A method and device for detecting steel bars of a fabricated shear wall edge component

Abstract

The application relates to a steel bar detection method and equipment for a fabricated shear wall edge component, and the method comprises the following steps: adopting an industrial camera to shoot steel bars of the fabricated shear wall edge component from multiple angles, obtaining shooting images, labeling steel bar regions and categories, establishing a target detection data set, training a target detection algorithm, and obtaining a steel bar recognition model; when on-site detection is performed, an auxiliary calibration plate is vertically erected at a central layer position of the fabricated shear wall edge component, a camera module is adopted to shoot and obtain detection images; sub-pixel corner point coordinates and pixel precision of the auxiliary calibration plate are analyzed, a standardized perspective transformation matrix is calculated, and the shooting angle of the images is corrected; geometric size and non-geometric size information is obtained by adopting the steel bar recognition model, and the information is compared with design information of the component to detect whether the component meets the specification. Compared with the prior art, the application has the advantages of automatic and accurate steel bar detection, high automation degree, simple operation and the like.

Description

Technical Field

[0001] This invention relates to the field of rebar detection technology, and in particular to an intelligent method and equipment for detecting rebar in edge components of prefabricated shear walls. Background Technology

[0002] Prefabricated buildings should adhere to the principle of equivalent cast-in-place construction, requiring edge members to be installed at both ends of shear walls and on both sides of openings. However, the construction requirements for shear wall edge nodes are complex, necessitating inspection of the connection methods and quantities of reinforcing bars (longitudinal and transverse bars), lap lengths, locations, and spacing of transverse bars. Traditional inspection methods typically require manual observation and measurement by relevant personnel, which is inefficient and may be subject to subjectivity and inaccuracy. In recent years, the development of technologies such as machine vision and deep learning has provided new solutions for the inspection of reinforcing steel structures.

[0003] Existing machine vision technology for inspecting steel reinforcement projects mainly targets single-layer planar steel mesh, identifying its diameter and spacing, or counting steel bundles through images of steel cross-sections. Its application is still limited to two-dimensional scenarios. A complete inspection system has not yet been formed for complex three-dimensional scenarios of shear wall edge components. Therefore, a targeted intelligent inspection technology and supporting inspection equipment are needed.

[0004] Regarding the exploration of intelligent inspection technology, existing technical approaches generally focus on improving upon traditional manual inspection methods by integrating 3D modeling and multimodal data analysis. This aims to overcome the bottlenecks of low efficiency and poor accuracy associated with traditional manual inspection. The general approach for 3D modeling is to use LiDAR combined with multi-view vision fusion technology to create 3D point cloud models of shear wall edge components. 3D point set registration algorithms are then used to match these models with design drawings to detect shape and position errors. The general approach for multimodal data analysis is to use high-resolution cameras, infrared imaging technology, and image segmentation algorithms to detect reinforcing bars.

[0005] The aforementioned existing intelligent exploration concepts all require a large amount of computing resources and are greatly affected by noise. They do not take into account the physical characteristics of the actual objects being detected, and their accuracy cannot be guaranteed. Summary of the Invention

[0006] The purpose of this invention is to overcome the defects of the prior art by providing a method and equipment for detecting reinforcing bars in the edge components of prefabricated shear walls, thereby achieving automated and precise reinforcing bar detection.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A method for inspecting the reinforcing bars in the edge members of a prefabricated shear wall includes the following steps:

[0009] Industrial cameras were used to capture images of the steel reinforcement of the edge components of the prefabricated shear wall from multiple angles. The captured images were preprocessed and data augmented to generate diverse training samples. The training samples were then labeled with steel reinforcement regions and categories to establish a target detection dataset.

[0010] Using a target detection dataset, a neural network-based target detection algorithm was trained to obtain a rebar recognition model;

[0011] During on-site testing, an auxiliary calibration plate is vertically mounted on the center layer of the prefabricated shear wall edge component. A camera module with multiple viewing angles is used to capture multiple sections of the prefabricated shear wall edge component, and the captured images include the auxiliary calibration plate to obtain the test images.

[0012] Based on the detected image, the sub-pixel corner coordinates and pixel accuracy of the auxiliary calibration board are analyzed, and the standardized perspective transformation matrix is ​​calculated to correct the shooting angle of the detected image.

[0013] For the corrected detection image, the detection frame coordinates and non-geometric dimension information of the steel bars are identified using a steel bar recognition model; based on the detection frame coordinates of the steel bars, the projection points of each steel bar on the central layer are determined, and the geometric dimension information between each steel bar is determined based on the pixel coordinates and pixel precision of the projection points.

[0014] The geometric and non-geometric dimensional information of the reinforcing bars are compared with the design information of the edge components of the prefabricated shear wall to check whether the reinforcing bars meet the specifications.

[0015] Furthermore, based on the auxiliary calibration board in the detected image, the OpenCV corner detection algorithm is used to calculate the sub-pixel corner coordinates to obtain the actual distance of the corner. and pixel distance ;

[0016] The expression for calculating the pixel precision is:

[0017]

[0018] In the formula, Where n represents pixel precision, and n×n represents the size of the auxiliary calibration board.

[0019] Furthermore, based on the parsed sub-pixel corner coordinates, the normalized perspective transformation matrix is ​​calculated using four sets of matching points. The selection rules for the four sets of matching points are as follows:

[0020] The calibration plate edge corner point closest to the center of the detection image is used as the base point. Select the remaining three points in a clockwise direction. ;

[0021] Find the vector that is parallel to the four standard direction vectors. Standard direction vector with the smallest included angle ; and take ;

[0022] The coordinates of the four selected matching points are as follows:

[0023]

[0024] In the formula, and As the first matching point, and The second matching point, and The third matching point, and This is the fourth matching point;

[0025] Based on four sets of matching points Based on the principle of image homography, the standardized perspective transformation matrix is ​​calculated.

[0026] Furthermore, the geometric dimension information includes the transverse and longitudinal spacing between the reinforcing bars, the lap length, the position of the reinforcing bars, and the position of the joints. The calculation expressions for the transverse and longitudinal spacing are as follows:

[0027]

[0028]

[0029] In the formula, For steel reinforcement i , j Horizontal spacing between them For steel reinforcement i , j The longitudinal spacing between them The coordinates of the projected points after standardization and perspective transformation correction. For pixel precision;

[0030] The lap length includes the lap length between transverse reinforcing bars and the lap length between longitudinal reinforcing bars. The calculation expressions for the lap length between transverse reinforcing bars and the lap length between longitudinal reinforcing bars are as follows:

[0031]

[0032]

[0033] In the formula, For horizontal reinforcement i , j The overlap length between them Longitudinal reinforcement i , j The overlap length between them The boundary coordinates within the intersection range of the two rebar detection frames after standardized perspective transformation correction.

[0034] The location of the reinforcing bars refers to the actual height of the transverse reinforcing bars from the floor surface. The expression for calculating the actual height of the transverse reinforcing bars from the floor surface is as follows:

[0035]

[0036] In the formula, For horizontal reinforcement i The actual height from the floor, h 0 To determine the actual height of the center point of the calibration board from the floor, The ordinate of the center point of the calibration board after standardization and perspective transformation correction.

[0037] The joint location includes the actual height of the longitudinal reinforcement joints near and away from the camera from the floor. Since the joint is located on an independent longitudinal reinforcement, the longitudinal coordinate of the joint needs to be corrected to the center layer. The expression for calculating the actual height of the longitudinal reinforcement joints near and away from the camera from the floor is as follows:

[0038]

[0039]

[0040]

[0041]

[0042] In the formula, The actual height of the longitudinal reinforcement joint near the camera from the floor surface. The actual height of the longitudinal reinforcement joint away from the floor is the distance from the camera. Points in the longitudinal reinforcement joint inspection frame after standardized perspective transformation correction. Correct the longitudinal coordinates of the longitudinal rib joints near the camera to the center layer. To correct the longitudinal coordinates of the longitudinal reinforcement joints located away from the camera to the center layer, The distance between the longitudinal reinforcement bars that are closer to and farther from the camera. The shooting distance from the camera to the center layer. The height is the image pixel height.

[0043] Furthermore, the method of photographing the reinforcing bars of the prefabricated shear wall edge members from multiple angles includes:

[0044] Images taken from the front and side of the edge components of the prefabricated shear wall;

[0045] The labeling of the reinforcement area and category includes:

[0046] For images taken from the front, the coordinates and categories of the reinforcing bars are labeled according to the principle of minimum bounding rectangle. For images taken from the side, the longitudinal bars and joints are labeled.

[0047] Furthermore, the preprocessing includes cropping and size standardization operations;

[0048] The data enhancements include flipping, brightness adjustment, and contrast adjustment.

[0049] Furthermore, the target detection algorithm is an anchorless neural network structure;

[0050] Alternatively, an anchored neural network structure can be used, and the aspect ratio of the predefined anchor frames in the anchored neural network structure can be adjusted to adapt to the shape of the reinforcing bars.

[0051] Furthermore, the multiple viewing angles of the camera module include an elevation angle, a level view, and a depression angle, and the auxiliary calibration board can be a single piece or multiple pieces;

[0052] The central layer is the virtual plane containing the center of the reinforcement cage of the prefabricated shear wall edge component, and it is parallel to the wall surface. Before taking the picture, the auxiliary calibration plate is fixed on the outer central layer of the reinforcement cage, and the auxiliary calibration plate is adjusted to be vertical to the floor. The height of the center point of the calibration plate from the floor is recorded at this time. h 0 .

[0053] Before the camera module takes pictures, the shooting distance is adjusted so that the camera module can sequentially capture the upper, middle and lower sections of the complete prefabricated shear wall edge component from each viewing angle. The steel bars captured in the three sections are allowed to overlap, but omissions are not allowed.

[0054] When there are multiple auxiliary calibration boards, at least one of the upper, middle, and lower sections of the image is included; when there is only one auxiliary calibration board, the middle section of the image is included.

[0055] Information is integrated from the captured images of the three intervals. When there are multiple auxiliary calibration plates, the position of the auxiliary calibration plate in each interval is used as the boundary of the detection interval information. When there is only one auxiliary calibration plate, the boundary of the overlapping area is used as the boundary of the detection interval information.

[0056] Furthermore, the non-geometric dimension information includes the number of steel bars, connection method, and edge member type of the prefabricated shear wall edge members. This information is statistically analyzed by keyword matching of the inference results from a deep learning model, and the number of longitudinal bars that may be obscured needs to be compensated.

[0057] The process of comparing the geometric and non-geometric dimensional information of the reinforcing bars with the design information of the edge components of the prefabricated shear wall is as follows:

[0058] Based on the design information of the prefabricated shear wall edge components, a parametric virtual model of the reinforcing steel bars of the prefabricated shear wall edge components is established;

[0059] A parametric solid model is established based on the detected geometric and non-geometric dimensional information of the reinforcing bars;

[0060] The parametric virtual model is compared with the parametric solid model to check whether the steel reinforcement meets the specifications.

[0061] This embodiment also provides an apparatus for implementing the above-described method for detecting the reinforcing bars of the edge members of a prefabricated shear wall, comprising:

[0062] A camera module includes a low-angle camera, a level-up camera, a high-angle camera, and a device housing. The low-angle camera is mounted on the device housing via an angle adjustment assembly and is equipped with an angle adjustment knob. The level-up camera is fixed to the device housing. The high-angle camera is mounted on the device housing via a tilt adjustment assembly and is equipped with a tilt adjustment knob.

[0063] The auxiliary calibration plate includes an interconnected spherical universal support and a calibration plate. The calibration plate is connected to a bubble level. The spherical universal support is vertically mounted on the center layer of the prefabricated shear wall edge component via a snap fastener.

[0064] The image processing and visualization module is connected to the camera module and is used for data transmission and execution of the calculation process of the rebar detection method.

[0065] Compared with the prior art, the present invention has the following advantages:

[0066] (1) Accuracy: The calibration plate, center layer projection and continuous information interval integration strategy are adopted. The sub-pixel corner coordinates of the calibration plate are analyzed according to the captured image, and the pixel accuracy and standardized perspective transformation matrix are calculated to realize perspective correction of the captured image, obtain a more accurate center layer projection point position, and on this basis, combine the pixel accuracy to calculate a more accurate geometric dimension information of the steel bars; the continuous information interval integration strategy also realizes the integration of captured images of the entire prefabricated shear wall edge components;

[0067] Overall, it avoids the error problems of traditional image stitching methods; the strategy of continuous information intervals ensures the image quality within each interval, provides more accurate detection results, and is suitable for processing complex 3D models such as tall edge components.

[0068] (2) High degree of automation: Through the combination of deep learning and machine vision technology, the detection of steel bars has been highly automated, reducing manual intervention and improving detection efficiency and reliability.

[0069] (3) Easy to operate: The detection equipment and process design of the present invention take into account the actual operation needs on site. It adopts automatic focusing, auxiliary calibration plate and camera module with specific angle, which makes the operation easier and reduces the requirements for the professional skills of the operators. Attached Figure Description

[0070] Figure 1 This is a flowchart illustrating a method for detecting reinforcing bars in the edge components of a prefabricated shear wall, as provided in an embodiment of the present invention.

[0071] Figure 2 This is a schematic diagram showing the angle between the camera imaging plane and the calibration plate plane of a four-calibration plate according to an embodiment of the present invention.

[0072] Figure 3 This is a schematic diagram showing the angle between the camera imaging plane and the calibration plate plane of two calibration plates provided in an embodiment of the present invention;

[0073] Figure 4 This is a schematic diagram showing the angle between the camera imaging plane and the calibration plate plane of a single calibration plate provided in an embodiment of the present invention;

[0074] Figure 5 This is a schematic diagram of the central layer of a prefabricated shear wall edge component provided in an embodiment of the present invention;

[0075] Figure 6 This is a schematic diagram of the projection point of the center layer of the transverse reinforcing steel provided in an embodiment of the present invention;

[0076] Figure 7 This is a schematic diagram of the overlapping area between the upper and middle sections during single calibration board shooting, provided in an embodiment of the present invention.

[0077] Figure 8 This is a schematic diagram of the side geometry of a longitudinal reinforcement joint provided in an embodiment of the present invention;

[0078] Figure 9 This is a front structural schematic diagram of an intelligent detection device for reinforcing bars of prefabricated shear wall edge components provided in an embodiment of the present invention;

[0079] Figure 10This is a schematic diagram of the front right view of an intelligent detection device for reinforcing bars of prefabricated shear wall edge components provided in an embodiment of the present invention;

[0080] Figure 11 This is a schematic diagram of the rear right structure of an intelligent detection device for reinforcing bars of prefabricated shear wall edge components provided in an embodiment of the present invention;

[0081] Figure 12 This is a schematic diagram showing the top and front views of a matching auxiliary calibration plate provided in an embodiment of the present invention.

[0082] In the diagram: 1. Main body of the equipment casing; 2. Automatic focusing industrial camera viewed from below; 3. Automatic focusing industrial camera viewed at eye level; 4. Automatic focusing industrial camera viewed from above; 5. Elevation adjustment knob; 6. Depression adjustment knob; 7. Shoulder strap hole; 8. Small light bulb; 9. Data cable expansion dock; 10. Data connection hole; 11. Image processing and visualization module; 12. Rib bayonet; 13. Sliding guide rail; 14. Spherical universal support; 15. Calibration plate; 16. Bubble level. Detailed Implementation

[0083] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0084] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0085] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0086] Example 1

[0087] like Figure 1 As shown in the figure, this embodiment provides a method for detecting the reinforcing bars of the edge members of a prefabricated shear wall, including the following steps:

[0088] S1: Use an industrial camera to photograph the reinforcing bars of the prefabricated shear wall edge components from multiple angles to obtain images;

[0089] An industrial camera with automatic focal length adjustment can be used to photograph the steel reinforcement of the prefabricated shear wall edge components from multiple angles, both front and side, within a range of 1-3 meters.

[0090] S2: Label the captured images, label the xywh coordinates and categories of the reinforcing bars according to the minimum bounding rectangle principle, and label only the longitudinal bars and joints in the side images to establish a target detection dataset;

[0091] S3: Preprocess the image, including cropping and resizing, to meet the input requirements of the convolutional neural network;

[0092] S4: Apply data augmentation techniques, including flipping, brightness adjustment, and contrast adjustment, to generate diverse training samples to improve the robustness of the model.

[0093] S5: Use deep learning target detection algorithms to train a rebar recognition model, and adopt an anchorless neural network structure or an anchored neural network structure with adjusted predefined anchor frames.

[0094] S6: During on-site testing, the auxiliary calibration plate is vertically erected at the center layer of the prefabricated shear wall edge component, and the height from the ground is recorded. h 0 The camera module with multiple viewing angles is used to capture multiple sections of the edge components of the prefabricated shear wall, and the captured images include an auxiliary calibration plate to obtain the detection image;

[0095] Preferably, the camera module has an elevation angle. α Eye level and downward angle β The camera module captures images of the upper, middle, and lower sections of the edge components of the prefabricated shear wall.

[0096] S7: Based on the detected image, analyze the sub-pixel corner coordinates and pixel accuracy of the auxiliary calibration board. ε Calculate the normalized perspective transformation matrix M To correct the shooting angle of the detected image;

[0097] For the corrected detection image, the detection frame coordinates of the steel bars are identified using a steel bar recognition model; based on the detection frame coordinates of the steel bars, the projection points of each steel bar on the central layer are determined, and the geometric dimension information between each steel bar is determined based on the pixel coordinates and pixel precision of the projection points.

[0098] S8: Based on the corrected image, apply the trained rebar recognition model to count non-geometric dimensional information such as the number of rebars, connection method, and edge component type; for longitudinal bars that may be occluded, perform quantity compensation;

[0099] S9: Based on design information, establish a parametric virtual model of the reinforcement of the edge members of the prefabricated shear wall on the BIM platform; and establish a parametric solid model using the information obtained from the inspection.

[0100] S10: Compare the virtual model with the solid model to check whether the connection method, quantity, lap length, position and spacing of the reinforcing bars meet the specifications.

[0101] Specifically,

[0102] In step S2, images taken from the front and side of the prefabricated shear wall edge components are labeled with xywh coordinates and categories of the reinforcing bars in the images according to the minimum bounding rectangle principle using a labeling tool. Since it is difficult to observe the type of transverse reinforcing bars from the side, the images taken from the side are labeled only with longitudinal bars and joints, thus establishing a target detection dataset.

[0103] Optionally, the wall extension reinforcement, additional connecting reinforcement, stirrups, longitudinal reinforcement, tie bars, and longitudinal reinforcement connection nodes of different types of prefabricated shear wall edge members can be labeled, and the xywh coordinates of the reinforcement in the image can be labeled according to the principle of minimum circumscribed rectangle. For the reinforcement category, the atlas "Construction of Connection Nodes of Prefabricated Shear Wall (Shear Wall)" 15G310-2 defines different forms of shear wall edge members. For example, for edge concealed column nodes, four forms are specified: Q3-1, Q3-2, Q3-3(1) and Q3-3(2) (Q3-3(1) and Q3-3(2) represent two different ways of Q3-3). The difference between the four concealed column forms is mainly in the form of transverse reinforcement. The reinforcement category in the image can be labeled according to the naming method in Table 1 below. Considering that the Ministry of Housing and Urban-Rural Development has restricted the use of flash butt welding for reinforcements larger than 22mm in the "List of Elimination of Construction Technology, Equipment and Materials that Endanger Production Safety in Housing Construction and Municipal Infrastructure Projects (First Batch)", the welding nodes are specially distinguished from flash butt welding.

[0104] Table 1. Labeling and Naming of Reinforcing Bars or Connections

[0105]

[0106] In step S5, a deep learning-based object detection algorithm is applied using the established image dataset. Currently, common object detection algorithms are mainly divided into two paradigms: anchorless and anchored. Anchorless methods do not rely on predefined anchor boxes and directly predict the key points, center points, or bounding boxes of the target. Anchored methods use predefined anchor boxes as references, matching positive and negative samples based on these anchor boxes and performing target position and size regression. An anchorless neural network structure or adjusting the aspect ratio of the predefined anchor boxes in an anchored neural network structure to adapt to the rebar shape can be used to train a rebar recognition model.

[0107] Specifically, because steel bars are slender, the bounding boxes often exhibit a relatively large aspect ratio. Anchorless target detection networks (such as FCOS, CenterNet, YOLOv8, and YOLOv11) avoid setting predefined anchor boxes and directly predict the center point and size of the target, allowing them to flexibly adapt to different shapes. Anchored target detection networks (such as Faster R-CNN, RetinaNet, YOLOv5, and YOLOv7) use a series of predefined anchor boxes and perform regression by matching predicted boxes with anchor boxes during training. For the characteristics of steel bars, the aspect ratio of the predefined anchor boxes can be adjusted. K-Means clustering algorithm can be used, with IoU as the distance metric, to automatically generate adaptive anchor boxes based on the bounding boxes in the training set, in order to better adapt to the slender shape of steel bars.

[0108] In step S6, during on-site testing, an auxiliary calibration plate is vertically erected at a certain height near the center layer of the prefabricated shear wall edge component, such as... Figure 5 As shown, record the height from the floor. h 0 Because the geometric characteristics of the edge components of prefabricated shear walls are narrow and tall (typically about 0.5m wide and 3m high), they are designed with three viewing angles. The camera module consists of three cameras arranged from top to bottom, each with an elevation angle. Eye level (0 degrees of elevation), downward angle Angle of elevation and angle of depression No more than One or more auxiliary calibration plates are installed on the side of the shear wall. By adjusting the shooting distance, the three cameras sequentially capture images of the upper, middle, and lower sections of the complete prefabricated shear wall edge component. When multiple calibration plates are used, at least one calibration plate must be captured in each of the three sections. When a single calibration plate is used, the middle section must capture the calibration plate. Figures 2-4 As shown. The angle between the camera's imaging plane and the calibration plate plane should not exceed 25°.

[0109] In step S7, for geometric dimensional information involving actual geometric length measurement, such as the spacing of transverse reinforcing bars, lap length, and position, the sub-pixel corner coordinates of the calibration plate are statistically analyzed from the images captured by the camera, and the pixel accuracy is determined. The normalized perspective transformation matrix is ​​calculated using four sets of matching points. The shooting angle of the corresponding image is corrected by applying a standardized perspective transformation matrix. The projection point of the rebar on the center layer of the original image is calculated by using the detection box coordinates inferred by the rebar detection model in step S5. The geometric dimensions between the rebars can be obtained by the corrected pixel coordinates and pixel precision.

[0110] The sub-pixel coordinates of corner points are calculated using an OpenCV corner detection algorithm based on an n×n calibration board in the mid-section image. The actual distance of the corner points is then calculated. Pixel distance is Then the pixel precision is:

[0111]

[0112] The selection rules for the four sets of matching points are as follows:

[0113] (1) Take the point closest to the image center among the four calibration plate edge corner points as the base point. Select the remaining three points in a clockwise direction. ;

[0114] (2) From the four standard direction vectors Find with Standard direction vector with the smallest included angle ;

[0115] (3) Take , for The unit standard direction vector rotated 90° clockwise.

[0116] (4) with , The four corresponding matching points are as follows:

[0117]

[0118] Furthermore, through four sets of pixel matching points Based on the principle of image homography, the normalized perspective transformation matrix is ​​calculated. If a single calibration board is used before shooting, then the elevation angle... Normalized perspective transformation matrix of the camera and angle of depression Normalized perspective transformation matrix of the camera They are respectively:

[0119]

[0120]

[0121] The normalized perspective transformation matrix is ​​applied to the corresponding image so that the calibration plate plane is parallel to the imaging plane and the calibration plate is corrected into a standard rectangle. After perspective distortion correction, the pixel ratio of points on the central plane in the image remains unchanged.

[0122] For measuring the spacing of transverse reinforcing bars, since the inference results of the neural network model include the xywh detection box information of the reinforcing bars, and the center of the detection box approximately falls on the center layer, the projection point of the transverse reinforcing bars on the center layer in the original image is... Approximate points within the detection frame, such as... Figure 6 As shown.

[0123] After standardizing perspective transformation and correcting the original image, the projection points of the central layer are... The new coordinates in the image after standardized perspective transformation are: Any two objects i , j lateral spacing between and longitudinal spacing They can be represented as follows:

[0124]

[0125]

[0126] For calculating the lap length, the boundary coordinates within the intersection area of ​​the two rebar detection frames are taken after standardized perspective transformation correction. Then the transverse reinforcement i , j The overlap length between them is The lap length between the longitudinal and longitudinal reinforcement bars is They are respectively:

[0127]

[0128]

[0129] For calculating the location of the reinforcing bars, the height of the slab above the floor will be used. Using the absolute coordinates, and then calculating the spacing between the reinforcing bars and the reference slab as the relative coordinates, the actual height of the transverse reinforcing bars from the floor surface is then determined. The calculation is as follows:

[0130]

[0131] The calculation of the actual height of the joint from the floor is divided into two cases: joints near and far from the camera. Since the joint is located on an independent longitudinal reinforcement bar, the calculation needs to be based on the distance between the longitudinal reinforcement bars. The shooting distance from the camera to the central layer and image pixel height The ordinate of the joint is corrected to the center layer, and its geometric relationship is as follows: Figure 8 As shown, the actual height of the longitudinal reinforcement joint near the camera from the floor surface. The actual height of the longitudinal reinforcement joints away from the floor from the camera. The calculation is as follows:

[0132]

[0133]

[0134]

[0135] In step S8, based on the image corrected by standardized perspective transformation, the rebar recognition model trained in step S5 is applied to statistically analyze non-geometric dimensional information such as the number of rebars, connection method, and edge member type of the prefabricated shear wall edge members. Since longitudinal bars may be obscured during shooting, if a certain longitudinal bar does not have a second longitudinal bar within a certain pixel threshold range, it is considered that the longitudinal bar is obscured, and the number of longitudinal bars needs to be increased by 1 when calculating the number of longitudinal bars.

[0136] Optionally, for statistics on non-geometric dimensions, the rebar connection method is obtained and the quantity of different types of rebar is counted based on the rebar category results inferred from the rebar detection model. Since longitudinal bars may be obstructed during image capture, 1 / 6 of the average width of the transverse rebar detection box in the image after standardized perspective transformation correction is taken as the threshold.

[0137] Furthermore, when multiple calibration plates are used during shooting, and the upper, middle, and lower images all contain the calibration plates, the calibration plates can be used as the boundaries of the detection interval. The physical spacing of adjacent transverse steel bars within the boundary and their distance to the calibration plates are calculated step by step from bottom to top. The transverse steel bars outside the boundary are not calculated, and the geometric position registration of the transverse steel bars is completed at the data level.

[0138] When a single calibration plate is used during shooting, the upper and lower image intervals do not contain the calibration plate, making it impossible to extract the corresponding information. In this case, geometrical registration of the reinforcing bars in the upper and lower image intervals is required. Overlapping areas will form between the upper / middle and middle / lower intervals, such as... Figure 7 As shown.

[0139] The proportion of the overlapping area between the upper and middle intervals in the upper interval image is:

[0140]

[0141] The proportion of the overlapping area between the middle and lower intervals in the lower interval image is:

[0142]

[0143] The middle section of the image is used to collect information on all horizontal reinforcing bars. Reinforcing bars in overlapping areas between upper and lower sections are not counted repeatedly. Assume the image height is... The detection range of the upper region image is narrowed down to pixels. The width range, the detection range of the lower interval image is narrowed to The width range is defined by using the boundary of the overlapping area as the boundary of the new upper and lower detection intervals, thereby completing the geometric position registration of the transverse reinforcement.

[0144] Furthermore, the set of reinforcement information calculated within any interval At the data level, the information from the three intervals is integrated to obtain the entire set of detection information for the shear wall. for .

[0145] In step S9, Building Information Modeling (BIM) is a technology and method based on digital 3D models used to create, manage, and share physical and functional characteristic information of buildings or infrastructure throughout the entire lifecycle of a building project. It helps optimize resource allocation, reduce costs, and contribute to sustainable building development and intelligent management. A parametric virtual model is established on the BIM platform based on the reinforcement design information of the prefabricated shear wall edge members. A parametric solid model is then established on the BIM platform based on the geometric dimension information obtained in step S7 and the non-geometric dimension information obtained in step 8.

[0146] Optionally, the design information should include the seismic grade, the form of the prefabricated shear wall edge members, the longitudinal reinforcement method, the story height, the connection method and quantity of the steel bars (longitudinal and transverse reinforcement); the lap length and location of the steel bars, the spacing of the transverse reinforcement, etc.

[0147] Write the interface for parametric modeling on the BIM platform, and define model classes for wall extension reinforcement, longitudinal reinforcement and connection joints, stirrups, additional connection reinforcement, tie bars, etc.

[0148] The design information and the actual detected information should share the same model class. A virtual model should be built based on the relevant parameters of the design information, and a solid model should be built based on the relevant parameters of the actual detected information.

[0149] In step S10, the virtual model from step S9 is compared with the solid model to check whether the connection method, quantity, lap length, position, and spacing of the reinforcing bars (longitudinal and transverse bars) meet the specifications.

[0150] Optionally, referring to Article 5.5.3 of the "Code for Acceptance of Construction Quality of Concrete Structures" GB 50204-2015, which specifies the allowable deviation for reinforcement installation, the virtual model and the solid model can be used to determine whether the construction of the edge components of the prefabricated shear wall is qualified according to the rules in Table 2 below.

[0151] Table 2. Rules for comparing and detecting virtual models and solid models.

[0152]

[0153] Note: In the table above, e represents the error generated by the image measurement method proposed in this application.

[0154] This embodiment also provides a rebar monitoring device for implementing the above-described method for detecting rebar in prefabricated shear wall edge members, comprising:

[0155] The camera module includes a low-angle camera, a level-up camera, a high-angle camera, and a device housing. The low-angle camera is mounted on the device housing via an angle adjustment assembly and is equipped with an angle adjustment knob. The level-up camera is fixed to the device housing. The high-angle camera is mounted on the device housing via a tilt adjustment assembly and is equipped with a tilt adjustment knob.

[0156] The auxiliary calibration plate includes an interconnected spherical universal support and a calibration plate. The calibration plate is connected to a bubble level. The spherical universal support is vertically mounted on the center layer of the prefabricated shear wall edge component via a snap fastener.

[0157] The image processing and visualization module connects to the camera module and is used for data transmission and execution of calculations for rebar detection methods.

[0158] Example 2

[0159] This embodiment proposes a deep learning model training method for intelligent detection of reinforcing bars in edge members of prefabricated shear walls, applicable to the scheme of Embodiment 1, including the following steps:

[0160] S1: Data acquisition. For the reinforcing bars of the prefabricated shear wall edge components that have not yet been poured with concrete, an industrial camera with an automatically adjustable focal length is used to acquire images of the reinforcing bars from multiple angles, including the front and side, within a usage scenario of 1-3m.

[0161] S2: Dataset creation involves annotating the images of the reinforcing bars of the prefabricated shear wall edge components with their xywh coordinates and corresponding categories according to the minimum bounding rectangle principle, and storing them as a txt annotation file with the same name as the image.

[0162] S3: Data preprocessing, preprocessing the images, and standardizing the size of all images.

[0163] S4: Data augmentation. Data augmentation methods such as flipping, brightness adjustment, and contrast adjustment are used to broaden the training set data, balance the number of training images of different categories, and improve the robustness of the deep learning model.

[0164] S5: Model training involves feature extraction and learning from the preprocessed images. For the task of detecting rebar in the edge components of prefabricated shear walls, an anchorless neural network structure can be used; alternatively, an anchored neural network structure can be used, and the aspect ratio of the rebar bounding boxes in the dataset can be statistically analyzed to adjust the shape of the predefined anchor boxes. For example, the YOLOv11 object detection network structure can be used. As an anchorless neural network structure, YOLOv11 employs a lightweight design in its classification head, using depthwise separable convolutional layers instead of traditional convolutional layers, thereby reducing computational complexity and the number of parameters. Its backbone and neck architecture, combined with a position-sensitive attention mechanism, enhances feature extraction capabilities, achieving performance breakthroughs through architectural innovation and training optimization. YOLOv11, with its high efficiency, low latency, and multi-task adaptability, has become a leading solution in the field of computer vision. After model training, it can predict the xywh coordinates of the detection boxes for rebar in the image and their corresponding categories.

[0165] Example 3

[0166] like Figures 8-11 As shown, this embodiment provides an intelligent detection device for the reinforcing bars of the edge components of a prefabricated shear wall used in Embodiment 1, including: a main body of the device shell 1; an industrial camera with automatic focusing from a low angle 2; an industrial camera with automatic focusing from a horizontal angle 3; an industrial camera with automatic focusing from a high angle 4; an elevation adjustment knob 5; a depression adjustment knob 6; a shoulder strap hole 7; a small light bulb 8; a data cable expansion dock 9; a data connection hole 10; an image processing and visualization module 11; a longitudinal reinforcement bayonet 12; a sliding guide rail 13; a spherical universal support 14; a calibration plate 15; and a bubble level 16.

[0167] Among them, the small light bulb 8 can be used to provide supplemental lighting in low-light shooting environments; the autofocus industrial camera group (2,3,4) is used to shoot the steel reinforcement of the edge components of prefabricated shear walls. 3 has no tilt angle, while 2 and 4 are connected to a gimbal. The tilt angle adjustment knob 5 and the pitch angle adjustment knob 6 adjust and read the tilt angle respectively. and angle of depression The autofocus industrial camera group (2,3,4) can be connected to a USB cable to a data cable expansion dock 9 and transmit data to the image processing and visualization module 11 through the data connection hole 10. The main body 1 of the equipment casing has a pre-drilled shoulder strap hole 7, allowing the entire device to be hung on the chest by attaching a shoulder strap to safety clothing. The guide rail 13 and the longitudinal rib clamp 12 are adjustable in length so that the buckle can engage the longitudinal rib. The calibration plate 15 is placed on the central layer near the edge component of the prefabricated shear wall, and its balance can be adjusted via the spherical universal support 14 and the bubble level 16.

[0168] Optionally, if multiple calibration plates 15 are used during testing, they should be placed as evenly as possible on the sides of the prefabricated shear wall edge members to ensure that the three divided sections are comparable. Assume the height of the shear wall edge members is... HThe first calibration plate, from bottom to top, is set at 0-1 / 3 of its height, depending on the site conditions. H Within the specified range, the last calibration plate is set at 1 / 3 of the height of the shear wall edge member, depending on the site conditions. H ~ H Within the range; if a single calibration plate 15 is used, it should be set at 1 / 3. H ~2 / 3 H Within the range.

[0169] Furthermore, the autofocus industrial camera 2 adjusts its tilt angle upwards, and the autofocus industrial camera 4 adjusts its tilt angle downwards. The autofocus industrial camera group (2,3,4) can capture images of the upper, middle, and lower detection zones simultaneously.

[0170] Furthermore, the equipment is intended for use at a distance of 1-3m from the edge components of the prefabricated shear wall.

[0171] Preferably, the entire device can be suspended in front of the testing personnel by threading a strap through the strap hole 7, thereby improving the convenience of on-site testing.

[0172] Example 4

[0173] This embodiment provides a field inspection system for intelligent detection of reinforcing bars in the edge components of prefabricated shear walls based on Embodiments 2 and 3, which consists of the following modules:

[0174] The data acquisition module inputs the shear wall reinforcement design information for the inspection items before testing, including seismic grade, type of prefabricated shear wall edge members, longitudinal reinforcement method, story height, reinforcement (longitudinal and transverse) connection method and quantity; reinforcement lap length, location, and transverse reinforcement spacing, etc. The auxiliary calibration plate from Example 3 is vertically erected at the center of the longitudinal reinforcement at a certain height using clips, and its balance is adjusted. The elevation angle of the autofocus industrial camera group from Example 3 is adjusted and recorded. and angle of depression Angle of elevation and angle of depression Not exceeding the range of visible angles This allows for a specific shooting angle, ensuring that the entire shear wall is captured in its entirety. An autofocus industrial camera system is used to acquire image data from three sections of the shear wall's reinforcing steel reinforcement in a single shot.

[0175] The deep learning model inference module infers the collected data using the deep learning model trained in Example 2 to obtain the rebar type and corresponding xywh coordinates in each of the upper, middle and lower intervals.

[0176] The geometric analysis post-processing module parses sub-pixel corner coordinates, calculates the standardized perspective transformation matrix of the three cameras (top, middle, and bottom) using four sets of matching points, applies the perspective transformation matrix to correct the shooting angle of the corresponding image, and calculates the geometric and non-geometric dimensions between the reinforcing bars through deep learning model inference results. Finally, it integrates the information at the data level based on the three continuous shooting intervals (top, middle, and bottom).

[0177] The parametric modeling module inputs design information and actual detected rebar information into the BIM platform's parametric modeling interface, respectively. It then builds a virtual model based on the relevant parameters of the design information and a solid model based on the relevant parameters of the actual detection.

[0178] The results evaluation module performs feature matching, geometric comparison, and deviation comparison on the virtual and solid models, and outputs the evaluation results based on certain evaluation rules.

[0179] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for detecting the reinforcing steel bars in the edge members of a prefabricated shear wall, characterized in that: Images of the reinforcing bars of the edge components of the prefabricated shear wall are captured from multiple angles using a camera; the captured images of the reinforcing bars are preprocessed and data augmented to generate diverse training samples, and the training samples are labeled with the reinforcing bar regions and categories to establish a target detection dataset; the target detection dataset is used to train a neural network-based target detection algorithm to obtain a reinforcing bar recognition model. During on-site testing, the auxiliary calibration plate is vertically erected at the center layer of the prefabricated shear wall edge component. The center layer is the virtual plane where the center of the steel cage of the prefabricated shear wall edge component is located, and it is parallel to the wall surface. Before taking pictures, the auxiliary calibration plate is fixed on the center layer outside the steel cage, and the auxiliary calibration plate is adjusted to keep it vertical to the floor. The height h0 of the center point of the calibration plate from the floor is recorded at this time. A camera module is used to capture images of multiple sections of the edge components of the prefabricated shear wall, and the captured images include the auxiliary calibration plate to obtain a detection image; Based on the detected image, the sub-pixel corner coordinates and pixel precision ε of the auxiliary calibration board are analyzed, and the standardized perspective transformation matrix is ​​calculated to correct the shooting angle of the detected image. For the corrected detection image, the detection frame coordinates and non-geometric dimension information of the steel bars are identified using a steel bar recognition model; based on the detection frame coordinates of the steel bars, the projection points of each steel bar on the central layer are determined, and the geometric dimension information between each steel bar is determined based on the pixel coordinates and pixel precision of the projection points. The geometric dimensional information includes the transverse and longitudinal spacing between reinforcing bars, lap length, reinforcing bar position, and joint position. The calculation expressions for the transverse and longitudinal spacing are as follows: In the formula, Δs x (i,j) represents the transverse spacing between steel bars i and j, Δs y (i,j) represents the longitudinal spacing between reinforcement bars i and j, (u m ,v m () represents the coordinates of the projected points after standardized perspective transformation correction; The lap length includes the lap length between transverse reinforcing bars and the lap length between longitudinal reinforcing bars. The calculation expressions for the lap length between transverse reinforcing bars and the lap length between longitudinal reinforcing bars are as follows: In the formula, Δd x (i,j) represents the lap length between transverse reinforcing bars i and j, Δd y (i,j) represents the lap length between longitudinal reinforcement bars i and j, (u d ,v d () represents the boundary coordinates within the intersection area of ​​the two rebar detection frames after standardized perspective transformation correction; The location of the reinforcing bars refers to the actual height of the transverse reinforcing bars from the floor surface. The expression for calculating the actual height of the transverse reinforcing bars from the floor surface is as follows: In the formula, h i h0 is the actual height of the transverse reinforcement i from the floor surface, and h0 is the actual height of the center point of the calibration slab from the floor surface. The ordinate of the center point of the calibration board after standardized perspective transformation correction; The joint location includes the actual height of the longitudinal reinforcement joints near and away from the camera from the floor. Since the joint is located on an independent longitudinal reinforcement, the longitudinal coordinate of the joint needs to be corrected to the center layer. The expression for calculating the actual height of the longitudinal reinforcement joints near and away from the camera from the floor is as follows: In the formula, The actual height of the longitudinal reinforcement joint near the camera from the floor surface. v represents the actual height of the longitudinal reinforcement joint away from the floor, away from the camera. c Points in the longitudinal reinforcement joint inspection frame after standardized perspective transformation correction. Correct the longitudinal coordinates of the longitudinal rib joints near the camera to the center layer. To correct the longitudinal coordinates of the longitudinal rib joints away from the camera to the center layer, d y The distance between the longitudinal ribs closest to and furthest from the camera, g is the shooting distance from the camera to the center layer, and H is the distance between the longitudinal ribs closest to and furthest from the camera. image Image pixel height; The geometric and non-geometric dimensional information of the reinforcing bars are compared with the design information of the edge components of the prefabricated shear wall to check whether the reinforcing bars meet the specifications.

2. The method for detecting reinforcing steel bars in the edge members of a prefabricated shear wall according to claim 1, characterized in that, Based on the auxiliary calibration board in the detected image, the OpenCV corner detection algorithm is used to calculate the sub-pixel corner coordinates, obtaining the actual distance d0 and pixel distance d of the corner. i,j ; The expression for calculating the pixel precision is: In the formula, ε represents the pixel precision, and n×n represents the specifications of the auxiliary calibration board.

3. The method for detecting reinforcing bars in the edge members of a prefabricated shear wall according to claim 2, characterized in that, Based on the parsed sub-pixel corner coordinates, the standardized perspective transformation matrix is ​​calculated using four sets of matching points. The selection rules for the four sets of matching points are as follows: Take the calibration plate edge corner point closest to the center of the detection image as the base point p1', and select the remaining three points p2', p3', and p4' in a clockwise direction; Find the vector that is parallel to the four standard direction vectors. Standard direction vector with the smallest included angle and take The coordinates of the four selected matching points are as follows: In the formula, p1 and p1' are the first matching points, p2 and p2' are the second matching points, p3 and p2' are the third matching points, and p4 and p4' are the fourth matching points; Based on the four sets of matching points p i '→p i Based on the principle of image homography, the standardized perspective transformation matrix is ​​calculated.

4. The method for detecting reinforcing bars in the edge members of a prefabricated shear wall according to claim 1, characterized in that, The steel reinforcement in the prefabricated shear wall edge components photographed from multiple angles includes: Images taken from the front and side of the edge components of the prefabricated shear wall; The labeling of the reinforcement area and category includes: For images taken from the front, the coordinates and categories of the reinforcing bars are labeled according to the principle of minimum bounding rectangle. For images taken from the side, the longitudinal bars and joints are labeled.

5. The method for detecting reinforcing bars in the edge members of a prefabricated shear wall according to claim 1, characterized in that, The preprocessing includes cutting and size standardization operations; The data enhancements include flipping, brightness adjustment, and contrast adjustment.

6. The method for detecting reinforcing bars in the edge members of a prefabricated shear wall according to claim 1, characterized in that, The target detection algorithm is an anchorless neural network structure; Alternatively, an anchored neural network structure can be used, and the aspect ratio of the predefined anchor frames in the anchored neural network structure can be adjusted to adapt to the shape of the reinforcing bars.

7. The method for detecting reinforcing bars in the edge members of a prefabricated shear wall according to claim 1, characterized in that, The camera module has cameras with three viewing angles, which can capture multiple sections of the edge components of the prefabricated shear wall at one time. The multiple viewing angles include an elevation angle, a level view, and a depression angle. The auxiliary calibration plate can be a single piece or multiple pieces. Before the camera module takes a picture, the shooting distance is adjusted within the range of 1-3m so that the camera module can sequentially capture the upper, middle and lower sections of the complete prefabricated shear wall edge component from each viewing angle. The steel bars captured in the three sections are allowed to overlap, but no omissions are allowed, and the images must be clear and the geometric information complete. When there are multiple auxiliary calibration boards, at least one of the upper, middle, and lower sections of the image is included; when there is only one auxiliary calibration board, the middle section of the image is included. Information is integrated from the captured images of the three intervals. When there are multiple auxiliary calibration plates, the position of the auxiliary calibration plate in each interval is used as the boundary of the detection interval information. When there is only one auxiliary calibration plate, the boundary of the overlapping area is used as the boundary of the detection interval information.

8. The method for detecting reinforcing bars in the edge members of a prefabricated shear wall according to claim 1, characterized in that, The non-geometric dimension information includes the number of steel bars, connection method, and edge member type of the prefabricated shear wall edge members. The statistics are obtained by keyword matching of the inference results of the deep learning model. The longitudinal bars that may be obscured need to be compensated for in quantity. The process of comparing the geometric and non-geometric dimensional information of the reinforcing bars with the design information of the edge components of the prefabricated shear wall is as follows: Based on the design information of the prefabricated shear wall edge components, a parametric virtual model of the reinforcing steel bars of the prefabricated shear wall edge components is established; A parametric solid model is established based on the detected geometric and non-geometric dimensional information of the reinforcing bars; The parametric virtual model is compared with the parametric solid model to check whether the steel reinforcement meets the specifications.

9. A reinforcement monitoring device for implementing the reinforcement detection method for the edge members of a prefabricated shear wall as described in any one of claims 1-8, characterized in that, include: A camera module includes a low-angle camera, a level-up camera, a high-angle camera, and a device housing. The low-angle camera is mounted on the device housing via an angle adjustment assembly and is equipped with an angle adjustment knob. The level-up camera is fixed to the device housing. The high-angle camera is mounted on the device housing via a tilt adjustment assembly and is equipped with a tilt adjustment knob. The auxiliary calibration plate includes an interconnected spherical universal support and a calibration plate. The calibration plate is connected to a bubble level. The spherical universal support is vertically mounted on the center layer of the prefabricated shear wall edge component via a snap fastener. The image processing and visualization module is connected to the camera module and is used for data transmission and execution of the calculation process of the rebar detection method.

Images