RGBD image-based intelligent detection method for spacing between reinforcing steel bars of prefabricated part
Through the depth camera, RGBD images are acquired and combined with semantic segmentation and clustering algorithms, the problem of time-consuming and labor-consuming traditional detection methods is solved, and the fast and accurate detection of the steel spacing of prefabricated components is achieved, which improves the intelligent management capabilities and efficiency of building safety. .
Patent Information
- Application Number
- CN202410933703.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-12
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2044-07-12
AI Technical Summary
Traditional steel bar spacing detection methods are time-consuming and labor-intensive, with low accuracy, and cannot meet the large-scale inspection requirements of prefabricated components. The existing detection technology based on pure point cloud or pure image has high accuracy but high cost, making it difficult to achieve low-cost and accurate detection.
The depth camera is used to collect RGBD images, and the steel bar pixels are segmented through the InternImage semantic segmentation network. Combined with DBSCAN clustering and KD-Tree algorithm, the steel bar point cloud data is extracted and enhanced, and the steel bar spacing is calculated to achieve automatic detection.
It realizes rapid and accurate detection of the steel bar spacing of prefabricated components, improves the intelligent management capabilities and efficiency of building safety, and reduces inspection costs.
Smart Images

Figure CN118918361A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of image processing, and in particular to an intelligent detection method for the spacing of steel bars of prefabricated components based on RGBD images. Background Art
[0002] Prefabricated components are an indispensable core component of prefabricated buildings. The steel bar spacing that meets the specifications is one of the key factors to ensure the safety and stability of the structure. In the traditional steel bar spacing detection, workers need to use rulers and tape measures for measurement. This method is not only time-consuming and labor-intensive, but also easily affected by factors such as personnel's professional skills, experience and fatigue, thus affecting the accuracy and reliability of the detection results, and cannot meet the needs of large-scale detection of prefabricated components. With the rise of prefabricated buildings, the demand for intelligent detection of steel bar spacing continues to increase, and most of the current detection technologies are based on pure point cloud or pure image detection solutions. Although the methods based on pure images or pure point clouds can effectively obtain the steel bar spacing, the pure image solution has low accuracy; and the pure point cloud solution uses a land-based 3D laser scanner for data collection, which is highly accurate but expensive. Therefore, there is an urgent need for an accurate and low-cost solution for intelligent detection of steel bar spacing.
[0003] In recent years, with the rapid development of depth camera technology, it has become one of the commonly used equipment in various size detection tasks with its advantages of low cost and high precision. The depth camera based on the ToF principle obtains the RGBD image data of the object being measured by emitting light pulses to the object being measured and detecting the reflected light pulses. The RGBD image is processed based on the camera's internal parameters to generate point cloud data, and the geometric information of the object being measured can be obtained. However, in view of the professional span influence of the current image information field and the building safety management field, few people can achieve a large-span professional technology integration. For example, although the Chinese patent CN117333501A discloses a double-layer steel bar intersection recognition method and system, it cannot accurately realize the intelligent detection of steel bar spacing. Summary of the invention
[0004] In response to the problems of the background technology, the present invention provides an intelligent detection method for the spacing between steel bars of prefabricated components based on RGBD images, which adopts depth camera technology to effectively and accurately realize the precise detection of the spacing between steel bars through RGBD images.
[0005] In order to achieve the above object, the technical solution adopted by the present invention is as follows:
[0006] An intelligent detection method for the spacing of steel bars of prefabricated components based on RGBD images, the key of which is to include the following steps:
[0007] S101: Data acquisition: using a depth camera to acquire a steel bar RGBD image of a prefabricated component, wherein the RGBD image includes an RGB image and a Depth image (depth image);
[0008] S102: semantic segmentation of steel bar pixels: using the trained InternImage semantic segmentation network to segment the steel bar pixels in the steel bar RGBD image to obtain RGB image data containing only steel bar pixels;
[0009] S103: Extracting steel bar point cloud: extracting steel bar point cloud data based on the internal parameters of the depth camera for the RGB image data obtained in step S102 and the Depth image collected in step S101;
[0010] S104: reinforcement point cloud data enhancement: using DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering algorithm to process the reinforcement point cloud data, and obtaining point cloud data of each reinforcement; and gradually performing linear point cloud screening, low discreteness point cloud screening and point cloud denoising through reinforcement point cloud data enhancement;
[0011] S105: Automatic calculation of steel bar spacing: Use KD-Tree (K-Dimensional tree) to establish efficient point cloud indexing, calculate the minimum spacing between each steel bar point cloud based on the point cloud processing algorithm, and obtain the spacing of each steel bar.
[0012] Optionally, in step S101, a single prefabricated component is placed on the ground, and a handheld depth camera is used to perform surround multi-site photography of the steel bar parts of the single prefabricated component to obtain RGBD image data of all steel bar parts of the prefabricated component.
[0013] Optionally, step S102 specifically includes:
[0014] S1021: Dataset preparation, using various prefabricated component steel bar images to create ADE20K format dataset;
[0015] S1022: Model training, using the stochastic gradient descent (SGD) optimization method and polynomial decay strategy to iterate the IntemImage network for a preset number of times, and the iterative update rule is:
[0016]
[0017] Among them, lr t represents the learning rate at the tth iteration; lr0 is the initial learning rate; t is the current number of iterations; max_iters is the maximum number of iterations; power is the power exponent;
[0018] S1023: InternImage segmentation of steel bar pixels: Use the trained IntemImage network to perform semantic segmentation on the RGB image collected in step S101 to obtain RGB image data containing only steel bar pixels.
[0019] Optionally, the extraction of steel bar point cloud data based on the internal parameters of the depth camera in step S103 is specifically mapped in the following manner:
[0020]
[0021] Among them, K is the intrinsic parameter matrix of the camera; f x and f y is the focal length of the camera; c x and c y is the principal point coordinate of the camera; (u, v) is the pixel coordinate; d is the depth corresponding to each pixel; (x, y, 1) is the normalized coordinate in the camera coordinate system; (X, Y, Z) is the three-dimensional coordinate of the point cloud;
[0022] After the mapping is completed, the (X, Y, Z) coordinate data of each point in the point cloud is output to obtain the steel point cloud data with the background environment information removed.
[0023] Optionally, in step S104, the DBSCAN clustering algorithm defines clusters through two parameters: neighborhood radius ∈ and minimum number of points MintPts. For each point in the point cloud, if the number of points within its ∈ radius is greater than or equal to MinPts, the point is a core point and is grouped into the same cluster with all the points in its neighborhood, thereby completing the point cloud data classification.
[0024] Optionally, in step S104, the principal component analysis method is used to calculate the first principal component variance ratio λ of each steel bar point cloud, and the specific calculation formula is as follows:
[0025]
[0026] Among them, λ1 is the maximum eigenvalue of the covariance matrix of the point cloud, corresponding to the variance of the first principal component; λ2 and λ3 are the second and third eigenvalues of the covariance matrix, corresponding to the variance of the second and third principal components respectively; by setting the minimum threshold of the variance ratio, the steel bar point cloud with obvious linear characteristics is screened, thereby realizing linear point cloud screening.
[0027] Optionally, for the steel bar point cloud with obvious linear characteristics obtained by screening, the standard deviation of each steel bar in the first principal component direction is calculated to screen the steel bar point cloud with low discreteness, specifically including:
[0028] (1) Decentralized processing: P = [P1, P2, ..., P n ], Pi =(x i ,y i , z i ),
[0029] Among them, P is a set of points; P i is a three-dimensional point; P′ is the result of point cloud decentralization, and n is the number of points;
[0030] (2) Covariance matrix calculation and eigendecomposition:
[0031] Cv i =λ i v i ;
[0032] Where C is the covariance matrix; λ i is the eigenvalue; v i is the feature vector;
[0033] (3) Project each steel bar point cloud toward its first principal component direction and calculate the standard deviation:
[0034]
[0035] Where v1 is the maximum eigenvalue vector corresponding to the first principal component direction; H i is the Euclidean distance from each point to the centroid after projection; σ is the standard deviation on the first principal component;
[0036] By comparing the standard deviation of each steel bar point cloud, the set number of roots with the smallest standard deviation is obtained to achieve low discreteness steel bar point cloud screening.
[0037] Optionally, step S104 also includes using a radius filter to analyze the neighborhood of each point, and based on the distance range from the point to all neighboring points, filter out some outliers that do not meet the requirements, thereby achieving point cloud denoising; the radius filter defines a search radius R and a minimum number of neighboring points N for each point min , calculate the number of neighboring points within the radius. If the number of neighboring points of a point is less than the preset threshold, then the point is considered as noise or outlier and is removed.
[0038] Optionally, the specific process of step S105 includes:
[0039] For any point p i In point cloud C i In point cloud C, find j The nearest neighbor point p in j , and calculate the distance d between the two points. The calculation formula is as follows:
[0040]
[0041] Among them, p i and p j is any point; d(p i , p j ) is the distance between two points;
[0042] For each steel bar point cloud C i , retain the shortest distance between all points, and set a maximum threshold d max If the shortest distance exceeds the threshold, the distance is considered invalid, and the minimum value of all valid shortest distances is taken as the two sets of point clouds C i and C j The overall minimum distance between is calculated as follows:
[0043] d min (C i , C j )=min{d(p i , p j )≤d max}
[0044] D min (C i , C j )=min{d min (C i , C j )}
[0045] Among them, d min is the shortest distance between points; d max is the maximum threshold; D min is the shortest distance between two steel bar point clouds;
[0046] Finally, traverse the point cloud set of each steel bar to obtain the shortest distance D between all steel bars i .
[0047] Optionally, when the shortest distance D between the steel bars i When it exceeds the preset range, an alarm will be given.
[0048] The remarkable effects of the present invention are:
[0049] The present invention proposes an intelligent detection method for the spacing between steel bars of prefabricated components based on RGBD images. The image data collected by the depth camera can conveniently, quickly and accurately realize the intelligent detection of the spacing between steel bars of prefabricated components, effectively improving the intelligent digitization, management capabilities and efficiency of building safety. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] The accompanying drawings of the present invention are as follows:
[0051] Figure 1It is a flow chart of a method for intelligently detecting the spacing between reinforcement bars of prefabricated components based on RGBD images disclosed in the present invention;
[0052] Figure 2 This is an example of the RGBD image acquisition result of the prefabricated component reinforcement in a specific embodiment, where Figure 2 (a) is an RGB image, Figure 2 (b) is the Depth image;
[0053] Figure 3 This is an example of the results of semantic segmentation of steel bar pixels by the InternImage network in a specific implementation example, where Figure 3 (a) is the image after segmentation mask. Figure 3 (b) is the steel bar pixel image;
[0054] Figure 4 This is an example diagram of the steel bar point cloud location extraction result in a specific embodiment;
[0055] Figure 5 This is a DBSCAN clustering result diagram in a specific embodiment;
[0056] Figure 6 This is an example diagram of the calculation result of the linear reinforcement point cloud based on the variance ratio of the first principal component in a specific embodiment;
[0057] Figure 7 This is an example diagram of the cloud computing result of a low-discrete steel bar point cloud based on the standard deviation of the first principal component direction in a specific embodiment;
[0058] Figure 8 This is an example diagram of the calculation results of the reinforcement spacing part based on the point cloud nearest neighbor algorithm in a specific embodiment;
[0059] Fig. 9 A top view and a side view of a point cloud of a prefabricated panel unit scanned by a land-based three-dimensional laser scanner in a specific embodiment;
[0060] Fig.10 The data comparison and data error of the vertical depth camera detection value and the land-based 3D laser scanner detection value of the left side of the prefabricated component in the specific embodiment
[0061] Fig.11 The data comparison and data error of the horizontal depth camera detection value and the land-based three-dimensional laser scanner detection value of the left side of the prefabricated component in the specific embodiment;
[0062] Fig.12 The data comparison and data error of the depth camera detection value in the vertical direction of the rib spacing on the right side of the prefabricated component and the land-based three-dimensional laser scanner detection value;
[0063] Fig.13It is the data comparison and data error of the horizontal depth camera detection value and the land-based three-dimensional laser scanner detection value of the right side reinforcement spacing of the prefabricated component in the specific embodiment. DETAILED DESCRIPTION
[0064] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.
[0065] like Figure 1 As shown, the present invention discloses an intelligent detection method for the spacing of steel bars of prefabricated components based on RGBD images, and the method comprises the following steps:
[0066] S101: Data collection: Use a depth camera to collect RGBD images of the steel bars of prefabricated components. Usually, the prefabricated components are placed on the ground, and the Azure Kinect depth camera is used to collect RGBD images of the steel bars of the prefabricated components in a surround multi-station manner to obtain RGBD image data of the steel bars of the prefabricated components. The collection results are as follows: Figure 2 As shown, the obtained RGBD image includes an RGB image and a Depth image;
[0067] S102: semantic segmentation of steel bar pixels: using the trained InternImage semantic segmentation network to segment the steel bar pixels in the steel bar RGBD image to obtain RGB image data containing only steel bar pixels;
[0068] In specific implementation, step S102 specifically includes:
[0069] S1021: Dataset preparation, using various prefabricated component steel bar images to create ADE20K format dataset;
[0070] S1022: Model training, using the stochastic gradient descent (SGD) optimization method and polynomial decay strategy to iterate the IntemImage network for a preset number of times, and the iterative update rule is:
[0071]
[0072] Among them, lr t represents the learning rate at the tth iteration; lr0 is the initial learning rate; t is the current number of iterations; max_iters is the maximum number of iterations; power is the power exponent;
[0073] S1023: InternImage segmentation of steel bar pixels: Use the trained IntemImage network to perform semantic segmentation on the RGB image collected in step S101 to obtain RGB image data containing only steel bar pixels.
[0074] In this example, the SGD optimization method and polynomial decay strategy are used to train the IntemImage network for 16,000 iterations. The initial learning rate lr0 = 0.01, the power exponent power = 0.9, and the decay weight is 0.0005. Some segmentation results are shown in Figure 2. Figure 3 As shown;
[0075] S103: Extracting steel bar point cloud: extracting steel bar point cloud data based on the internal parameters of the depth camera for the RGB image data obtained in step S102 and the Depth image collected in step S101;
[0076] The mapping is performed in the following way:
[0077]
[0078] Among them, K is the intrinsic parameter matrix of the camera; f x and f y is the focal length of the camera; c x and c y is the principal point coordinate of the camera; (u, v) is the pixel coordinate; d is the depth corresponding to each pixel; (x, y, 1) is the normalized coordinate in the camera coordinate system; (X, Y, Z) is the three-dimensional coordinate of the point cloud;
[0079] After the mapping is completed, the (X, Y, Z) coordinate data of each point in the point cloud is output to obtain the steel point cloud data with the background environment information removed.
[0080] In this example, the steel bar pixels are mapped to the point cloud based on the camera intrinsic parameters, and the camera intrinsic parameters f are set. x =912.310242,f y =912.028198, C x =962.345581, C y =548.116638, the result of extracting the steel bar point cloud data is as follows Figure 4 As shown;
[0081] S104: reinforcement point cloud data enhancement: using the DBSCAN clustering algorithm to process the reinforcement point cloud data to obtain point cloud data of each reinforcement; and gradually performing linear point cloud screening, low discreteness point cloud screening and point cloud denoising through reinforcement point cloud data enhancement;
[0082] In the specific implementation, the neighborhood radius ∈=0.06m and the minimum number of points MintPts=80 are set to complete the point cloud data classification, such as Figure 5 As shown, the classified point cloud data of each steel bar are obtained.
[0083] against Figure 5The clustered point cloud data shown above uses the principal component analysis method to calculate the first principal component variance ratio λ of each steel bar point cloud. The higher the ratio, the stronger the linear feature of the point cloud. The specific calculation formula is as follows:
[0084]
[0085] Among them, λ1 is the maximum eigenvalue of the covariance matrix of the point cloud, corresponding to the variance of the first principal component; λ2 and λ3 are the second and third eigenvalues of the covariance matrix, corresponding to the variance of the second and third principal components; by setting the minimum threshold of the variance ratio, the steel bar point cloud with obvious linear characteristics is screened, thereby realizing the linear point cloud screening. In this example, according to the characteristics of the steel bar, the minimum threshold of the variance ratio is set to min =0.94, we get Figure 6 The steel bar point cloud with obvious linear features is shown;
[0086] Furthermore, for Figure 6 The screening shown above obtains a steel bar point cloud with obvious linear characteristics, and the standard deviation of the first principal component direction of each steel bar is calculated to screen the steel bar point cloud with low discreteness, including:
[0087] (1) Decentralized processing: P = [P1, P2, ..., P n ], P i =(x i ,y i , z i ),
[0088] Among them, P is a set of points; P i is a three-dimensional point; P′ is the result of point cloud decentralization, and n is the number of points;
[0089] (2) Covariance matrix calculation and eigendecomposition:
[0090] Cv i =λ i v i ;
[0091] Where C is the covariance matrix; λ i is the eigenvalue; v i is the feature vector;
[0092] (3) Project each steel bar point cloud toward its first principal component direction and calculate the standard deviation:
[0093]
[0094] Where v1 is the maximum eigenvalue vector corresponding to the first principal component direction; H iis the Euclidean distance from each point to the centroid after projection; σ is the standard deviation on the first principal component;
[0095] By comparing the standard deviation of each steel bar point cloud, the five with the smallest standard deviation are selected and the Figure 7 The low-dispersion reinforcement point cloud is shown.
[0096] In addition, a radius filter is used to analyze the neighborhood of each point. Based on the distance range from the point to all neighboring points, some outliers that do not meet the requirements are filtered out, thereby achieving point cloud denoising; the search radius R = 0.01m and the minimum number of neighboring points N are set. min = 100 parameter, calculate the number of neighboring points within the radius. If the number of neighboring points of a point is less than the preset threshold, then this point is considered as noise or outlier and is removed, thus completing the point cloud data preprocessing work of discrete point removal;
[0097] S105: Automatic calculation of steel bar spacing: Use KD-Tree to establish efficient point cloud indexing, calculate the minimum spacing between each steel bar point cloud based on the point cloud processing algorithm, and obtain the spacing of each steel bar.
[0098] In specific implementation, for any point p i In point cloud C i In point cloud C, find j The nearest neighbor point p in j , and calculate the distance d between the two points. The calculation formula is as follows:
[0099]
[0100] Among them, p i and p j is any point; d(p i , p j ) is the distance between two points;
[0101] For each steel bar point cloud C i , retain the shortest distance between all points, and set a maximum threshold d max If the shortest distance exceeds the threshold, the distance is considered invalid, and the minimum value of all valid shortest distances is taken as the two sets of point clouds C i and C j The overall minimum distance between is calculated as follows:
[0102] d min (C i , C j )=min{d(p i , p j )≤d max}
[0103] Din (C i , C j )=min{d min (C i , C j )}
[0104] Among them, d min is the shortest distance between points; d max is the maximum threshold; D min is the shortest distance between two steel bar point clouds;
[0105] Finally, traverse the point cloud set of each steel bar to obtain the shortest distance D between all steel bars i Some results are as follows Figure 8 shown.
[0106] In this embodiment, there are 272 steel bars on both sides of the prefabricated panel unit, and 135 sets of steel bar spacing on one side. Therefore, 34 depth cameras are set to shoot on the left and right sides, and each station is set at two angles, vertical and horizontal. Each shooting can obtain 5 steel bar spacings, and the spacing detection values of the upper and middle parts of the steel bars can be obtained. At the same time, the prefabricated panel unit is scanned in color using a high-precision land-based 3D laser scanner FARO S150, with the resolution set to 1 / 4 and the point cloud quality set to 4X, and at least 3 public target balls are ensured between adjacent measuring stations. The scanning results are as follows: Fig. 9 FARO S150 has a ranging accuracy of ±1mm within the detection range of 0.6m-150m, so this device can be used to verify the feasibility of the proposed method.
[0107] like Figure 10-13 They are the comparison results of the spacing detection values of the upper and middle parts of all the steel bars in the example using this method and the point cloud processing detection values collected by the land-based three-dimensional laser scanner.
[0108] The above experimental results prove that the method for intelligent detection of the spacing between steel bars of prefabricated components based on RGBD images disclosed in the present invention can complete the intelligent detection of the spacing between steel bars of prefabricated components, can use a computer to accurately calculate the spacing between steel bars of prefabricated components, and effectively improve the intelligent and digital management capabilities and efficiency of building safety.
[0109] Finally, it should be noted that the embodiments described in this specification are merely illustrative of the present invention, and those skilled in the art should understand that various changes may be made in form and detail without departing from the spirit and scope of the present invention as defined in the appended claims.
Claims
1. An intelligent detection method for the spacing of steel bars of prefabricated components based on RGBD images, characterized in that: The following steps are involved: S101: Data acquisition: using a depth camera to acquire a steel bar RGBD image of a prefabricated component, wherein the RGBD image includes an RGB image and a Depth image; S102: semantic segmentation of steel bar pixels: using the trained InternImage semantic segmentation network to segment the steel bar pixels in the steel bar RGBD image to obtain RGB image data containing only steel bar pixels; S103: Extracting steel bar point cloud: extracting steel bar point cloud data based on the internal parameters of the depth camera for the RGB image data obtained in step S102 and the Depth image collected in step S101; S104: reinforcement point cloud data enhancement: using the DBSCAN clustering algorithm to process the reinforcement point cloud data to obtain point cloud data of each reinforcement; and gradually performing linear point cloud screening, low discreteness point cloud screening and point cloud denoising through reinforcement point cloud data enhancement; S105: Automatic calculation of steel bar spacing: Use KD-Tree to establish efficient point cloud indexing, calculate the minimum spacing between each steel bar point cloud based on the point cloud processing algorithm, and obtain the spacing of each steel bar.
2. The intelligent detection method for prefabricated component reinforcement spacing based on RGBD images according to claim 1 is characterized in that: In step S101, a single prefabricated component is placed on the ground, and a handheld depth camera is used to perform surround multi-site photography of the steel bar portion of the single prefabricated component to obtain RGBD image data of all steel bar portions of the prefabricated component.
3. The intelligent detection method for prefabricated component reinforcement spacing based on RGBD images according to claim 1 or 2, characterized in that: Step S102 specifically includes: S1021: Dataset preparation, using various prefabricated component steel bar images to create ADE20K format dataset; S1022: Model training, using the stochastic gradient descent (SGD) optimization method and polynomial decay strategy to iterate the InternImage network for a preset number of times, and the iterative update rule is: Among them, lr t represents the learning rate at the tth iteration; lr0 is the initial learning rate; t is the current number of iterations; max_iters is the maximum number of iterations; power is the power exponent; S1023: InternImage segmentation of steel bar pixels: Use the trained InternImage network to perform semantic segmentation on the RGB image collected in step S101 to obtain RGB image data containing only steel bar pixels.
4. The intelligent detection method for prefabricated component reinforcement spacing based on RGBD images according to claim 3 is characterized in that: The steel bar point cloud data extracted based on the internal parameters of the depth camera in step S103 is specifically mapped in the following manner: Among them, K is the intrinsic parameter matrix of the camera; f x and f y is the focal length of the camera; c x and c y is the principal point coordinate of the camera; (u, v) is the pixel coordinate; d is the depth corresponding to each pixel; (x, y, 1) is the normalized coordinate in the camera coordinate system; (X, Y, Z) is the three-dimensional coordinate of the point cloud; After the mapping is completed, the (X, Y, Z) coordinate data of each point in the point cloud is output to obtain the steel point cloud data with the background environment information removed.
5. The intelligent detection method for prefabricated component reinforcement spacing based on RGBD images according to claim 4 is characterized in that: In step S104, the DBSCAN clustering algorithm defines clusters through two parameters: neighborhood radius ∈ and minimum number of points MintPts. For each point in the point cloud, if the number of points within its ∈ radius is greater than or equal to MinPts, the point is a core point and is grouped into the same cluster with all the points in its neighborhood, completing the point cloud data classification.
6. The intelligent detection method for prefabricated component reinforcement spacing based on RGBD images according to claim 5 is characterized in that: In step S104, the principal component analysis method is used to calculate the first principal component variance ratio λ of each steel bar point cloud. The specific calculation formula is as follows: Among them, λ1 is the maximum eigenvalue of the covariance matrix of the point cloud, corresponding to the variance of the first principal component; λ2 and λ3 are the second and third eigenvalues of the covariance matrix, corresponding to the variance of the second and third principal components respectively; by setting the minimum threshold of the variance ratio, the steel bar point cloud with obvious linear characteristics is screened, thereby realizing linear point cloud screening.
7. The intelligent detection method for prefabricated component reinforcement spacing based on RGBD images according to claim 6 is characterized in that: For the steel bar point cloud with obvious linear characteristics obtained by screening, the standard deviation of the first principal component direction of each steel bar is calculated to screen the steel bar point cloud with low discreteness, including: (1) Decentralized processing: P = [P1, P2, ...., P n ], P i =(x i ,y i ,z i ), Among them, P is a set of points; P i is a three-dimensional point; P' is the result of point cloud decentralization, and n is the number of points; (2) Covariance matrix calculation and eigendecomposition: Cv i =λ i v i Where C is the covariance matrix; λ i is the eigenvalue; v i is the feature vector; (3) Project each steel bar point cloud toward its first principal component direction and calculate the standard deviation: Where v1 is the maximum eigenvalue vector corresponding to the first principal component direction; H i is the Euclidean distance from each point to the centroid after projection; σ is the standard deviation on the first principal component; By comparing the standard deviation of each steel bar point cloud, the set number of roots with the smallest standard deviation is obtained to achieve low discreteness steel bar point cloud screening.
8. The intelligent detection method for prefabricated component reinforcement spacing based on RGBD images according to claim 7 is characterized in that: Step S104 also includes using a radius filter to analyze the neighborhood of each point, and based on the distance range from the point to all neighboring points, filter out some outliers that do not meet the requirements, thereby achieving point cloud denoising; the radius filter defines a search radius R and a minimum number of neighboring points N for each point min , calculate the number of neighboring points within the radius. If the number of neighboring points of a point is less than the preset threshold, then the point is considered as noise or outlier and is removed.
9. The intelligent detection method for prefabricated component reinforcement spacing based on RGBD images according to any one of claims 4 to 8, characterized in that: The specific process of step S105 includes: For any point p i In point cloud C i In point cloud C, find j The nearest neighbor point p in j , and calculate the distance d between the two points. The calculation formula is as follows: Among them, p i and p j is any point; d(p i ,p j ) is the distance between two points; For each steel bar point cloud C i , retain the shortest distance between all points, and set a maximum threshold d max If the shortest distance exceeds the threshold, the distance is considered invalid, and the minimum value of all valid shortest distances is taken as the two sets of point clouds C i and C j The overall minimum distance between is calculated as follows: the min (C i ,C j )=min{d(p i ,p j )≤d max } D min (C i ,C j )=min{d min (C i ,C j )} Among them, d min is the shortest distance between points; d max is the maximum threshold; D min is the shortest distance between two steel bar point clouds; Finally, traverse the point cloud set of each steel bar to obtain the shortest distance D between all steel bars i .
10. The method for intelligent detection of steel bar spacing of prefabricated components based on RGBD images according to claim 9, characterized in that: When the shortest distance between the steel bars is D i When it exceeds the preset range, an alarm will be given.
Citation Information
Patent Citations
Rebar spacing identification method and system, electronic equipment and storage medium
CN117058240A
User-customized group exercise program provision system, method and application using the same
KR1020250025129A
Method for Extracting Planes from 3D Point Cloud Sensor Data
US20150154467A1
Automatic measurement method and apparatus suitable for long reinforcement cage
WO2024099007A1
Cited By
Precast beam reinforcement cage quality detection method and system based on depth vision
CN120182175A
Reinforcing mesh automatic identification method based on unmanned aerial vehicle laser point cloud data
CN121658919A
Multi-layer reinforcing steel bar intelligent counting method and system based on deep learning
CN122391273A
A multi-layer steel bar intelligent counting method and system based on deep learning
CN122391273B
