A cabinet mounting method based on BIM and computer vision technology

Abstract

The application discloses a cabinet installation method based on BIM and computer vision technology, comprising the following steps: 1) establishing a BIM model template; 2) 2D visual detection and ROI extraction; 3) obtaining and down-sampling 3D point cloud data; 4) processing the down-sampled 3D point cloud data by using point cloud correction filtering, including removing point cloud outliers by using an improved DBSCAN algorithm and point cloud completion based on a grid method, so as to improve the point cloud quality; and 5) registering the point cloud and the BIM model template, and verifying whether the installation space meets the requirements of cabinet installation. The method combines 2D and 3D visual technologies to realize efficient and accurate space information acquisition and cabinet installation feasibility evaluation, reduces the labor cost, improves the design accuracy, and provides technical support for efficient reconstruction of urban power distribution rooms.

Description

Technical Field

[0001] This invention belongs to the field of power distribution room equipment management and renovation technology, specifically involving a method for adding cabinets based on BIM and computer vision technology. Background Technology

[0002] Urban power distribution rooms are widely distributed and scattered throughout cities. As electricity demand changes, these rooms often need to be renovated, usually by adding new switchboards. Traditional methods rely on on-site surveys to determine equipment space, cable trench layout, and switchboard installation locations. This method is not only labor-intensive but also prone to inaccurate or missing data. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies. This invention proposes a method for installing power distribution cabinets based on BIM and computer vision technologies. It combines 2D and 3D vision technologies to achieve efficient and accurate spatial information acquisition and feasibility assessment of power distribution cabinet installation. This not only reduces labor costs but also improves the accuracy of design, providing technical support for the efficient transformation of urban power distribution rooms.

[0004] To achieve the above objectives, the present invention adopts the following technical solution.

[0005] A method for installing a cabinet based on BIM and computer vision technology includes the following steps:

[0006] Step S1: Create a BIM model template

[0007] BIM software is used to construct a BIM model template for the power distribution room. The BIM model template contains key features of the cable trench and cabinet of the power distribution room, which serve as a reference framework for subsequent visual inspection and point cloud registration. This ensures the spatial consistency between the point cloud and the BIM model template. Through selective input, only feature data related to the installation of the cabinet is input into the BIM model template, reducing unnecessary data processing and optimizing the system's resource utilization efficiency.

[0008] Step S2: 2D visual inspection and ROI extraction

[0009] RGB images of the power distribution room are acquired using a 2D camera mounted on a 3D camera, and then pre-trained.

[0010] The Nanodet target detection algorithm identifies key feature areas of cable trenches and cabinets, and marks the Region of Interest (ROI) in RGB images, providing accurate spatial location references for 3D point cloud data processing;

[0011] Step S3: Acquire and downsample 3D point cloud data

[0012] Within the region of interest (ROI) marked in step S2, point cloud data is acquired using a 3D camera. The point cloud is divided into spherical regions using a uniform sampling method. The center point of each spherical region is selected as the sampling point to retain the main geometric features of the spherical region while reducing data density.

[0013] Step S4: Use point cloud correction filtering to process the downsampled 3D point cloud data to improve the point cloud quality;

[0014] Step S5: Registration of point cloud with BIM model template

[0015] By extracting key contour features from the BIM model template and point cloud data, calculating the normal vector and Euclidean distance of the feature points, weighting the similarity between the two, and using the KD tree structure to achieve nearest neighbor search matching, the point cloud data and BIM model template are accurately registered, and the installation space is evaluated to determine whether it meets the requirements for cabinet installation.

[0016] Specifically, the algorithm network structure of the pre-trained target detection algorithm Nanodet described in step S2 includes:

[0017] Step S21: Adopt the lightweight ShuffleNetV2 structure, remove the last convolutional layer of the algorithm network, and retain the 16x and 32x downsampling layers to improve the inference speed of the model in detection.

[0018] Step S22: In the Neck section, a feature fusion module is used. The feature fusion module includes a 3*3 convolution and two dilated convolutions with different dilation rates. Then, the Prelu function is used to enhance the nonlinearity of the model and avoid the problem of neuron death. Subsequently, the Concat operation is used to fuse feature information, and the dimensionality is reduced to the dimension of the input feature through a 1*1 convolution. Then, two convolutional layers are used to extract richer deep features, and a residual connection operation is performed with the input features to avoid the loss of shallow information during transmission.

[0019] Specifically, step S4 involves processing the downsampled 3D point cloud data using point cloud correction filtering, including the following steps:

[0020] Step S41: Remove outliers from the point cloud using the improved DBSCAN algorithm.

[0021] Based on the geometric properties of the cable or cabinet, the scanning radius R and the minimum number of points M are set as algorithm parameters. The improved DBSCAN algorithm is used to mark the core points whose neighborhood density meets the conditions and recursively expand to form reasonable clusters. For points that do not meet the neighborhood density conditions, they are marked as noise points and removed from the point cloud data to obtain optimized point cloud data.

[0022] Step S42: Point cloud completion based on mesh method

[0023] Furthermore, the improved DBSCAN algorithm described in step S41 specifically includes the following:

[0024] Step S411: Based on the geometric properties of the cable and the cabinet, set the scanning radius R and the minimum number of points M;

[0025] Step S412: Substitute R and M, traverse the point cloud data, check the number of points in the neighborhood of each point, divide the core points and use them as the starting point of the cluster, divide the points that are not in the cluster as noise, recursively expand each core point until all point cloud data has been traversed.

[0026] Step S413: Calculate the center point coordinates (x_mean, y_mean) of the region of interest (ROI). Use the clustering in steps S411 and S412 to obtain the center value Z_mean of the object in the Z direction, and then obtain the centroid coordinates P(x_mean, y_mean, z_mean) of the object.

[0027] Step S414: Calculate the distance and direction from each core point and noise point to P. Based on the prior features of the object's geometric properties, set a threshold to automatically remove core points with unreasonable ranges and misjudged noise points, thereby obtaining a more reasonable clustering result.

[0028] Furthermore, the point cloud completion based on the mesh method described in step S42 specifically includes the following steps:

[0029] Step S421: Divide the denoised point cloud data into a three-dimensional mesh;

[0030] Step S422: Fill in the missing points within the divided 3D mesh using an interpolation algorithm.

[0031] Specifically, the registration of the point cloud with the BIM model template in step S5 includes the following steps:

[0032] Step S51: Extract contour features from the BIM model template and point cloud data and perform interval sampling to reduce computational burden;

[0033] Step S52: Calculate the normal vector of each feature point, and after normalization based on the Euclidean distance and the difference in normal vector direction between feature points, perform weighted summation to obtain the similarity score;

[0034] Step S53: By constructing a KD tree structure and performing nearest neighbor search, the point with the highest similarity in the point cloud is matched with the feature point in the BIM model template, thereby achieving accurate registration between the point cloud and the BIM model template.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. The method of the present invention reduces manual surveying and measurement errors: It obtains complete spatial information of the power distribution room through a single three-dimensional scan, avoiding the problem of multiple on-site surveys required by traditional manual surveying, which is labor-intensive, and at the same time reduces measurement errors caused by environmental limitations in manual surveying.

[0037] 2. The method of the present invention can effectively improve design efficiency. By combining BIM and computer vision technology, it can automatically calculate the installation space of the cabinet and quickly assess whether the available length of the cable trench meets the requirements, which greatly shortens the design time.

[0038] 3. The BIM model template of the present invention can automatically analyze the installation space and determine whether the size and position of the cabinet meet the installation requirements, thereby improving the accuracy and effectiveness of the installation plan.

[0039] 4. The method of this invention supports remote planning, and enables remote design and feasibility analysis through digital three-dimensional models, thereby achieving intelligent and efficient power distribution room renovation and reducing manpower and time costs. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this disclosure and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart of a method for installing a screen cabinet based on BIM and computer vision technology according to the present invention;

[0042] Figure 2 This is a schematic diagram of the network structure of the pre-trained Nanodet algorithm of this invention;

[0043] Figure 3 This is a schematic diagram of the feature fusion module architecture of the present invention;

[0044] Figure 4 This is a flowchart of the point cloud correction filtering method of the present invention;

[0045] Figure 5 This is a schematic diagram of the BIM model of the power distribution room in an example of the present invention;

[0046] Figure 6 This is a schematic diagram of cabinet location prediction and ROI generation in an example of the present invention;

[0047] Figure 7 This is a schematic diagram of the point cloud data processing results of the display cabinet in an example of the present invention;

[0048] Figure 8 This is a schematic diagram of the registration of the BIM model and the point cloud of the cabinet in an example of the present invention. Detailed Implementation

[0049] To facilitate understanding and implementation of the present invention by those skilled in the art, the various steps of the method proposed in this invention are described in detail below. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various modifications or alterations to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0050] Example

[0051] like Figure 1 As shown, this invention discloses a method for installing a cabinet based on BIM and computer vision technology, including the following steps:

[0052] Step S1: Create a BIM model template

[0053] BIM software is used to construct a BIM model template for the power distribution room. The BIM model template contains key features of the cable trench and cabinet of the power distribution room, which serve as a reference framework for subsequent visual inspection and point cloud registration. This ensures the spatial consistency between the point cloud and the BIM model template. Through selective input, only feature data related to the installation of the cabinet is input into the BIM model template, reducing unnecessary data processing and optimizing the system's resource utilization efficiency.

[0054] Step S2: 2D visual inspection and ROI extraction

[0055] RGB images of the power distribution room are acquired using a 2D camera mounted on a 3D camera, and then pre-trained.

[0056] The Nanodet target detection algorithm identifies key feature areas of cable trenches and cabinets, and marks the Region of Interest (ROI) in RGB images, providing accurate spatial location references for 3D point cloud data processing;

[0057] Step S3: Acquire and downsample 3D point cloud data

[0058] Within the region of interest (ROI) marked in step S2, point cloud data is acquired using a 3D camera. The point cloud is divided into spherical regions using a uniform sampling method. The center point of each spherical region is selected as the sampling point to retain the main geometric features of the spherical region while reducing data density.

[0059] Step S4: Use point cloud correction filtering to process the downsampled 3D point cloud data to improve the point cloud quality;

[0060] Step S5: Registration of point cloud with BIM model template

[0061] By extracting key contour features from the BIM model template and point cloud data, calculating the normal vector and Euclidean distance of the feature points, weighting the similarity between the two, and using the KD tree structure to achieve nearest neighbor search matching, the point cloud data and BIM model template are accurately registered, and the installation space is evaluated to determine whether it meets the requirements for cabinet installation.

[0062] Specifically, such as Figure 2 As shown, the algorithm network structure of the pre-trained target detection algorithm Nanodet described in step S2 includes:

[0063] Step S21: Adopt the lightweight ShuffleNetV2 structure, remove the last convolutional layer of the algorithm network, and retain the 16x and 32x downsampling layers to improve the inference speed of the model in detection.

[0064] Step S22: Use the feature fusion module in the Neck section, such as... Figure 3 As shown, the feature fusion module includes a 3*3 convolution and two dilated convolutions with different dilation rates. The Prelude function is then used to enhance the nonlinearity of the model and avoid the problem of neuron death. Subsequently, the Concat operation is used to fuse feature information, and the dimensionality is reduced to the dimension of the input feature through a 1*1 convolution. Then, two convolutional layers are used to extract richer deep features, and a residual connection operation is performed with the input features to avoid the loss of shallow information during transmission.

[0065] Specifically, such as Figure 4 As shown, step S4, which involves processing the downsampled 3D point cloud data using point cloud correction filtering, includes the following steps:

[0066] Step S41: Remove outliers from the point cloud using the improved DBSCAN algorithm.

[0067] Based on the geometric properties of the cable or cabinet, the scanning radius R and the minimum number of points M are set as algorithm parameters. The improved DBSCAN algorithm is used to mark the core points whose neighborhood density meets the conditions and recursively expand to form reasonable clusters. For points that do not meet the neighborhood density conditions, they are marked as noise points and removed from the point cloud data to obtain optimized point cloud data.

[0068] Step S42: Point cloud completion based on mesh method

[0069] Furthermore, the improved DBSCAN algorithm described in step S41 specifically includes the following:

[0070] Step S411: Based on the geometric properties of the cable and the cabinet, set the scanning radius R and the minimum number of points M;

[0071] Step S412: Substitute R and M, traverse the point cloud data, check the number of points in the neighborhood of each point, divide the core points and use them as the starting point of the cluster, divide the points that are not in the cluster as noise, recursively expand each core point until all point cloud data has been traversed.

[0072] Step S413: Calculate the center point coordinates (x_mean, y_mean) of the region of interest (ROI). Use the clustering in steps S411 and S412 to obtain the center value Z_mean of the object in the Z direction, and then obtain the centroid coordinates P(x_mean, y_mean, z_mean) of the object.

[0073] Step S414: Calculate the distance and direction from each core point and noise point to P. Based on the prior features of the object's geometric properties, set a threshold to automatically remove core points with unreasonable ranges and misjudged noise points, thereby obtaining a more reasonable clustering result.

[0074] Furthermore, the point cloud completion based on the mesh method described in step S42 specifically includes the following steps:

[0075] Step S421: Divide the denoised point cloud data into a three-dimensional mesh;

[0076] Step S422: Fill in the missing points within the divided 3D mesh using an interpolation algorithm.

[0077] Specifically, the registration of the point cloud with the BIM model template in step S5 includes the following steps:

[0078] Step S51: Extract contour features from the BIM model template and point cloud data and perform interval sampling to reduce computational burden;

[0079] Step S52: Calculate the normal vector of each feature point, and after normalization based on the Euclidean distance and the difference in normal vector direction between feature points, perform weighted summation to obtain the similarity score;

[0080] Step S53: By constructing a KD tree structure and performing nearest neighbor search, the point with the highest similarity in the point cloud is matched with the feature point in the BIM model template, thereby achieving accurate registration between the point cloud and the BIM model template.

[0081] The feasibility of the method of the present invention will be further verified through specific installation examples below.

[0082] The traditional feasibility assessment process for manual installation is as follows:

[0083] 1. Before each installation, an infrared rangefinder must be used to accurately measure the dimensions of the wall on site.

[0084] 2. Draw a preliminary sketch by hand on site.

[0085] 3. Use specialized software to convert sketches into CAD drawings to address on-site judgment issues arising from the irregular shapes of some power distribution rooms.

[0086] 4. Finally, the decision on whether or not the installation can be carried out will be based on the CAD drawings.

[0087] Feasibility assessment process for the installation of the method of this invention:

[0088] 1. Obtain the BIM model: Based on a systematic survey, establish a BIM model of the power distribution rooms within the urban area; such as... Figure 5 As shown, find the corresponding model based on the power distribution room that needs to be modified.

[0089] 2. Generate ROI: such as Figure 6 As shown, the improved Nanodet algorithm is used to predict the location and confidence level of the cabinet. The predicted area is defined as the ROI, meaning that point cloud data only needs to be processed within this area.

[0090] 3. Point Cloud Processing: Extract point cloud data from the ROI and perform point cloud post-processing to obtain point cloud data that can be used for registration, such as... Figure 7 The image shows the point cloud processing result of a screen cabinet.

[0091] 4. Accurately register the BIM model and the panel point cloud, such as... Figure 8 As shown, the BIM model and the point cloud of the cabinet are precisely registered to determine whether the installation can be carried out.

[0092] In summary, comparing the method of this invention with traditional methods for assessing the feasibility of installations, it is evident that: First, the method of this invention effectively reduces errors in manual surveying and measurement: by acquiring complete spatial information of the power distribution room through a single 3D scan, it avoids the problem of multiple on-site surveys required by traditional manual surveying, which is labor-intensive, and reduces measurement errors caused by environmental limitations in manual surveying; Second, the method of this invention effectively improves design efficiency: by combining BIM and computer vision technologies, it automatically calculates the installation space of the cabinets and quickly assesses whether the available length of the cable trench meets the requirements, greatly shortening the design time and improving the accuracy and effectiveness of the installation plan; Finally, the method of this invention supports remote planning: through digital 3D models, it enables remote design and feasibility analysis, achieving intelligent and efficient power distribution room renovation and reducing manpower and time costs.

[0093] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for installing a cabinet based on BIM and computer vision technology, characterized in that, Includes the following steps: Step S1: Create a BIM model template BIM software is used to construct a BIM model template for the power distribution room. The BIM model template contains key features of the cable trench and cabinet of the power distribution room, which serve as a reference framework for subsequent visual inspection and point cloud registration. This ensures the spatial consistency between the point cloud and the BIM model template. Through selective input, only feature data related to the installation of the cabinet is input into the BIM model template. Step S2: 2D visual inspection and ROI extraction RGB images of the power distribution room are acquired using a 2D camera mounted on a 3D camera, and then pre-trained. The Nanodet target detection algorithm identifies key feature areas of cable trenches and cabinets, and marks the Region of Interest (ROI) in RGB images, providing accurate spatial location references for 3D point cloud data processing; Step S3: Acquire and downsample 3D point cloud data Within the region of interest (ROI) marked in step S2, point cloud data is acquired using a 3D camera. The point cloud is divided into spherical regions using a uniform sampling method. The center point of each spherical region is selected as the sampling point to retain the main geometric features of the spherical region while reducing data density. Step S4: Use point cloud correction filtering to process the downsampled 3D point cloud data to improve the point cloud quality; Step S5: Registration of point cloud with BIM model template By extracting key contour features from the BIM model template and point cloud data, calculating the normal vector and Euclidean distance of the feature points, weighting the similarity between the two, and using the KD tree structure to achieve nearest neighbor search matching, the point cloud data and BIM model template are accurately registered, and the installation space is evaluated to determine whether it meets the requirements for cabinet installation.

2. The method for installing a cabinet based on BIM and computer vision technology according to claim 1, characterized in that, The algorithm network structure of the pre-trained target detection algorithm Nanodet described in step S2 includes: Step S21: Adopt the lightweight ShuffleNetV2 structure, remove the last convolutional layer of the algorithm network, and retain the 16x and 32x downsampling layers to improve the inference speed of the model in detection. Step S22: In the Neck section, a feature fusion module is used. The feature fusion module includes a 3*3 convolution and two dilated convolutions with different dilation rates. Then, the Prelu function is used to enhance the nonlinearity of the model and avoid the problem of neuron death. Subsequently, the Concat operation is used to fuse feature information, and the dimensionality is reduced to the dimension of the input feature through a 1*1 convolution. Then, two convolutional layers are used to extract richer deep features, and a residual connection operation is performed with the input features to avoid the loss of shallow information during transmission.

3. The method for installing a cabinet based on BIM and computer vision technology according to claim 1, characterized in that, Step S4 describes processing the downsampled 3D point cloud data using point cloud correction filtering, which includes the following steps: Step S41: Remove outliers from the point cloud using the improved DBSCAN algorithm. Based on the geometric properties of the cable or cabinet, the scanning radius R and the minimum number of points M are set as algorithm parameters. The improved DBSCAN algorithm is used to mark the core points whose neighborhood density meets the conditions and recursively expand to form reasonable clusters. For points that do not meet the neighborhood density conditions, they are marked as noise points and removed from the point cloud data to obtain optimized point cloud data. Step S42: Point cloud completion based on mesh method.

4. The method for installing a cabinet based on BIM and computer vision technology according to claim 3, characterized in that, The improved DBSCAN algorithm described in step S41 specifically includes the following: Step S411: Based on the geometric properties of the cable and the cabinet, set the scanning radius R and the minimum number of points M; Step S412: Substitute R and M, traverse the point cloud data, check the number of points in the neighborhood of each point, divide the core points and use them as the starting point of the cluster, divide the points that are not in the cluster as noise, recursively expand each core point until all point cloud data has been traversed. Step S413: Calculate the center point coordinates (x_mean, y_mean) of the region of interest (ROI). Use the clustering in steps S411 and S412 to obtain the center value Z_mean of the object in the Z direction, and then obtain the centroid coordinates P(x_mean, y_mean, z_mean) of the object. Step S414: Calculate the distance and direction from each core point and noise point to P. Based on the prior features of the object's geometric properties, set a threshold to automatically remove core points with unreasonable ranges and misjudged noise points, thereby obtaining a more reasonable clustering result.

5. The method for installing a cabinet based on BIM and computer vision technology according to claim 3, characterized in that, The point cloud completion based on the mesh method described in step S42 specifically includes the following steps: Step S421: Divide the denoised point cloud data into a three-dimensional mesh; Step S422: Fill in the missing points within the divided 3D mesh using an interpolation algorithm.

6. The method for installing a cabinet based on BIM and computer vision technology according to claim 1, characterized in that, The registration of the point cloud with the BIM model template in step S5 specifically includes the following steps: Step S51: Extract contour features from the BIM model template and point cloud data and perform interval sampling to reduce computational burden; Step S52: Calculate the normal vector of each feature point, and after normalization based on the Euclidean distance and the difference in normal vector direction between feature points, perform weighted summation to obtain the similarity score; Step S53: By constructing a KD tree structure and performing nearest neighbor search, the point with the highest similarity in the point cloud is matched with the feature point in the BIM model template, thereby achieving accurate registration between the point cloud and the BIM model template.

Images