Building engineering quality intelligent monitoring method and system based on image processing

Abstract

The invention discloses a building engineering quality intelligent monitoring method and system based on image processing, and relates to the technical field of intelligent identification, and the method mainly comprises the steps: collecting image data of each block space of building engineering from multiple angles, and constructing a block type three-dimensional space; on the basis of image point cloud matching, local detail images collected on site are fused with the corresponding block type three-dimensional space; constructing a block type three-dimensional building model in a lightweight manner, and rendering a building organization structure according to the observation distance and angle; performing improved ellipse detection processing on the original positioning point feature set of the block type three-dimensional building model to generate a target center sub-pixel coordinate; and according to the target center sub-pixel coordinates, the installation deviation of each block of the current building engineering is corrected in real time, and the precision, high efficiency and intelligence of construction quality monitoring are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent identification, and in particular to a building engineering quality intelligent monitoring method and system based on image processing. BACKGROUND

[0002] Building engineering quality, as one of the core goals of project management, is interrelated and interactive with schedule management and cost management. Today, the construction field urgently needs to accelerate the transformation and upgrading of the construction industry, and digitalization, intelligentization and sustainable development have become the main development trend of the engineering construction industry. At present, the digitalization degree of China's construction industry is low, and the integration of digital technologies represented by artificial intelligence, image data processing, Internet of Things and BIM construction technology and the construction industry needs to be accelerated.

[0003] At present, the Chinese invention patent with application number 202410217997.X discloses a building engineering quality intelligent acceptance management method and system, the main method comprising: arranging a plurality of image acquisition devices in the building engineering construction area; collecting image information in the building engineering construction area; from the collected image information in the building engineering construction area, according to all images of any target building collected by each image acquisition device, filtering out the optimal acceptance image of any target building at the current time, and adding the optimal acceptance image set; adding a time label to each image in the filtered optimal acceptance image set; training a corresponding construction prediction model for any target building based on the optimal acceptance image set after adding the time label; outputting an image prediction result using the construction prediction model; and matching the image prediction result with the expected result at the later time to obtain an acceptance result.

[0004] The above-mentioned technology has a large amount of memory calculation, cannot effectively generate a three-dimensional space model for the collected images, and lacks priority dynamic regulation and control of construction quality. SUMMARY

[0005] The technical problem solved by the present application is that the memory calculation amount is large, the three-dimensional space model cannot be effectively generated for the collected images, and the priority dynamic regulation and control of construction quality is lacking.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, a building engineering quality intelligent monitoring method based on image processing comprises the following steps:

[0008] Step S1: Collecting image data of each block space of the building engineering from multiple angles, and constructing a block type three-dimensional space;

[0009] Step S2: Based on image point cloud matching, fusing the local detail image collected on site with the corresponding block type three-dimensional space;

[0010] Step S3, constructing a block type three-dimensional building model, rendering the building organization architecture according to the observation distance and angle;

[0011] Step S4, performing improved ellipse detection processing on the original positioning point feature set of the block type three-dimensional building model to generate a target center sub-pixel coordinate;

[0012] Step S5, correcting the installation deviation of each block of the current construction project in real time according to the target center sub-pixel coordinate.

[0013] Preferably, the step S1 comprises the following sub-steps:

[0014] Step S11, dividing the construction site into a plurality of block spaces according to the construction type, and deploying three industrial-grade global shutter cameras in each block space, the industrial-grade global shutter cameras comprising a main camera, a first auxiliary camera and a second auxiliary camera;

[0015] Step S12, aligning the lens of the main camera to the vertical surface of the block space, installing the first auxiliary camera at a left inclined angle of 45 degrees of the main camera, and installing the second auxiliary camera at a right inclined angle of 45 degrees of the main camera, adjusting the distance between the first auxiliary camera and the second auxiliary camera to a unit length threshold distance of the relative space according to the length, width and height of the block space, and making the visual angle reach a set overlap rate;

[0016] Step S13, controlling the three industrial-grade global shutter cameras to be synchronously exposed by using a PLC trigger signal to obtain image data of each block space of the construction project, the image data comprising a first image, a second image and a third image, acquiring camera feature parameters comprising a feature intrinsic matrix and an external parameter between each camera, matching the same feature points of the image data by using multi-view corresponding points, and solving three-dimensional space point coordinates by using a least square method, and building a block type three-dimensional space according to the three-dimensional space point coordinates.

[0017] Preferably, the specific method of matching the same feature points of the image data by using multi-view corresponding points and solving three-dimensional space point coordinates by using a least square method comprises:

[0018] Taking the image shot by the main camera as a world coordinate, randomly labeling a space point p in the block space, obtaining the position projection of the space point p in the main camera, the first auxiliary camera and the second auxiliary camera, and establishing a three-camera projection equation by using the camera feature parameters and the p point projection of each camera, the three-camera projection equation being:

[0019] ;

[0020] wherein, K is the intrinsic parameter of each camera, K is the extrinsic parameter of each camera, is the horizontal projection of point p in each camera, is the vertical projection of point p in each camera, is the coordinate of point p in three-dimensional space; the three-camera projection equation is solved by SVD decomposition, and the least squares solution is obtained as the three-dimensional space point coordinate, a plurality of space points are generated based on image point cloud technology and the corresponding three-dimensional space point coordinates are obtained, and the block type three-dimensional space is drawn according to the three-dimensional space point coordinates. Space, the block type three-dimensional space is used to divide the whole building into different three-dimensional space shapes.

[0021] Preferably, the step S2 specifically comprises:

[0022] Collecting local detail image information, the local detail image is an RGB image, the local detail image information is texture, new geometry and spatial features, and the local detail image information is fused into the corresponding block type three-dimensional space;

[0023] The specific method of fusing the local detail image information into the corresponding block type three-dimensional space comprises:

[0024] The first feature point is extracted from the texture image corresponding to the block three-dimensional space, the second feature point is extracted based on the image feature principle, the second feature point is projected to the first feature point, and the block type three-dimensional space is updated.

[0025] The first feature point is extracted by nearest neighbor search, and the second feature point is extracted by KNN matching.

[0026] Preferably, the step S3 comprises the following sub-steps:

[0027] Step S31, the block feature points, feature edges and feature surfaces of each block type three-dimensional space are output three-dimensional mesh data by using surface reconstruction algorithm, and the three-dimensional mesh data is used for fragmenting storage block type three-dimensional space;

[0028] Step S32, the sum of the storage space size occupied by the three-dimensional mesh data is calculated, and the sum of the storage space size occupied by each block type three-dimensional space is calculated, and a lightweight difference value is generated;

[0029] Step S33, the ratio of the lightweight difference value to the sum of the storage space size occupied by each block type three-dimensional space is calculated as a storage compression ratio, and the storage compression ratio is used to represent the lightweight degree of the lightweight three-dimensional building model.

[0030] Step S34, a fragment generator is constructed, and adaptive rendering shading is performed according to the distance of each edge and corner in the three-dimensional space mesh data.

[0031] Preferably, according to the distance of each corner in the three-dimensional space grid data, the adaptive rendering coloring method specifically comprises the following steps:

[0032] Extracting the position coordinates of different key node connections of the three-dimensional space grid data, randomly selecting one key node as a starting point, calculating the Hausdorff distance from the key node to any other key node, setting a position threshold, classifying the key nodes into regions, generating a key region type, and rendering different colors as visual divisions according to the key region type;

[0033] When the Hausdorff distance of the key node is less than the set position threshold, rendering it as red as a key region type;

[0034] When the Hausdorff distance of the key node is within the set position threshold, rendering it as blue as a non-key region;

[0035] When the Hausdorff distance of the key node is greater than the set position threshold, rendering it as green as a simulated key region.

[0036] Preferably, the step S4 specifically comprises the following steps:

[0037] Extracting feature points of the block type three-dimensional model as an original positioning feature set, removing outlier points using a Gaussian filter, introducing RANSAC to remove abnormal value features using an improved ellipse detection algorithm, randomly sampling three points in the original positioning feature set to generate an ellipse model, calculating the distance of other feature points to the ellipse model, setting a distance threshold and an iteration number, selecting the model with the highest proportion of internal points of the ellipse model as the optimal ellipse, labeling the center point of the optimal ellipse in the three-dimensional model as an image coordinate, measuring the actual target position coordinates using a laser tracker, calculating the directional deviation of the image coordinates and the position coordinates, generating a directional error, and when the error is less than a certain accuracy, taking the image coordinates as the target center sub-pixel coordinates.

[0038] Preferably, the step S5 comprises the following sub-steps:

[0039] According to the target center sub-pixel coordinates and the position coordinates, performing position deviation decomposition and angle deviation decomposition, the position deviation decomposition is used to calculate the offset of the current coordinates and the planned coordinates, and the angle deviation decomposition is used to decompose into Euler angles through a rotation matrix, setting a deviation level evaluation standard, and real-time correcting the construction building splicing point position.

[0040] Preferably, the deviation level evaluation standard specifically comprises the following steps:

[0041] The deviation level is set as normal, warning, adjustment and emergency, the deviation range includes a first range, a second range, a third range and a fourth range, when the offset and Euler angle are in the first range, a continue monitoring strategy is implemented, when the offset and Euler angle are in the second range, an adjustment construction splicing point position is prepared according to a unit error, when the offset and Euler angle are in the third range, a correction compensation strategy is executed, and when the offset and Euler angle are in the fourth range, the construction operation is immediately stopped.

[0042] In a second aspect, an image processing-based construction quality intelligent monitoring system comprises a collection module, a point cloud matching module, a rendering visualization module and a calibration correction module.

[0043] The collection module is configured to collect image data of each block space of a construction project from multiple angles, and to construct a block-based three-dimensional space.

[0044] The point cloud matching module is configured to match image point clouds, and to fuse local detail images collected on site with corresponding block-based three-dimensional spaces.

[0045] The rendering visualization module is configured to construct a block-based three-dimensional building model in a lightweight manner, to fragmentize the three-dimensional building model, and to render a building organizational framework according to an observation distance and an angle.

[0046] The calibration correction module is configured to perform improved ellipse detection processing on a set of original positioning point features of the block-based three-dimensional building model, to generate target center sub-pixel coordinates, and to develop deviation level evaluation standards, and to correct installation deviations of each block of a current construction project in real time.

[0047] The present application has the following advantages: through the technical closed loop of multi-angle image collection, point cloud fusion, lightweight modeling, sub-pixel positioning and real-time deviation correction, the precision, efficiency and intelligence of construction quality monitoring are realized, firstly, the holographic three-dimensional model constructed by the multi-camera cooperative collection and cross-modal matching technology has macroscopic integrity and microscopic detail, which significantly improves the efficiency compared with traditional measurement, and effectively eliminates the single-view blind area, secondly, the innovative LOD grading strategy and GPU optimization algorithm compress the model data while still maintaining part of the key details, realizing smooth rendering of large scenes on mobile terminals and shortening the loading time, thirdly, the improved sub-pixel ellipse detection technology improves the deviation detection threshold, and the "detection-analysis-correction" closed loop system can realize automatic correction response and accurate positioning of construction installation, and finally, the whole-process digitization penetrates the quality data chain, and the deep integration with the BIM system supports error traceability analysis, realizes the paradigm shift from experience judgment to data-driven quality control, and provides a reusable technical framework for intelligent construction. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A flowchart of a method for intelligent monitoring of construction project quality based on image processing according to an embodiment of the present invention;

[0049] Figure 2 A schematic diagram of the basic flow of an intelligent monitoring system for construction project quality based on image processing provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0051] Example 1, with reference to Figure 1 , provides a construction engineering quality intelligent monitoring method based on image processing, comprising the following steps:

[0052] Step S1, collecting image data of each block space of the construction project from multiple angles to construct a block-type three-dimensional space;

[0053] Step S2: Based on image point cloud matching, the local detail image collected on site is fused with the corresponding block-type three-dimensional space;

[0054] Step S3: lightweight construction of a block-type three-dimensional building model, rendering the building organization structure according to the observation distance and angle;

[0055] Step S4, performing improved ellipse detection processing on the original positioning point feature set of the block-type three-dimensional building model to generate sub-pixel coordinates of the target center;

[0056] Step S5: Correct the installation deviation of each block of the current construction project in real time according to the sub-pixel coordinates of the target center.

[0057] Through the technology loop of multi-angle image acquisition, point cloud fusion, lightweight modeling, sub-pixel positioning and real-time deviation correction, the precision, efficiency and intelligence of construction quality monitoring are realized. First, the holographic three-dimensional model constructed by multi-camera cooperative acquisition and cross-modal matching technology has macroscopic integrity and microscopic detail, which significantly improves the efficiency of traditional measurement and effectively eliminates single-view blind area. Second, the innovative LOD grading strategy and GPU optimization algorithm compress the model data while maintaining some key details, enabling smooth rendering of large scenes on mobile devices and shortening loading time. Third, the improved sub-pixel ellipse detection technology improves the deviation detection threshold, forming a "detection-analysis-correction" closed-loop system that can automatically respond to deviation correction and achieve precise positioning of construction and installation. Finally, the whole-process digitization connects the quality data chain and supports error traceability analysis through deep integration with BIM systems, realizing the paradigm shift from experience judgment to data-driven quality control and providing a reusable technical framework for intelligent construction.

[0058] Step S1 includes the following sub-steps:

[0059] Step S11, divide the construction site into several block spaces according to the construction type, and deploy three industrial-grade global shutter cameras in each block space, including a main camera, a secondary camera 1 and a secondary camera 2.

[0060] Step S12, align the lens of the main camera to the vertical surface of the block space, install the secondary camera 1 at a 45-degree angle to the left of the main camera, and install the secondary camera 2 at a 45-degree angle to the right of the main camera. According to the internal length, width and height of the block space, adjust the distance between the secondary camera 1 and the secondary camera 2 to the unit length threshold distance of the relative space, so that the visual angle reaches the set overlap rate.

[0061] Step S13, use PLC trigger signal to control the synchronous exposure of the three industrial-grade global shutter cameras, obtain the image data of each block space of the construction project, the image data includes first image, second image and third image, obtain camera feature parameters, camera feature parameters include feature intrinsic matrix and external parameters between each camera, use multi-view corresponding points to match the same feature points of image data, and solve three-dimensional space point coordinates by least square method, build block type three-dimensional space according to three-dimensional space point coordinates.

[0062] First, according to the construction flow section (such as steel structure hoisting unit, concrete pouring section) to divide the block, the recommended size of a single block is 5m x 5m x 3m, ensure that the three camera visual field is fully covered, select industrial camera (2440 million pixels, global shutter), match 35mm (main camera) and 28mm (secondary camera) fixed focus lens, the optical axis of the main camera is perpendicular to the vertical surface (Z axis direction), the inclination angle of the secondary camera is 45°, use PLC trigger pulse to control three camera exposure, through the 45° cross field design of three cameras, effectively detect key parts such as steel structure node and curtain wall joint, PLC hard synchronization cooperates with global shutter, can stably collect in the vibration environment such as tower crane operation, the whole process from image acquisition to three-dimensional reconstruction is automated, improve the model building speed, in addition, the combination of the main camera front view and the secondary camera oblique view can capture the texture details of the facade (such as crack width 0.1mm) and reconstruct complex three-dimensional structures (such as pipeline intersection), modular block design supports parallel operation, meets the full range coverage demand of large construction site.

[0063] The specific method for matching the same feature points of image data by using multi-view corresponding points and solving the three-dimensional space point coordinates by least square method includes:

[0064] Take the image shot by the main camera as the world coordinate, randomly label a space point p in the block space, obtain the position projection of the space point p in the main camera, secondary camera 1 and secondary camera 2, use the camera feature parameters and the p point projection of each camera to establish a three camera projection equation, the three camera projection equation is:

[0065] ;

[0066] Wherein, is the internal parameter of each camera, K is the external parameter of each camera, is the horizontal projection of p point in each camera, is the vertical projection of p point in each camera, is the coordinate of p point in three-dimensional space; use SVD decomposition to solve the three camera projection equation, obtain the least square solution as the three-dimensional space point coordinate, generate a plurality of space points based on image point cloud technology and obtain the three-dimensional space point coordinates corresponding to the space points, draw a block type three-dimensional space according to the three-dimensional space point coordinates, the block type three-dimensional space is used to divide the whole building into different three-dimensional space shapes.

[0067] A right-hand system is established with the principal camera optical center as the origin and the optical axis as the Z axis, a projection equation coordinate system is constructed, and the projection equation is expanded into a linear mode. Three cameras generate a total of 6 equation groups, and the values of three unknowns X, Y and Z are solved to form an over-determined equation group. The three-dimensional coordinates of the P point are calculated according to the least square solution, the three-dimensional coordinates of all feature points are calculated by repeating the above process, and a mesh model is generated by Poisson reconstruction triangulation. In this embodiment, the noise influence of illumination change and local occlusion is suppressed by solving the over-determined equation group based on redundant observation, and the high-fidelity conversion of the construction scene from two-dimensional image to three-dimensional digital twin is realized through multi-angle geometric constraint adjustment.

[0068] Step S2 includes:

[0069] Local detail image information of the scene is collected, the local detail image is an RGB image, the local detail image information is texture, new geometry and spatial features, and the local detail image information is fused into the corresponding block three-dimensional space;

[0070] The specific method of fusing the local detail image information into the corresponding block three-dimensional space includes:

[0071] The first feature point is extracted from the texture image corresponding to the block three-dimensional space, the second feature point is extracted based on the image feature principle, the second feature point is projected to the first feature point, and the block three-dimensional space is updated.

[0072] The first feature point is extracted by nearest neighbor search, and the second feature point is extracted by KNN matching.

[0073] The first feature point is extracted from the pre-stored texture image, nearest neighbor search is used, a radius of 0.5 pixels is set to accelerate matching, the second feature point is extracted using a network, about 1500 points per frame are set, reliable corresponding points are screened by KNN matching (k=2, ratio test threshold 0.7), a transformation matrix of the local image to the three-dimensional space is solved by PnP algorithm, and the re-projection error is controlled within 0.8 pixels. This method simultaneously processes RGB images, infrared thermal images (detecting hollow), and other multi-source data, realizes the compatibility of multi-modal data, and finally matches the error between the two-dimensional image feature points and the three-dimensional image feature points, effectively improves the model anti-interference performance, and realizes the seamless connection of the construction scene from local to global data.

[0074] Step S3 includes the following sub-steps:

[0075] In step S31, the block feature points, feature edges and feature surfaces of each block three-dimensional space are output as three-dimensional mesh data by a surface reconstruction algorithm, and the three-dimensional mesh data is used for fragmented storage of the block three-dimensional space.

[0076] Step S32, calculate the sum of the storage space size occupied by the three-dimensional grid data, and the sum of the storage space size occupied by each block three-dimensional space, and generate a lightweight difference value;

[0077] Step S33, calculate the ratio of the lightweight difference value to the sum of the storage space size occupied by each block three-dimensional space, as the storage compression ratio, which represents the degree of lightweight of the lightweight three-dimensional building model;

[0078] Step S34, construct a segment generator to adaptively render shading according to the distance of each edge corner in the three-dimensional space grid data.

[0079] In this embodiment, the data compression ratio is improved through QEM simplification and texture compression, and the GPU load rate is effectively reduced according to the view distance driven LOD strategy. This method solves the industry problem that large BIM models cannot achieve precision and efficiency by combining quantitative compression evaluation and dynamic rendering technology.

[0080] The method of adaptively rendering shading according to the distance of each edge corner in the three-dimensional space grid data specifically includes:

[0081] Extract the position coordinates of the connection of different key nodes of the three-dimensional space grid data, randomly select one key node as the starting point, calculate the Hausdorff distance from the key node to any other key node, set a position threshold, classify the key nodes by region, generate a key region type, and render different colors according to the key region type as a visual division;

[0082] When the Hausdorff distance of the key node is less than the set position threshold, render it as red as the key region type;

[0083] When the Hausdorff distance of the key node is within the set position threshold, render it as blue as the non-key region;

[0084] When the Hausdorff distance of the key node is greater than the set position threshold, render it as green as the simulation key region.

[0085] The method realizes intelligent identification and visual distinction of key areas in building three-dimensional models through Hausdorff distance measurement and dynamic coloring strategy. The technical logic is to first extract key nodes (such as beam-column connection points, pipe intersections, etc.) in the grid data, calculate the Hausdorff distance between nodes with a random starting point as the reference. The distance can effectively represent the maximum spatial deviation between structures. In the implementation, a double threshold (0.5m and 2m) is set to divide the area into three categories: when the distance is less than 0.5m, it is marked as a red key area (such as steel structure welds, equipment installation nodes, etc.), indicating that there is a high-precision construction requirement or potential collision risk; when the distance is between 0.5-2m, it is displayed as a blue non-key area (regular walls, floors, etc.), indicating the standard construction tolerance range; and when the distance exceeds 2m, it is rendered as a green simulation area (such as temporary support, construction gap, etc.), used for virtual construction rehearsal. The advantage of this method is that it replaces manual experience judgment with mathematical measurement, improving the accuracy of key area identification. The dynamic coloring mechanism reduces the time-consuming of visual analysis, and the supervisor can quickly locate the high-risk nodes. The threshold can be adjusted to adapt to different construction scenes such as steel structure (threshold 0.3m) and curtain wall (0.8m), and combined with the BIM system to improve the quality inspection efficiency, providing an intuitive spatial relationship cognition tool for intelligent construction.

[0086] Step S4 specifically includes:

[0087] The feature points of the block type three-dimensional model are extracted as the original positioning feature set, the Gaussian filter is used to remove outlier points, the improved ellipse detection algorithm is used, and the RANSAC is introduced to remove abnormal value features. Three points are randomly sampled from the original positioning feature set to generate an ellipse model, the distance of other feature points to the ellipse model is calculated, the distance threshold and the iteration number are set, the model with the highest proportion of internal points of the ellipse model is selected as the optimal ellipse, and the center point of the optimal ellipse in the three-dimensional model is marked as the image coordinate. The actual target position coordinates are measured by the laser tracker, the direction deviation of the image coordinates and the position coordinates is calculated, the direction error is generated, and when the error is less than a certain accuracy, the image coordinates are taken as the target center sub-pixel coordinates.

[0088] Firstly, the feature point set is extracted from the three-dimensional model, and after removing outliers by Gaussian filtering (σ=0.5mm), the RANSAC optimization ellipse fitting is adopted: 3 feature points are randomly selected to generate the ellipse parameter equation, the algebraic distance (threshold value is set to 0.3 pixels) of other points to the ellipse is calculated, and through 500 iterations, the ellipse model with the highest proportion of inliers (coverage rate needs to be >85%) is screened. The center of the ellipse is projected to the image coordinate system, and the directional deviation is compared with the measured coordinates of the laser tracker. When the error is <0.05 pixels, it is determined as an effective target center. The improved ellipse detection algorithm reduces the positioning error of the traditional Hough transform through the RANSAC anti-noise mechanism; the laser tracker closed-loop verification system eliminates the cumulative error and improves the absolute accuracy of the target coordinate. The full-automatic processing flow (time consumption <3 seconds / point) greatly improves the efficiency of manual total station measurement. Through the improved ellipse detection algorithm combined with the laser tracker calibration, the sub-pixel level high-precision positioning of the target center in the construction engineering is realized.

[0089] Step S5 includes the following sub-steps:

[0090] The position deviation decomposition is used to calculate the offset of the current coordinate and the planned coordinate, and the angle deviation decomposition is used to decompose the rotation matrix into Euler angles. The deviation level evaluation standard is set to correct the position of the splicing point of the construction building in real time.

[0091] Through multi-dimensional deviation analysis of the sub-pixel coordinates and the measured data of the laser tracker, real-time detection and dynamic correction of the installation error of the construction engineering are realized. Firstly, the sub-pixel coordinates (accuracy ±0.1mm) of the target center and the BIM design coordinates are decomposed for position deviation, and the three-dimensional offset (ΔX, ΔY, ΔZ) is calculated. At the same time, the singular value decomposition of the rotation matrix is used to obtain the Euler angle deviation (pitch angle Δα, yaw angle Δβ, roll angle Δγ). In the implementation, a multi-level evaluation standard is set: the first level deviation (Δ<1mm / 0.1°) is only recorded without alarm; the second level deviation (1mm≤Δ<3mm or 0.1°≤Δ<0.3°) triggers the sound-light warning; the third level deviation (Δ≥3mm / 0.3°) automatically links the hydraulic correction system for compensation, and the correction response time is <200ms. The advantage of this method is that the full-parameterized deviation decomposition model can identify 6-degree-of-freedom error, and the progressive correction strategy is adopted. Through the PID control algorithm, the installation accuracy of large components (such as 30m steel beams) is stably controlled within ±1.5mm. The method is deeply integrated with the BIM system, and the deviation data automatically generates a quality traceability report, avoiding the risk of high-altitude rework.

[0092] The deviation level evaluation standard specifically includes:

[0093] The deviation level is set as normal, warning, adjustment and emergency, the deviation range includes first range, second range, third range and fourth range, when the offset and Euler angle are in the first range, the implementation continues to monitor the strategy, when the offset and Euler angle are in the second range, the construction building splicing point position is adjusted according to the unit error, when the offset and Euler angle are in the third range, the correction compensation strategy is executed, and when the offset and Euler angle are in the fourth range, the construction operation is immediately stopped.

[0094] Through the four-level deviation evaluation system and the hierarchical response mechanism, the dynamic and intelligent control of the installation precision of the building engineering is realized. Firstly, based on the real-time comparison of sub-pixel coordinates and BIM design values, the deviation is divided into four levels: normal level (ΔX / Y / Z≤1mm and Δα / β / γ≤0.1°) only records data and continuously monitors; warning level (1mm<Δ≤3mm or 0.1°<Δ≤0.3°) automatically generates adjustment plan, adjustment level (3mm<Δ≤5mm or 0.3°<Δ≤0.5°) triggers automatic compensation system, which is completed by hydraulic jack (precision ±0.2mm) and servo motor (±0.01°) to correct the pose; emergency level (Δ>5mm or Δ>0.5°) immediately cuts off the power supply of the construction equipment and starts the sound and light alarm to prevent structural instability, and the parameters are optimized by engineering verification: the steel structure adopts more stringent standards (emergency threshold is set to 3mm / 0.3°), and the concrete structure is appropriately relaxed (5mm / 0.5°), which provides a standardized and expandable precision control paradigm for intelligent construction.

[0095] Embodiment 2, refer to Figure 2 provides a building engineering quality intelligent monitoring system based on image processing, comprising an acquisition module, a point cloud matching module, a rendering visualization module and a calibration correction module;

[0096] The acquisition module is used for multi-angle acquisition of image data of each block space of the building engineering, and a block type three-dimensional space is constructed;

[0097] The point cloud matching module is used for image point cloud matching, and the locally collected detail images are fused with the corresponding block type three-dimensional space;

[0098] The rendering visualization module is used for lightweight construction of the block type three-dimensional building model, fragmentation processing of the three-dimensional building model, and rendering of the building organization architecture according to the observation distance and angle;

[0099] The calibration correction module is used for improved ellipse detection processing of the original positioning point feature set of the block type three-dimensional building model, generation of target center sub-pixel coordinates, and development of deviation level evaluation standards, and real-time correction of installation deviation of each block of the current building engineering.

[0100] In this embodiment, the intelligent control of the whole process from data collection to deviation correction is realized through multi-module cooperation, and a reusable intelligent construction quality control paradigm is formed.

[0101] Those skilled in the art will understand that embodiments of the present application can be provided as methods, systems or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media having computer-usable program code embodied in the medium. The storage media can be realized by any type of volatile or non-volatile storage devices or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. These computer program instructions can also be stored in a computer readable storage medium that can guide a computer or other programmable data processing device to work in a specific way, so that the instructions stored in the computer readable storage medium produce a product including instruction devices, which realize the processes Figure 1 one or more processes and / or blocks Figure 1 the functions specified in one or more blocks.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalent ones without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. An intelligent monitoring method for construction project quality based on image processing, characterized in that: The steps include: Step S1, collecting image data of each block space of the construction project from multiple angles to construct a block-type three-dimensional space; Step S2: Based on image point cloud matching, the local detail image collected on site is fused with the corresponding block-type three-dimensional space; Step S3: lightweight construction of a block-type three-dimensional building model, rendering the building organization structure according to the observation distance and angle; Step S4, performing improved ellipse detection processing on the original positioning point feature set of the block-type three-dimensional building model to generate sub-pixel coordinates of the target center; Step S5: Correct the installation deviation of each block of the current construction project in real time according to the sub-pixel coordinates of the target center.

2. The method for intelligent monitoring of construction project quality based on image processing according to claim 1, characterized in that: The step S1 includes the following sub-steps: Step S11: Divide the construction site into several block spaces according to the construction type, and deploy three industrial-grade global shutter cameras in each block space. The industrial-grade global shutter cameras include a main camera, auxiliary camera 1, and auxiliary camera 2; Step S12: Align the lens of the main camera with the front elevation of the block space, install the auxiliary camera 1 at a 45-degree left angle to the main camera, and install the auxiliary camera 2 at a 45-degree right angle to the main camera. Based on the internal length, width, and height of the block space, adjust the distance between the auxiliary cameras 1 and 2 to a unit length threshold distance of the relative space so that the visual angle reaches a set overlap ratio; Step S13, using the PLC trigger signal to control three industrial-grade global shutter cameras for synchronous exposure, obtain image data of each block space of the construction project, the image data including the first image, the second image and the third image, obtain camera feature parameters, the camera feature parameters including the feature intrinsic parameter matrix and the extrinsic parameters between each camera, use multi-viewpoint corresponding phase points to match the same feature point of the image data, and solve the three-dimensional space point coordinates through the least squares method, and build a block-type three-dimensional space according to the three-dimensional space point coordinates.

3. The method for intelligent monitoring of construction project quality based on image processing according to claim 2, characterized in that: The specific method of using corresponding phase points from multiple perspectives to match the same feature point of image data and solving the coordinates of the three-dimensional space point by the least squares method includes: Using the image captured by the main camera as the world coordinate, randomly mark a spatial point p in the block space, obtain the position projection of the spatial point p in the main camera, auxiliary camera 1 and auxiliary camera 2, and use the camera characteristic parameters and the projection of point p of each camera to establish the three-camera projection equation. The three-camera projection equation is: ； in, is the internal parameter of each camera, K is the external parameter of each camera, is the horizontal projection of point p in each camera, is the vertical projection of point p in each camera, is the coordinate of point p in three-dimensional space; the three-camera projection equation is solved by SVD decomposition, and the least squares solution is obtained as the three-dimensional space point coordinates. Based on the image point cloud technology, several spatial points are generated and the three-dimensional space point coordinates corresponding to the spatial points are obtained. The block-type three-dimensional space is drawn according to the three-dimensional space point coordinates. The block-type three-dimensional space is used to divide the entire building into different three-dimensional space shapes.

4. The method for intelligent monitoring of construction project quality based on image processing according to claim 3, characterized in that: The step S2 comprises: Collecting local detail image information of the scene, where the local detail image is an RGB image and includes texture, newly added geometric figures, and spatial features, and fusing the local detail image information into the corresponding block-type three-dimensional space; The specific method of fusing the local detail image information into the corresponding block-type three-dimensional space includes: Extracting a first feature point from a texture image corresponding to the block three-dimensional space, extracting a second feature point from the local detail image based on an image feature principle, projecting the second feature point onto the first feature point, and updating the block three-dimensional space; The first feature points are extracted by nearest neighbor search, and the second feature points are extracted by KNN matching.

5. The method for intelligent monitoring of construction project quality based on image processing according to claim 4, characterized in that: The step S3 includes the following sub-steps: Step S31, outputting three-dimensional mesh data from the block feature points, feature edges, and feature faces of each block-type three-dimensional space using a surface reconstruction algorithm, wherein the three-dimensional mesh data is used for fragmented storage of the block-type three-dimensional space; Step S32, calculating the sum of the storage space occupied by the three-dimensional grid data, and simultaneously calculating the sum of the storage space occupied by each block-type three-dimensional space, and generating a lightweight difference; Step S33: Calculate the ratio of the lightweight difference to the sum of the storage spaces occupied by the three-dimensional block spaces as a storage compression ratio, wherein the storage compression ratio is used to represent the lightweight degree of the lightweight three-dimensional building model; Step S34: constructing a fragment generator to adaptively render and color according to the distances between corners in the three-dimensional space grid data.

6. The method for intelligent monitoring of construction project quality based on image processing according to claim 5, characterized in that: According to the distances of the corners in the three-dimensional space grid data, the method of adaptive rendering and shading specifically includes: Extracting the position coordinates of the connection points of different key nodes of the three-dimensional spatial grid data, randomly selecting one of the key nodes as the starting point, calculating the Hausdorff distance from the key node to any other key node, setting a position threshold, classifying the key nodes into regions, generating key region types, and rendering different colors according to the key region types as visual divisions; When the Hausdorff distance of the key node is less than the set position threshold, it is rendered in red as a key area type; When the Hausdorff distance of the key node is within the set position threshold, it is rendered in blue as a non-key area; When the Hausdorff distance of the key node is greater than the set position threshold, it is rendered in green as a simulation key area.

7. The method for intelligent monitoring of construction project quality based on image processing according to claim 6, characterized in that: The step S4 specifically includes: The feature points of the block-type three-dimensional model are extracted as the original positioning feature set, the outlier points are removed using a Gaussian filter, the improved ellipse detection algorithm is used, and the RANSAC feature is introduced to eliminate the outlier feature, three points are randomly sampled in the original positioning feature set to generate an ellipse model, the distance from other feature points to the ellipse model is calculated, the distance threshold and the number of iterations are set, the model with the highest proportion of internal points of the ellipse model is selected as the optimal ellipse, the center point of the optimal ellipse is marked in the three-dimensional model as the image coordinate, the actual target position coordinates are measured using a laser tracker, the direction deviation of the image coordinates and the position coordinates is calculated to generate a direction error, and when the error is less than a certain accuracy, the image coordinates are used as the sub-pixel coordinates of the target center.

8. The method for intelligent monitoring of construction project quality based on image processing according to claim 7, characterized in that: The step S5 includes the following sub-steps: Position deviation decomposition and angle deviation decomposition are performed based on the sub-pixel coordinates of the target center and the position coordinates. The position deviation decomposition is used to calculate the offset between the current coordinates and the planned coordinates. The angle deviation decomposition is used to decompose into Euler angles through a rotation matrix, set the deviation level evaluation standard, and correct the position of the construction building splicing point in real time.

9. The method for intelligent monitoring of construction project quality based on image processing according to claim 8, characterized in that: The deviation level assessment criteria specifically include: The deviation levels are set to normal, warning, adjustment and emergency. The deviation ranges include the first range, the second range, the third range and the fourth range. When the offset and the Euler angle are in the first range, the continued monitoring strategy is implemented. When the offset and the Euler angle are in the second range, the position of the construction building splicing point is adjusted according to the unit error. When the offset and the Euler angle are in the third range, the correction compensation strategy is executed. When the offset and the Euler angle are in the fourth range, the construction operation is stopped immediately.

10. An intelligent monitoring system for construction project quality based on image processing, which is applied to an intelligent monitoring method for construction project quality based on image processing according to any one of claims 1 to 9, characterized in that: Includes acquisition module, point cloud matching module, rendering visualization module and calibration correction module; The acquisition module is used to collect image data of each block space of the construction project from multiple angles to construct a block-type three-dimensional space; The point cloud matching module is used for image point cloud matching, fusing the local detail image collected on site with the corresponding block-type three-dimensional space; The rendering and visualization module is used to construct a lightweight block-type three-dimensional building model, fragment the three-dimensional building model, and render the building organization structure according to the observation distance and angle; The calibration and correction module is used to perform improved ellipse detection processing on the original positioning point feature set of the block-type three-dimensional building model, generate sub-pixel coordinates of the target center, and formulate deviation level evaluation standards to correct the installation deviation of each block of the current construction project in real time.

Images