Based on the matching result, the radar position and angle information are updated to help the robot achieve precise localization during movement.

[0094] S204, During the inspection process, based on the environmental type of different locations in the actual scene, a dynamic path planning method is used to generate the inspection path of the inspection robot at the corresponding location;

[0095] In this embodiment of the invention, environment type refers to the spatial characteristic classification of different areas in the construction scene, which can be achieved by identifying parameters such as channel width and obstacle density through LiDAR or visual sensors. During the inspection process, the real-time collected environment type data triggers dynamic switching of the path planning algorithm. For example, when a narrow channel is detected, one path planning method is used to extract the boundaries of walls and obstacles to avoid collisions. When entering an open area, another path planning method is used to identify large open areas. This segmented path planning method, through the linkage between the environmental perception module and the algorithm library, ensures that the robot can both avoid temporary obstacles in unfinished areas and quickly cover the planned area. Through the preset mapping relationship between the path planning algorithm and scene features, the algorithm selection is dynamically adapted to the construction progress, ultimately forming a complete inspection path and corresponding control commands.

[0096] Specifically, the inspection robot identifies environmental features using sensors such as LiDAR and RGB-D cameras. It extracts wall and obstacle boundaries using clustering algorithms (e.g., DBSCAN), calculates path curvature (higher curvature indicates a more complex environment), and classifies the environment type based on deep learning models (e.g., ResNet). It also determines spatial openness by combining depth information. During inspection, the robot determines the environment type at different locations based on point cloud density and on-site structural images. The final environment types are categorized into four types: long straight corridors, open spaces, winding and narrow passages, and dynamic obstacle environments. Based on the environment type at different locations in the actual scene, different path planning methods (i.e., dynamic path planning methods) are used to generate the inspection path for the robot at the corresponding location.

[0097] Specifically, when the lidar detects parallel long walls on both sides, with a uniform path width and no obvious obstacles, the corresponding area is determined to be a long, straight corridor; when the lidar detects low point cloud density and the RGB-D camera identifies the corresponding site structure image as a large open area, the corresponding area is determined to be an open space; when the lidar detects multiple short walls or dense obstacles, with frequent changes in path direction, the corresponding area is determined to be a winding, narrow passage; and when the RGB-D camera identifies moving personnel or temporarily piled materials in the corresponding site structure image, the corresponding area is determined to be a dynamic obstacle environment.

[0098] For long, straight corridor environments, the A* algorithm is used for path planning. In long, straight corridor environments during construction (such as stairwells and utility tunnels in unfinished buildings), the inspection robot needs to quickly plan a globally optimal path while avoiding collisions with walls and temporary material piles. Due to its heuristic search characteristics, the A* algorithm is the optimal choice for such structured environments. The specific steps are as follows: First, extract the IFC file (such as the IfcWall entity) from the BIM database to generate a two-dimensional grid map. The map is divided into three types of pixels: black, white, and gray. Black pixels represent walls / fixed obstacles (impassable); white pixels represent passable areas; and gray pixels represent temporary obstacles (such as stacked building materials, which need to be dynamically updated). Then, real-time scanning is performed using LiDAR. The ICP algorithm is used to align the current scanned point cloud with the BIM map to eliminate positioning errors. Newly added point clouds are clustered (e.g., DBSCAN) and marked as temporary obstacles (gray grid). The following describes the execution process of the A* algorithm. First, the relevant parameters are initialized, and the Manhattan distance is chosen as the heuristic function, suitable for a grid environment. Then, path searching is performed, with the movement cost represented by 1 for a straight line and 1 for a diagonal. It indicates that robots are encouraged to prioritize straight lines; the safety weight is set to 1.5 in the early stages of construction and 1.0 in the later stages, increasing the cost in the early stages to move away from unfixed walls, and ultimately outputting a safe movement path.

[0099] For open space environments, the RRT* algorithm (Rapid Extended Random Tree algorithm) is used for path planning. In open space environments, such as unfinished building lobbies or open-air work areas, inspection robots need to efficiently explore areas lacking structured features and avoid temporary obstacles (such as material piles and equipment). The RRT* algorithm, with its random sampling and asymptotic optimality, becomes the optimal choice for such unstructured environments. The specific steps are as follows: First, the point cloud obtained from the LiDAR scan is processed. The LiDAR scan generates a 3D point cloud, which is then filtered by voxel lattice filtering and statistical filtering to remove noise and extract passable areas. Next, data fusion is performed using semantic information from the RGB-D camera. The RGB image is input into an improved U-Net model to segment the ground, temporary obstacles (marked in red), and equipment (marked in blue). The depth map is converted into a 2.5D height map, and potential danger areas are marked. Finally, a free space map is constructed, and a binary raster map is output, with dynamic obstacles marked as temporary obstacles. The following is the execution flow of the RRT* algorithm: First, initialize the parameter step size and target bias probability (probabilities greater than 30% accelerate convergence). Initialize the rewiring radius (the neighborhood radius considered when optimizing the path) and the maximum number of iterations. Then, continue to expand the environment with random numbers, remove redundant nodes through rewiring, perform path planning, and finally output the planned path.

[0100] For winding and narrow passage environments, the *D Lite algorithm (Dynamic Incremental Path Planning algorithm) is used for path planning. In such environments, the inspection robot needs to avoid obstacles in real time within a limited space and deal with sudden obstacles, such as temporary material stacks and mobile equipment. The *D Lite algorithm, with its reverse search and incremental update characteristics, is the optimal choice for this type of dynamic and narrow environment. The specific steps are as follows: First, the structural information of the narrow passage is extracted from the BIM database to generate a high-resolution grid map. The map is divided into three types of pixels: black, white, and purple. Black pixels represent walls / fixed obstacles (impassable); white pixels represent passable areas; and purple pixels represent dynamic obstacles (marked in real time by LiDAR). Then, dynamic obstacle detection is performed through real-time scanning with LiDAR, and new obstacles are marked based on the differences in point clouds between adjacent frames. Finally, the robot pose is updated in real time by fusing IMU and wheel encoder data. In narrow passages, tightly coupled SLAM (such as the LOAM algorithm) is enabled to avoid cumulative errors. The execution flow of the *D Lite algorithm is as follows: initialization of relevant parameters, heuristic weights, replanning frequency, and safety distance. Then, sudden obstacle detection is performed, and finally, path replanning is achieved.

[0101] For dynamic obstacle environments, the APF algorithm (Artificial Potential Field Method) is used for path planning.

[0102] In dynamic obstacle environments during construction (such as those with moving personnel, transport vehicles, or temporary material stockpiles), inspection robots need to perceive and avoid sudden obstacles in real time while maintaining path smoothness and safety. The Artificial Potential Field (APF) method simulates attractive and repulsive forces using a virtual potential field to achieve rapid response and real-time obstacle avoidance. The specific steps are as follows: First, dynamic obstacles are detected. Using an RGB-D camera, based on a YOLOv8 model, personnel and vehicles are detected in real time. Depth maps are used to calculate obstacle distances and mark movement speeds. LiDAR detects movement trajectories and updates obstacle positions through adjacent frame point cloud registration (ICP algorithm). Then, an environmental map is constructed. Fixed structures (walls, equipment) are extracted from the BIM model as the basis for the repulsive force field, constructing a static obstacle map. Moving obstacles are marked with timestamp grids, and their position and velocity vectors are updated in real time to construct dynamic obstacle markers. Finally, target points are set to achieve path planning. The following is the execution flow of the APF algorithm:

[0103] (1) Potential field modeling formulas, including attractive force field modeling formulas and repulsive force field modeling formulas.

[0104] The specific formula for modeling the attraction field is as follows:

[0105] ;

[0106] in, For attraction, As the attractiveness coefficient, The distance between the inspection robot and the target point is Euclidean.

[0107] The specific formula for modeling a repulsive force field is as follows:

[0108] ;

[0109] in, It is a repulsive force. This is the repulsive force coefficient, which is temporarily increased when a high-speed obstacle is detected. d represents the shortest distance between the inspection robot and the obstacle, and d0 represents the radius of influence of the obstacle.

[0110] (2) Calculation of resultant force and motion control.

[0111] The formula for the total potential field is as follows:

[0112] ;

[0113] in, The total potential force.

[0114] Motion direction decision:

[0115] ;

[0116] in, The direction of movement for the inspection robot. Let represent the component of the total potential force along the y-axis in Cartesian coordinates. Let x be the component of the total potential force along the x-axis in Cartesian coordinates. The inspection robot moves towards... Direction of movement, speed and The amplitude is directly proportional.

[0117] In the initial stage, the robot moves towards the target along a straight line. When a dynamic environment is detected, the repulsive force is calculated according to the potential field modeling formula. The APF updates the direction of the repulsive force in real time. After the obstacle is removed, the attractive force takes over, the path returns to a straight line, and the final planned path is obtained.

[0118] S205, Based on the inspection path, control the inspection robot to inspect the construction building and obtain the three-dimensional point cloud data and on-site structural image of the construction building;

[0119] According to the inspection path generated by the above method, the present invention controls the inspection robot to perform multiple inspections on the construction building, so as to obtain the three-dimensional point cloud data and on-site structural images of the construction building through sensors such as LiDAR and RGB-D cameras.

[0120] Specifically, when the inspection robot is deployed to a building under construction, a 3D laser scanner scans and acquires a 3D point cloud of the building within its current detection range. This point cloud is then added to a matching group. Simultaneously, the top-view density map of the 3D point cloud is processed and uploaded as a 2D planar map of the building to the inspection command platform as the current 2D map display. (Because the density of the walls formed by the 3D point cloud is significantly higher than that of other spaces indoors in the top view, a 2D grid map of the current building is generated based on the density; low-density areas are blank grids, and high-density areas are occupied grids.) A set of state parameters is then used to represent the robot's position and orientation. Each parameter has a weight, indicating the degree of matching between the parameter and the observed data. Based on data acquired by the robot's sensors (such as LiDAR, IMU, etc.), the weight of each parameter is updated, and the parameters are resampled according to their weights. High-weight parameters are copied, while low-weight parameters may be discarded. This method enables accessibility and safety analysis of the next target location, selecting the most suitable next target location. Upon arrival at the new target location, the LiDAR continues to scan the 3D point cloud of the current building and adds it to the same matching group. The final aligned 3D point cloud map is uploaded to the inspection command platform as the current inspection 3D map display. Simultaneously, the 3D point cloud map is stored in a database for subsequent 3D BIM generation.

[0121] S206, preprocess the three-dimensional point cloud data to obtain preprocessed three-dimensional point cloud data;

[0122] In one optional embodiment, the three-dimensional point cloud data is preprocessed to obtain preprocessed three-dimensional point cloud data, including:

[0123] The 3D point cloud data is denoised to obtain denoised 3D point cloud data.

[0124] The denoised 3D point cloud data is downsampled to obtain preprocessed 3D point cloud data.

[0125] In this embodiment of the invention, noise points may be introduced into the 3D point cloud data during the acquisition process due to equipment accuracy limitations or dynamic environmental factors, such as outliers caused by sensor jitter or artifacts formed by dust interference. Denoising processing, such as using statistical filtering or radius filtering methods, can effectively identify and remove such abnormal data, restoring the true geometric structure of the point cloud and obtaining denoised 3D point cloud data. Subsequently, to address the redundancy problem of high-density point cloud data, downsampling processing, through spatial partitioning or random sampling, retains sufficient point cloud density in key feature regions while reducing the amount of data in non-critical regions, resulting in preprocessed 3D point cloud data. For example, the voxel mesh method can be used to divide the point cloud into uniform cubic units, retaining only the center point or centroid point within each unit, thereby significantly reducing the data size while maintaining the accuracy of the building structure outline.

[0126] S207, The preprocessed 3D point cloud data is input into a preset segmentation model to obtain point cloud data for different regions;

[0127] In this embodiment of the invention, the preset segmentation model is an improved U-Net semantic segmentation model. The input point cloud (preprocessed 3D point cloud data) is classified and identified as walls, beams, columns and pipelines through the trained model weight file, and the output point cloud subset with component category labels is obtained to obtain point cloud data of different regions.

[0128] The improved U-Net semantic segmentation model consists of four modules: an input module, an encoding layer module, a decoding layer module, and an output module. The input receives a point cloud dataset, where each point cloud is N×d in size. N represents the number of points, and d is the feature dimension. The feature dimension mainly includes point cloud coordinates (x, y, z) and color values ​​(r, g, b), timestamps, and reflection intensity. Compared to traditional U-Net-based image segmentation network models, this model considers the spatial location and color channels of indoor architectural 3D point clouds. In each encoding layer, it innovatively uses a Dual Attention Feature Aggregation Module (DAFA) and a Random Sampling Method (RS). DAFA focuses on aggregating the spatial location (x, y, z) and different channels (r, g, b) of the feature map, weighting and fusing the location and channels to retain more local information of the current point. After five encoding layers, the feature dimension gradually increases, extracting more complete and accurate architectural point cloud information. The decoding layer module also consists of five layers, combining feature extraction through upsampling and fully connected layers to expand the number of point clouds and reduce the feature dimension. Features from the encoding layer are combined through skip connections. The output layer consists of the last two fully connected layers (FC), and after multiple layers of processing, it outputs the semantic classification result for each point, performing point-by-point semantic label determination.

[0129] S208. The RANSAC algorithm and OPTICS clustering algorithm are used to fit the point cloud data of the different regions to obtain the building information model of the construction building.

[0130] In an optional embodiment, the RANSAC algorithm and the OPTICS clustering algorithm are used to fit the point cloud data of the different regions to obtain the building information model of the construction building, including:

[0131] The RANSAC algorithm is used to fit the wall point cloud and column point cloud in the point cloud data of the different regions to obtain the planar wall model and column wall model.

[0132] The OPTICS clustering algorithm is used to fit other point cloud data in the point cloud data of the different regions to obtain the component models and model labels corresponding to the other point cloud data.

[0133] Based on the model tags and corresponding component models, the planar wall model, and the column wall model, a building information model of the construction building is generated.

[0134] In this embodiment of the invention, the segmented point cloud data (point cloud data from different regions) is classified to address the differences in morphological characteristics of unfinished building structures in a construction scenario. For wall and column structures with obvious planar features, the RANSAC algorithm is used to iteratively fit the plane equation parameters, and its strong noise resistance is used to extract linear geometric models. For planar walls and column walls, planar models and cylindrical models are selected respectively to obtain planar wall models and column wall models. For irregular components such as pipes and equipment, the OPTICS algorithm is used to perform hierarchical clustering based on the point cloud density distribution characteristics, and independent components are identified by setting neighborhood radii and minimum point count thresholds. In the model fusion stage, the parametric geometric model and the labeled clustering model are spatially registered, and construction information such as component type and installation progress is written into the model attribute fields. Using existing technologies (such as importing the above information into Revit), a building information model containing complete semantic information is formed.

[0135] S209, Input the on-site structural image into a preset quality monitoring network to calculate the construction status monitoring data of the building under construction;

[0136] In this embodiment of the invention, the quality monitoring network is a YOLOv8 model with pre-trained weight parameters. The inspection robot uses camera vision perception and deep learning algorithms to input images of the current building structure and output construction status monitoring data. The construction status monitoring data includes the current original image and defect detection target boxes.

[0137] In an alternative embodiment, construction status monitoring data can be combined with a two-dimensional planar map to generate a flatness heat map of the corresponding building structure, thereby enabling rapid flatness measurement.

[0138] S210, Based on the building information model and the construction status monitoring data, generate a construction progress inspection report for the construction building;

[0139] In this embodiment of the invention, construction status monitoring data is associated with component numbers in the building information model, and a progress report with specific problem locations is generated.

[0140] This invention discloses a construction inspection method that solves the problem that existing inspection methods are difficult to implement for accurate path planning, 3D modeling, and quality monitoring due to the dynamic changes in the environment and incomplete structures in construction scenarios, resulting in insufficient accuracy and timeliness of construction progress monitoring. This method lays an important foundation for accurate construction inspection.

[0141] Example 3

[0142] Please see Figure 3 , Figure 3 This is a structural block diagram of an embodiment of a construction inspection device according to the present invention. The device includes:

[0143] Module 301 is used to acquire a two-dimensional plan map of the construction site;

[0144] The inspection module 302 is used to plan the path of the construction building according to the environmental type of different locations in the actual scene and in combination with the two-dimensional planar map, so as to control the inspection robot to inspect the construction building and obtain the three-dimensional point cloud data and on-site structural image of the construction building.

[0145] The fitting module 303 is used to fit the three-dimensional point cloud data based on a preset segmentation model using the RANSAC algorithm and the OPTICS clustering algorithm to obtain the building information model of the construction building.

[0146] The calculation module 304 is used to input the on-site structural image into a preset quality monitoring network and calculate the construction status monitoring data of the construction building.

[0147] The generation module 305 is used to input the on-site structural image into a preset quality monitoring network and calculate the construction status monitoring data of the construction building.

[0148] In an optional embodiment, the acquisition module 301 includes:

[0149] The acquisition submodule is used to acquire 2D point cloud data of the building under construction, as well as all 2D plan map templates of buildings in the BIM database;

[0150] The extraction submodule is used to extract the template with the highest similarity to the features of the two-dimensional point cloud data from all the two-dimensional planar map templates of the building based on the template matching algorithm, sliding window method and Hausdorff distance, so as to obtain the two-dimensional planar map of the construction building.

[0151] In an optional embodiment, the extraction submodule includes:

[0152] A generation unit is used to generate a planar structure diagram corresponding to the two-dimensional point cloud data;

[0153] The first extraction unit is used to extract key feature data from the planar structure diagram;

[0154] The second extraction unit is used to extract the template with the highest similarity to the key feature data from all the two-dimensional planar map templates of the building based on the template matching algorithm, the sliding window method and Hausdorff distance, so as to obtain the two-dimensional planar map of the construction building.

[0155] In an optional embodiment, the generation unit includes:

[0156] The processing subunit is used to perform noise reduction and simplification processing on the two-dimensional point cloud data to obtain the processed two-dimensional point cloud data.

[0157] A sub-unit is generated to generate a planar structure diagram based on the processed two-dimensional point cloud data.

[0158] In an optional embodiment, the second extraction unit includes:

[0159] The calculation subunit is used to calculate the feature similarity between the key feature data and the corresponding positions in all the two-dimensional building map templates based on Hausdorff distance using a template matching algorithm and a sliding window method.

[0160] Extract sub-units to extract the maximum value from the feature similarity;

[0161] A sub-unit is defined to determine the two-dimensional plan map template of the building corresponding to the maximum value as the two-dimensional plan map of the building under construction.

[0162] In an optional embodiment, the inspection module 302 includes:

[0163] The guidance submodule is used to guide the inspection robot to perform inspections using the two-dimensional planar map.

[0164] The generation submodule is used to generate the inspection path of the inspection robot at the corresponding location based on the environmental type of different locations in the actual scene during the inspection process, using a dynamic path planning method.

[0165] The control submodule is used to control the inspection robot to inspect the construction building based on the inspection path, and obtain the three-dimensional point cloud data and on-site structural image of the construction building.

[0166] In an optional embodiment, the fitting module 303 includes:

[0167] The preprocessing submodule is used to preprocess the three-dimensional point cloud data to obtain preprocessed three-dimensional point cloud data.

[0168] The input submodule is used to input the preprocessed 3D point cloud data into a preset segmentation model to obtain point cloud data for different regions;

[0169] The fitting submodule is used to fit the point cloud data of the different regions using the RANSAC algorithm and the OPTICS clustering algorithm to obtain the building information model of the construction building.

[0170] In an optional embodiment, the preprocessing submodule includes:

[0171] A denoising unit is used to denoise the three-dimensional point cloud data to obtain denoised three-dimensional point cloud data.

[0172] The downsampling unit is used to downsample the denoised 3D point cloud data to obtain preprocessed 3D point cloud data.

[0173] In an optional embodiment, the fitting submodule includes:

[0174] The first fitting unit is used to fit the wall point cloud and column point cloud in the point cloud data of the different regions using the RANSAC algorithm to obtain the planar wall model and the column wall model.

[0175] The second fitting unit is used to fit other point cloud data in the point cloud data of the different regions using the OPTICS clustering algorithm to obtain the component model and model label corresponding to the other point cloud data.

[0176] The modeling unit is used to generate the building information model of the construction building based on the model tags and corresponding component models, the planar wall model and the column wall model.

[0177] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0178] In the several embodiments provided in this application, it should be understood that the methods and apparatus disclosed in this invention can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between devices or units, and may be electrical, mechanical, or other forms.

[0179] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0180] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0181] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned readable storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0182] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A construction inspection method, characterized in that, include: Obtain a two-dimensional plan view of the building under construction; Based on the environmental types at different locations in the actual scene, and combined with the two-dimensional planar map, path planning is performed on the construction building to control the inspection robot to inspect the construction building and obtain the three-dimensional point cloud data and on-site structural image of the construction building. This includes: guiding the inspection robot to perform inspections using the two-dimensional planar map; during the inspection process, using a dynamic path planning method to generate the inspection path for the inspection robot at the corresponding location based on the environmental types at different locations in the actual scene; and controlling the inspection robot to inspect the construction building based on the inspection path to obtain the three-dimensional point cloud data and on-site structural image of the construction building. The RANSAC and OPTICS clustering algorithms are used to fit the three-dimensional point cloud data based on a preset segmentation model to obtain the building information model of the construction building. The on-site structural image is input into a preset quality monitoring network to calculate the construction status monitoring data of the building under construction. Based on the building information model and the construction status monitoring data, a construction progress inspection report for the construction building is generated.

2. The construction inspection method according to claim 1, characterized in that, Obtain a two-dimensional plan view of the building under construction, including: Obtain 2D point cloud data of the building under construction, and obtain all 2D plan map templates of buildings in the BIM database; Based on template matching algorithm, sliding window method and Hausdorff distance, the template with the highest similarity to the features of the two-dimensional point cloud data is extracted from all the two-dimensional planar map templates of the building to obtain the two-dimensional planar map of the building under construction.

3. The construction inspection method according to claim 2, characterized in that, Based on template matching algorithms, sliding window methods, and Hausdorff distance, the template with the highest similarity to the two-dimensional point cloud data features is extracted from all the aforementioned two-dimensional building plan map templates to obtain the two-dimensional plan map of the construction building, including: Generate a planar structure diagram corresponding to the two-dimensional point cloud data; Key feature data are extracted from the planar structure diagram; Based on template matching algorithm, sliding window method and Hausdorff distance, the template with the highest similarity to the key feature data is extracted from all the two-dimensional plan map templates of the building to obtain the two-dimensional plan map of the building under construction.

4. The construction inspection method according to claim 3, characterized in that, Generating a planar structure diagram corresponding to the two-dimensional point cloud data includes: The two-dimensional point cloud data is subjected to denoising and simplification processing to obtain the processed two-dimensional point cloud data; Based on the processed two-dimensional point cloud data, a planar structure diagram is generated.

5. The construction inspection method according to claim 3, characterized in that, Based on template matching algorithms, sliding window methods, and Hausdorff distance, the template with the highest similarity to the key feature data is extracted from all the aforementioned 2D building plan map templates to obtain the 2D plan map of the construction building, including: Using template matching algorithm and sliding window method, based on Hausdorff distance, the feature similarity between the key feature data and the corresponding positions in all the building two-dimensional planar map templates is calculated. The maximum value is extracted from the feature similarity; The building 2D plan map template corresponding to the maximum value is determined as the 2D plan map of the building under construction.

6. The construction inspection method according to claim 1, characterized in that, Using the RANSAC and OPTICS clustering algorithms, and based on a preset segmentation model, the 3D point cloud data is fitted to obtain the building information model of the construction structure, including: The three-dimensional point cloud data is preprocessed to obtain preprocessed three-dimensional point cloud data; The preprocessed 3D point cloud data is input into a preset segmentation model to obtain point cloud data for different regions; The RANSAC and OPTICS clustering algorithms were used to fit the point cloud data of the different regions to obtain the building information model of the construction building.

7. The construction inspection method according to claim 6, characterized in that, The three-dimensional point cloud data is preprocessed to obtain preprocessed three-dimensional point cloud data, including: The 3D point cloud data is denoised to obtain denoised 3D point cloud data. The denoised 3D point cloud data is downsampled to obtain preprocessed 3D point cloud data.

8. The construction inspection method according to claim 6, characterized in that, The RANSAC and OPTICS clustering algorithms are used to fit the point cloud data of the different regions to obtain the building information model of the construction building, including: The RANSAC algorithm is used to fit the wall point cloud and column point cloud in the point cloud data of the different regions to obtain the planar wall model and column wall model. The OPTICS clustering algorithm is used to fit other point cloud data in the point cloud data of the different regions to obtain the component models and model labels corresponding to the other point cloud data. Based on the model tags and corresponding component models, the planar wall model, and the column wall model, a building information model of the construction building is generated.

9. A construction inspection device, characterized in that, include: The acquisition module is used to acquire a two-dimensional plan view of the building under construction. The inspection module is used to plan the path of the construction building according to the environmental type of different locations in the actual scene and in combination with the two-dimensional planar map, so as to control the inspection robot to inspect the construction building and obtain the three-dimensional point cloud data and on-site structural image of the construction building. The fitting module is used to fit the three-dimensional point cloud data based on a preset segmentation model using the RANSAC algorithm and the OPTICS clustering algorithm to obtain the building information model of the construction building. The calculation module is used to input the on-site structural image into a preset quality monitoring network and calculate the construction status monitoring data of the construction building. The generation module is used to generate a construction progress inspection report for the construction building based on the building information model and the construction status monitoring data. The inspection module includes: The guidance submodule is used to guide the inspection robot to perform inspections using the two-dimensional planar map. The generation submodule is used to generate the inspection path of the inspection robot at the corresponding location based on the environmental type of different locations in the actual scene during the inspection process, using a dynamic path planning method. The control submodule is used to control the inspection robot to inspect the construction building based on the inspection path, and obtain the three-dimensional point cloud data and on-site structural image of the construction building.

Images