A Precision Detection Method for Concealed Engineering of Prefabricated Components

Abstract

The present invention belongs to the field of concealed engineering detection, and involves the application of laser precision detection technology to realize non-destructive detection of concealed engineering, and specifically provides a method for precision detection of concealed engineering of prefabricated components. The method first pre-processes the photographed image, and automatically generates a matching template to match the on-site environment based on the industrial drawing information of the superimposed plate layer. The position of the prefabricated component is preliminarily detected based on the gradient information of the image and the feature extraction algorithm. Next, the prefabricated components in the superimposed plate are matched and positioned for precise positioning. The superimposed plate layer is precisely scanned with an anti-interference laser, and the prefabricated components are segmented according to the grayscale and the expected position. The final detection result is obtained by combining the detection results of the aforementioned steps. This method can improve the accuracy and coverage of hidden detection and reduce the burden on quality inspectors.

Description

Technical Field

[0001] The invention belongs to the field of concealed engineering detection, relates to the application of laser precision detection technology to realize non-destructive detection of concealed engineering, and specifically provides a method for precise detection of concealed engineering of prefabricated components. Background Art

[0002] Hidden engineering inspection refers to the process of inspecting and evaluating those parts of the project that are inside the structure and difficult to observe and access for quality inspection. Generally, the molds, embedded parts, steel bars and trusses are measured manually, and the position deviation and quantity are checked by comparing with the drawings, and finally the quality inspection report of the inspection parts is written manually. However, manual output will lead to some quality inspection problems, such as focusing on finished product inspection and neglecting hidden inspection, as well as inaccurate manual measurement, missed inspection, and wrong inspection.

[0003] For the existing solutions in the field of concealed engineering inspection, 3D laser scanning is already used. However, during the 3D laser scanning process, the acquisition of point cloud data is often affected by factors such as object occlusion and uneven lighting, which can easily cause blind spots in the scanning of complex-shaped objects and form holes. At the same time, due to the limited scanning measurement range, large-sized objects or large-scale scenes cannot be fully measured at one time, and multiple scanning measurements are required. Therefore, the scanning results are often multiple pieces of point cloud data with different coordinate systems and noise, which cannot fully meet people's requirements for the authenticity and real-time performance of digital models.

[0004] Therefore, the above problems need to be solved urgently. Summary of the invention

[0005] Purpose of the invention: In order to overcome the above shortcomings, the purpose of the present invention is to provide a method for accurate detection of concealed works of prefabricated components, which mainly uses anti-interference laser linear array scanning and is supplemented by cameras to identify the features of the object being measured, solve the problems of missed detection and non-standard detection in the existing concealed engineering detection using manual measurement, improve the accuracy and coverage of concealed detection, and reduce the burden on quality inspectors. And the system automatically records quality inspection records, so that the quality of components can be effectively traced.

[0006] Technical solution: In order to achieve the above-mentioned purpose, the present invention provides a method for accurately detecting concealed engineering of prefabricated components, comprising:

[0007] Step S10: Acquire the laminated board layer image and perform preprocessing to improve the image contrast;

[0008] Step S11: obtaining a template matching the on-site environment according to the drawing information, and using the template to match the laminated board image preprocessed in step S10;

[0009] Step S12: collecting the gradient information of the image after matching in step S11, and using a feature extraction algorithm to preliminarily detect the position of the prefabricated component;

[0010] Step S13: extracting the geometric shape of the prefabricated component in the drawing, matching the prefabricated component features in the drawing with the prefabricated component features obtained in step S12, and achieving accurate matching and positioning of the prefabricated component;

[0011] Step S14: using laser scanning to obtain point cloud data of the laminated plate layer, retaining the endpoint and intersection features in the point cloud, processing the point cloud data, and accurately locating the prefabricated components;

[0012] Step S15: Input the point cloud data processed in step S14 and the prefabricated component position information and feature extraction results in the plane image into the quality inspection system. The quality inspection system compares the geometric shape of the prefabricated components and the precise position of the prefabricated components in the drawing in step S13 with the input information, and outputs the detection results of the characteristics, structure and prefabricated components inside the composite plate layer.

[0013] Based on the gradient information and feature extraction algorithm of the image, the prefabricated components in the laminated plate layer can be accurately identified, and the quantity, type, size and spatial position of the prefabricated components can be accurately identified.

[0014] Furthermore, in the above-mentioned method for precise detection of concealed works of prefabricated components, in step S14, the point cloud data processing flow includes:

[0015] Step S20: Acquire point cloud data of the laminated board layer from a laser scanner;

[0016] Step S21: pre-processing the point cloud data to remove noise and outliers;

[0017] Step S22: extracting endpoints and intersections from the point cloud processed in step S21 as features, using a clustering algorithm to segment the prefabricated components, and retaining the endpoints and intersections of the prefabricated components as features after segmentation for subsequent matching;

[0018] Step S23: reconstructing the point cloud surface of the point cloud data preprocessed in step S21 to facilitate visualization analysis;

[0019] Step S24: performing feature matching and target detection, matching the prefabricated component features extracted from step S22 with the geometric features of the endpoints and intersections of the prefabricated components in the corresponding drawings, and accurately locating the positions of the prefabricated components;

[0020] Step S25: Perform visualization processing on the point cloud data.

[0021] The laser precision detection technology is used to provide an efficient point cloud processing algorithm, which can effectively process tens of millions of point cloud data, improve the accuracy and coverage of hidden detection, and solve the problems of missed detection and irregular detection. Finally, on the premise of maintaining the point cloud features, the point cloud data is simplified to reduce the complexity of calculation and processing, and realize the three-dimensional reconstruction and precise positioning of the superimposed plate layers.

[0022] Furthermore, in the above-mentioned method for precise detection of concealed works of prefabricated components, the operation of preprocessing the image in step S10 is specifically to convert it into a grayscale image using the cv2.cvtColor function and cv2.GaussianBlur, and to apply Gaussian blur using cv2.GaussianBlur to reduce noise.

[0023] Furthermore, in the above-mentioned method for precise detection of concealed engineering of prefabricated components, the specific method for preliminary detection of prefabricated components in step S12 is a dual threshold detection algorithm, and two thresholds are set, a threshold A and a threshold B. Pixels with a gradient amplitude greater than threshold A are marked as strong edges, and pixels with a gradient amplitude between threshold B and threshold A are marked as weak edges. By connecting strong edge pixels, weak edge pixels are converted into strong edge pixels to form a continuous edge, and preliminary positioning of prefabricated components is performed.

[0024] Furthermore, in the above-mentioned method for precise detection of concealed works of prefabricated components, the specific method for accurate matching and positioning in step S13 is to read the image of the prefabricated component in the drawing, match the features of the prefabricated component with the features of the prefabricated component in the image, eliminate erroneous matching results, and manually perform fine-tuning based on the matching results.

[0025] Furthermore, in the above-mentioned method for precise detection of concealed works of prefabricated components, the method for removing noise and outliers in step S21 includes using the remove_noise function to perform noise reduction processing, and using the voxel_down_sample function of the point cloud to perform voxel down-sampling on the point cloud to reduce the number of points and noise.

[0026] Furthermore, in the above-mentioned method for precise detection of concealed works of prefabricated components, the specific method for segmenting the prefabricated components in step S22 is to divide the points in the point cloud into different clusters, exclude clusters that do not conform to the geometric properties of the prefabricated components, and thus separate the points of the prefabricated components from other noise points and background points.

[0027] Furthermore, in the above-mentioned method for precise detection of concealed works of prefabricated components, the specific method for cloud surface reconstruction in step S23 is to voxelize the preprocessed point cloud data, map the point cloud data to a three-dimensional grid space, store the information of the point cloud data in each voxel, calculate the normal of each voxel, and use the Poisson equation to reconstruct the surface.

[0028] Furthermore, in the above-mentioned method for precise detection of concealed works of prefabricated components, in step S25, the visualization method is to use the visualization.draw_geometries function of the Open3D library to display the input point cloud object, and the function automatically draws the point cloud in the graphical interface.

[0029] Furthermore, in the above-mentioned method for accurate detection of concealed works of prefabricated components, the detection results output in step S15 can be input into the production management information system, which generates a record sheet and archives it by comparing the detection results with the production indicators. The introduction of an intelligent detection system realizes the linkage between the input of detection indicators and the output of quality inspection results, thereby improving the efficiency of quality management of concealed works.

[0030] It can be seen from the above technical scheme that the present invention has the following beneficial effects: the method for accurately detecting hidden engineering of prefabricated components of the present invention adopts laser precision detection technology, provides an efficient point cloud processing algorithm, can effectively process tens of millions of point cloud data, improve the accuracy and coverage of hidden detection, and solve the problems of missed detection and non-standard detection; according to the gradient information and feature extraction algorithm of the image, the prefabricated components in the laminated plate layer are accurately identified, and the number, type, size and spatial position of the prefabricated components can be accurately identified; plane vision detection is used to assist laser precision detection. It is used to solve the problem that thin-walled embedded parts or highly reflective objects are prone to weak detection in laser scanning; according to the current standards, it covers the detection items of hidden engineering side molds, steel bars, and embedded parts. Visual processing is performed on different detection parts to facilitate the detection of quality inspectors, so that one quality inspector can be responsible for multiple shifts and multiple lines, improve detection efficiency, and reduce the load of quality inspectors; through data-driven, the production data of BIM prefabricated components is analyzed, and the target value of the detection item is obtained. Compared with the traditional EXCEL table, the data is relatively complete, reducing the problem of manual retrieval of drawings and secondary entry, which is easy to make mistakes. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Schematic diagram of the detection method flow chart;

[0032] Figure 2 This is the point cloud data processing flow chart;

[0033] Figure 3 This is a flow chart of the quality inspection system. DETAILED DESCRIPTION

[0034] like Figure 1The detection method flow shown in the figure first uses a planar vision camera to collect images of the stacked board layers, uses cv2.cvtColor function and cv2.GaussianBlur function to convert the image into a grayscale image, and further uses cv2.GaussianBlur function to apply Gaussian blur to reduce noise.

[0035] Based on the CAD drawing information of the laminated board layer, a matching template suitable for the on-site environment is generated. This matching template is used for image matching of the actual on-site environment.

[0036] According to the gradient information of the matched image, the feature extraction algorithm is used to preliminarily detect the position of the steel bars. The specific feature extraction algorithm is the Canny edge detection algorithm, which is implemented using the cv2.Canny function in the opencv library. The basic derivation steps of the Canny edge detection algorithm are as follows:

[0037] The Sobel operator is used to calculate the gradient magnitude and direction of each pixel in the image.

[0038] Assume that the input image is I(x,y), and the horizontal and vertical components of the gradient are G x (x,y) and G y (x,y), the gradient amplitude is G(x,y), and the gradient direction is θ(x,y). The gradient calculation formula is as follows:

[0039]

[0040]

[0041]

[0042]

[0043] I(x,y): pixel value of the input image.

[0044] G x (x,y): The gradient component of the image in the x direction.

[0045] G y (x,y): The gradient component of the image in the y direction.

[0046] G(x,y): The gradient magnitude of the image.

[0047] θ(x,y): The gradient direction of the image.

[0048] Using the non-maximum suppression algorithm, for each pixel, check the gradient magnitude of its two adjacent pixels in the gradient direction. If the gradient magnitude of the current pixel is the largest between the gradient magnitudes of the two adjacent pixels, then retain the pixel, otherwise suppress it to 0. The algorithm derivation process is as follows:

[0049] 1. For each pixel point (x, y), calculate its gradient magnitude G(x, y) and gradient direction θ(x, y);

[0050] 2. Discretize the gradient direction θ(x,y) and divide it into four directions;

[0051] 3. According to the discretized gradient direction, determine the two adjacent pixels of the current pixel in the gradient direction;

[0052] 4. For each direction, determine whether the gradient magnitude of the current pixel is the local maximum among its neighboring pixels. If so, retain the gradient magnitude of the current pixel, otherwise suppress it to 0. For example, for the horizontal direction, compare the gradient magnitude G(x,y) of the current pixel (x,y) with the gradient magnitudes of the neighboring pixels (x+1,y) and (x-1,y). If G(x,y) is the largest of these two gradient magnitudes, retain G(x,y), otherwise suppress it to 0.

[0053] 5. Repeat steps 3 and 4 to judge all four directions and obtain the final edge map.

[0054] A dual threshold detection algorithm is used, and two thresholds are set, a high threshold and a low threshold. The high threshold is set to 90% of the gradient intensity, and the low threshold is set to half of the high threshold. Pixels with gradient magnitude greater than the high threshold are marked as strong edges, and pixels with gradient magnitude between the low threshold and the high threshold are marked as weak edges. Then, by connecting the strong edge pixels, the weak edge pixels are converted into strong edge pixels to form a continuous edge.

[0055] After the initial positioning is completed, the geometric shape of the steel bars extracted from the CAD drawing is used to further match and locate the steel bars in the composite slab. First, use the cv2.imread function to read the steel bar image automatically generated in the CAD project, and then use the cv2.ORB_create function to create the ORB (Oriented FAST and Rotated BRIEF) feature detector and descriptor extractor function. Use a brute force matcher to perform feature matching on the steel bars in the drawing, and use the cv2.BFMatcher function to match the steel bar features in the drawing with the actual steel bar features. Use the cv2.findHomography function to calculate the optimal single mapping transformation matrix between multiple two-dimensional point pairs. To achieve the matching of the two-dimensional points of the steel bar image with the prefabricated steel bar points.

[0056]

[0057] The function also uses the RANSAC algorithm to exclude false matches. According to the confidence of the required matching model, the RANSAC threshold is set from 2 to 5. If you want the estimated model to have high accuracy, choose a smaller threshold; if the scene of the detected object is complex, a larger threshold may be required. Finally, based on the matching results of the steel bars, manual fine-tuning is performed to achieve accurate positioning.

[0058] On the basis of precise positioning by plane vision camera-assisted technology, anti-interference laser precision detection is used for further precise positioning. Anti-interference laser scanning technology is used to precisely scan the laminated board layers. After scanning, the tens of millions of point cloud data are processed. First, point cloud splicing is performed to splice the point cloud data of multiple laser scans to form an overall point cloud model. Then the spliced ​​point cloud model is denoised to remove unnecessary noise and abnormal points. On the premise of maintaining the point cloud features, the point cloud data is processed to reduce the complexity of calculation and processing.

[0059] The specific processing flow of point cloud data is as follows: Figure 2 As shown, the first step is to load the point cloud data obtained from the laser scanner and use the "np.loadtxt" function to read the point cloud data.

[0060] In the second step, the point cloud data is preprocessed and filtered to remove noise and outliers. The remove_noise function of the Open3D library is used for denoising, which is based on the nearest neighbor noise method (FAST). The point cloud is downsampled using the voxel_down_sample function of the point cloud to reduce the number of points and noise.

[0061] The third step is to extract specific types of features from the point cloud and use clustering algorithms to segment the steel bars. The steel bars usually exist in the form of lines or strips in the point cloud, and the points in the point cloud are divided into different clusters. The steel bar points can be separated from other noise points and background points by excluding clusters that do not conform to the geometric properties of the steel bars. After segmentation, the endpoints and intersections of the steel bars are retained as feature points for subsequent matching processes.

[0062] The fourth step is to use the Poisson method to reconstruct the point cloud surface using the preprocessed point cloud data. First, the preprocessed point cloud data is voxelized and mapped to a three-dimensional grid space. Each voxel stores the information of the point cloud data. For each voxel, its normal is calculated, which can be achieved by estimating the local normal of the point cloud within the voxel. Finally, the Poisson equation is used to reconstruct the surface. The Poisson equation describes the scalar function The basic form of the Poisson equation is as follows:

[0063]

[0064] in is the Laplace operator, which represents the function Find the sum of the second-order spatial partial derivatives.

[0065] is the objective function to be reconstructed.

[0066] ρ is the source term of the divergence, representing the normal information.

[0067] This step allows the creation of a more structured 3D model for easier visualization and analysis.

[0068] In the fifth step, the feature_matching function is used to perform feature matching and target detection. The feature points obtained in the third step are used to match the corresponding geometric features from the extracted feature points based on the geometric features of the endpoints and intersections of the steel bars, thereby determining the position of the steel bars.

[0069] The sixth step is to visualize the point cloud data and use the visualization.draw_geometries function of the Open3D library to display the passed point cloud object. The function can automatically draw the point cloud in the graphical interface, making it easier for users to view and analyze the processing results.

[0070] The position information and feature extraction results of the steel bars in the processed point cloud files and plane images are input into the quality inspection system, and the parts to be inspected are accurately located based on the geometric shape of the inspection part model in the prefabricated laminate layer. The steel bars are segmented from the image according to the grayscale and the expected position. This step involves the positioning and classification of the steel bars. The inspection results of the above steps are combined to obtain the final hidden engineering inspection results. This result includes information such as the characteristics, structure and prefabricated components inside the laminate layer.

[0071] The process of quality inspection system is as follows Figure 3As shown, through LAN transmission, the hidden engineering intelligent detection system feeds back the quality inspection results to the central control system. The system identifies and calculates the inspection items, and obtains the unqualified items by comparing the inspection piece information in the production plan with the actual position information of the inspection piece. The quality inspection results are fed back to the central control system, and the errors of the unqualified items are highlighted on the display. The central control system transfers to the production management system (MES system), and the production management system generates an inspection record sheet. The generated inspection record sheet contains the following information: uploaded to the production management information system through the hidden engineering quality inspection system, the system can obtain the production plan and design documents / project inspection indicators from the factory's existing production management system, or establish a project component database in the hidden engineering quality inspection system, and index the inspection indicators according to the production plan. Compare the test results with the indicator values ​​to generate a component hidden inspection record sheet. Finally, the quality management system transmits and archives the quality inspection data to meet the requirements of the group and regulatory authorities.

[0072] The above embodiments are exemplary, and their purpose is to illustrate the technical concept and features of the present invention so that people familiar with the technology in this field can understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for precise detection of concealed works of prefabricated components, characterized in that: include: Step S10: Acquire the laminated board layer image and perform preprocessing to improve the image contrast; Step S11: generating a template matching the on-site environment according to the drawing information, and using the template to match the laminated board image preprocessed in step S10; Step S12: collecting the gradient information of the image after matching in step S11, and using a feature extraction algorithm to preliminarily detect the position of the prefabricated component; Step S13: extracting the geometric shape of the prefabricated component in the drawing, matching the prefabricated component features in the drawing with the prefabricated component features obtained in step S12, and achieving accurate matching and positioning of the prefabricated component; Step S14: using laser scanning to obtain point cloud data of the laminated plate layer, retaining the endpoint and intersection features in the point cloud, processing the point cloud data, and accurately locating the prefabricated components; Step S15: Input the point cloud data processed in step S14 and the prefabricated component position information and feature extraction results in the plane image into the quality inspection system. The quality inspection system compares the prefabricated component geometry and the precise position of the prefabricated components in the drawing in step S13 with the input information, and outputs the features, structure and prefabricated component detection results inside the composite board layer.

2. The method for accurate detection of concealed works of prefabricated components according to claim 1 is characterized in that: In step S14, the point cloud data processing flow includes: Step S20: Acquire point cloud data of the laminated board layer from a laser scanner; Step S21: pre-processing the point cloud data to remove noise and outliers; Step S22: extracting endpoints and intersections from the point cloud processed in step S21 as features, using a clustering algorithm to segment the prefabricated components, and retaining the endpoints and intersections of the prefabricated components as features after segmentation for subsequent matching; Step S23: reconstructing the point cloud surface of the point cloud data preprocessed in step S21 to facilitate visualization analysis; Step S24: performing feature matching and target detection, matching the prefabricated component features extracted from step S22 with the geometric features of the endpoints and intersections of the prefabricated components in the corresponding drawings, and accurately locating the positions of the prefabricated components; Step S25: Perform visualization processing on the point cloud data.

3. The method for accurate detection of concealed works of prefabricated components according to claim 1 is characterized in that: The operation of preprocessing the image in step S10 is specifically to convert it into a grayscale image using the cv2.cvtColor function and cv2.GaussianBlur, and to apply Gaussian blur using cv2.GaussianBlur to reduce noise.

4. The method for accurate detection of concealed works of prefabricated components according to claim 1 is characterized in that: The specific method for preliminary detection of prefabricated components in step S12 is a dual-threshold detection algorithm, which sets two thresholds, a threshold A and a threshold B; pixels with a gradient amplitude greater than threshold A are marked as strong edges, and pixels with a gradient amplitude between threshold B and threshold A are marked as weak edges; by connecting strong edge pixels, weak edge pixels are converted into strong edge pixels to form a continuous edge, and preliminary positioning of prefabricated components is performed.

5. The method for accurate detection of concealed works of prefabricated components according to claim 1 is characterized in that: The specific method for accurate matching and positioning in step S13 is to read the image of the prefabricated component in the drawing, match the features of the prefabricated component with the features of the prefabricated component in the image, eliminate erroneous matching results, and manually perform fine-tuning based on the matching results.

6. The method for accurate detection of concealed works of prefabricated components according to claim 2 is characterized in that: The method for removing noise and outliers in step S21 includes using the remove_noise function to perform noise reduction processing, and using the voxel_down_sample function of the point cloud to perform voxel down sampling on the point cloud to reduce the number of points and noise.

7. The method for accurate detection of concealed works of prefabricated components according to claim 2 is characterized in that: The specific method of segmenting the prefabricated components in step S22 is to divide the points in the point cloud into different clusters, exclude clusters that do not conform to the geometric properties of the prefabricated components, and thus separate the points of the prefabricated components from other noise points and background points.

8. The method for accurate detection of concealed works of prefabricated components according to claim 2 is characterized in that: The specific method of cloud surface reconstruction in step S23 is to voxelize the preprocessed point cloud data, map the point cloud data to a three-dimensional grid space, store the information of the point cloud data in each voxel, calculate the normal of each voxel, and use the Poisson equation to reconstruct the surface.

9. The method for accurate detection of concealed works of prefabricated components according to claim 2 is characterized in that: In the step S25, the visualization method is to use the visualization.draw_geometries function of the Open3D library to display the input point cloud object, and the function automatically draws the point cloud in the graphical interface.

10. The method for accurate detection of concealed works of prefabricated components according to claim 1 is characterized in that: The test results output in step S15 can be input into the production management information system, and the production management information system generates a record sheet and archives it by comparing the test results with the production indicators.

Images