Machine vision-based intelligent laying method and system for optical cable of electric power working well

Abstract

The invention discloses an electric power working well optical cable intelligent laying method and system based on machine vision, and relates to the technical field of optical cable laying, and the method comprises the following steps: obtaining visual data and three-dimensional point cloud data in an electric power working well, and extracting a multi-scale visual feature map and point cloud geometric key points; the multi-scale visual feature map comprises a depth grayscale image, a color feature map and a semantic feature map; based on the multi-scale visual feature map and the three-dimensional point cloud data, constructing semantic color point cloud data; generating a high-precision semantic color 3D model based on the semantic color point cloud data; processing the high-precision semantic color 3D model through a multi-target adaptive algorithm to generate an optical cable laying optimal path; laying feasibility verification based on physical simulation is carried out on the optical cable laying optimal path, and a final laying path is generated. The method is used for solving the problem that the path is difficult to accurately plan during optical cable laying.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of optical cable laying, more particularly, the present application relates to a power manhole optical cable intelligent laying method and system based on machine vision. BACKGROUND

[0002] As a key link of power communication network construction, the rationality and efficiency of the planning scheme of optical cable laying path planning directly affect the construction cost and communication transmission quality. Traditional power manhole optical cable laying mainly relies on manual survey and experience decision-making. Such methods show obvious problems of low efficiency, poor safety and insufficient adaptability when facing complex well environment, dynamic obstacles and diversified laying constraints. Therefore, how to optimize the laying planning process of optical cable through intelligent technology and improve the accuracy and reliability of the path scheme has become one of the key tasks of the intelligent operation and maintenance of current power infrastructure.

[0003] For example, the invention patent with the announcement number CN115862012A announced a point cloud data semantic segmentation method, device, electronic equipment and storage medium, including: acquiring three-dimensional point cloud data and color image collected in the target mine area; the three-dimensional point cloud data and the color image are fused to obtain the three-dimensional color point cloud data to be segmented; the three-dimensional color point cloud data to be segmented is input into a semantic segmentation model to obtain a semantic segmentation result of the three-dimensional color point cloud data to be segmented. In this way, by collecting point cloud data and corresponding color image in the mine operation scene, the point cloud data and the corresponding color image are fused to obtain color point cloud data, so that the semantic segmentation of the color point cloud data by the semantic segmentation model can effectively improve the accuracy of point cloud semantic segmentation.

[0004] For example, the invention patent with the announcement number CN119204527A announced an underground power cable management method, system, computer equipment and readable storage medium, including: dividing the underground power cable into regions, acquiring data of sensors in each partition; establishing a well risk assessment model according to the well structure and environmental factors, obtaining a well risk assessment result, inputting the cable operation data into a fault prediction model to obtain a fault prediction result, determining the maintenance task field corresponding to the two results, determining the maintenance personnel matched with the task field according to the professional of the candidate maintenance personnel; importing the three-dimensional point cloud data into the geographic information system, reconstructing the spatial position to form a three-dimensional model of the cable, constructing a graph structure of the laying area to be laid based on the model, obtaining the corresponding laying path based on the nodes and edges in the graph structure; visualizing the three-dimensional model, the well risk assessment result, the fault prediction result, the laying path and the maintenance personnel information.

[0005] The above disclosed technical scheme has at least the following technical problems: Manual laying relies on manual surveying and experience-based decision-making, which is cumbersome and prone to rework, resulting in low overall efficiency. Personnel need to work in confined spaces with risks such as collapses and toxic gases, making it difficult to guarantee personal safety. The human brain has difficulty accurately balancing multiple constraints such as distance, bending radius, and obstacle avoidance in complex three-dimensional environments, making it impossible to generate the globally optimal path.

[0006] To address the above problems, this invention proposes a solution. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a machine vision-based intelligent optical cable laying method and system for power manholes. By using visual data and 3D point cloud data inside the power manhole, semantic color point cloud data is constructed and a high-precision semantic color 3D model is generated. Based on the model, the optimal path for optical cable laying is generated to solve the problem of difficulty in accurately planning the path for laying optical cables.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A machine vision-based intelligent fiber optic cable laying method for power wells includes the following steps: acquiring visual data and 3D point cloud data inside the power well; extracting multi-scale visual feature maps and geometric key points of the point cloud, wherein the multi-scale visual feature maps include depth grayscale images, color feature maps, and semantic feature maps; constructing semantic color point cloud data based on the multi-scale visual feature maps and 3D point cloud data; generating a high-precision semantic color 3D model based on the semantic color point cloud data; processing the high-precision semantic color 3D model through a multi-objective adaptive algorithm to generate the optimal fiber optic cable laying path; verifying the feasibility of the optimal fiber optic cable laying path based on physical simulation, and generating the final laying path.

[0009] In a preferred embodiment, constructing semantic color point cloud data specifically involves: Spatial matching is performed between point cloud geometric keypoints and semantic feature maps to construct point cloud-semantic spatial relationships; Color key points are extracted based on color feature maps and combined with deep grayscale images to generate three-dimensional color feature points. Based on 3D color feature points and point cloud geometric key points, a point cloud registration transformation matrix is ​​constructed using the ICP algorithm; Based on the point cloud-semantic space relationship, color key points, and point cloud registration transformation matrix, semantic color point cloud data is generated.

[0010] In a preferred embodiment, the construction of point cloud-semantic space relations specifically includes: Based on the shooting perspective information of RGB images, calculate the semantic projection set of key points on the semantic feature map of geometric key points in point cloud; The semantic projection set of key points is matched with the semantic feature map to extract the optimal semantic matching relationship; Based on the optimal semantic matching relationship, a semantic registration transformation matrix is ​​constructed between 3D point cloud data and multi-scale visual feature maps to obtain the point cloud-semantic space relationship.

[0011] In a preferred embodiment, constructing the point cloud registration transformation matrix specifically involves: Based on color keypoints and depth grayscale images, calculate the keypoint color back projection set of the color feature map on point cloud data; The key point color back projection set and point cloud geometric key points are matched as 3D point cloud base points. All point cloud data are registered using the ICP algorithm, and a point cloud registration transformation matrix for all point cloud data is generated.

[0012] In a preferred embodiment, generating semantic color point cloud data specifically involves: Based on the point cloud-semantic space relationship, semantic information of geometric key points in the point cloud is extracted; Based on the key point color back projection set and point cloud geometric key points, extract the color information of point cloud geometric key points; Based on the semantic and color information of geometric key points in the point cloud, semantic color point cloud data is generated.

[0013] In a preferred embodiment, generating a high-precision semantic color 3D model specifically involves: Based on equipment motion data, prior monitoring of motion state is performed to construct absolute scale information; Based on semantic color point cloud data and absolute scale information, a semantic color 3D model is constructed. A high-precision semantic color 3D model is generated by iterative correction based on semantic color 3D model and multi-scale visual feature map.

[0014] In a preferred embodiment, the semantic color 3D model and the multi-scale visual feature map are iteratively corrected, specifically as follows: Render a semantic color 3D model from a two-dimensional perspective to generate a virtual two-dimensional perspective image set. Semantic extraction is performed on virtual two-dimensional viewpoint image groups to generate multi-scale rendering feature maps; Based on the differences between multi-scale rendered feature maps and multi-scale visual feature maps, the semantic color 3D model is adjusted to generate a high-precision semantic color 3D model.

[0015] In a preferred embodiment, the multi-objective adaptive path planning algorithm includes: Based on a high-precision semantic color 3D model, laying distance, path smoothness and safety redundancy distance are extracted, and a multi-objective cost function is constructed. Based on scene characteristics, a preset weight set is assigned to the multi-objective cost function, and the optimal path for optical cable laying is generated by using the multi-objective A* search algorithm.

[0016] In a preferred embodiment, the final laying step specifically involves collision detection based on real-time feedback data from distance sensors. If a path conflict is detected, path adjustment information is generated and fed back to the path planning module for re-optimization of the path.

[0017] A machine vision-based intelligent fiber optic cable laying system for power wells includes: The data acquisition module is used to acquire visual data and 3D point cloud data inside the power well, and extract multi-scale visual feature maps and point cloud geometric key points. The multi-scale visual feature maps include color feature maps and semantic feature maps. The data processing module is used to construct semantic color point cloud data based on multi-scale visual feature maps and 3D point cloud data; The environment modeling module is used to generate high-precision semantic color 3D models based on semantic color point cloud data; The path planning module is used to combine a high-precision semantic color 3D model and generate the optimal path for optical cable laying through a multi-objective adaptive algorithm. The simulation verification module is used to verify the feasibility of laying the optimal optical cable path based on physical simulation and generate the final laying path.

[0018] The technical effects and advantages of the machine vision-based intelligent fiber optic cable laying method and system for power wells in this invention are as follows: This invention constructs a high-precision semantic 3D model using visual data and 3D point cloud data from inside power wells, systematically recording color and semantic information to provide a reliable basis for path planning. It adjusts path planning weights based on environmental features, generates the optimal optical cable laying path using multi-objective search, and verifies the path's feasibility through physical simulation. A verification feedback mechanism is employed to continuously optimize the planning scheme, achieving accurate planning and safe verification of optical cable laying paths, effectively solving the problem of difficult-to-accurate path planning for optical cable laying. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the intelligent laying method for optical cables in power wells based on machine vision, provided in an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the intelligent optical cable laying system for power wells based on machine vision, provided in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1, Figure 1 This invention presents a machine vision-based intelligent fiber optic cable laying method for power wells, comprising the following steps: S1, acquire visual data and 3D point cloud data inside the power well, and extract multi-scale visual feature maps and point cloud geometric key points. The multi-scale visual feature maps include depth grayscale images, color feature maps and semantic feature maps. S2, based on multi-scale visual feature maps and 3D point cloud data, constructs semantic color point cloud data; S3 generates high-precision semantic color 3D models based on semantic color point cloud data; S4, through multi-objective adaptive algorithm to process high-precision semantic color 3D model, generates the optimal path for optical cable laying; S5 verifies the feasibility of laying the optimal optical cable path based on physical simulation and generates the final laying path.

[0023] S1, acquire visual data and 3D point cloud data inside the power well, and extract multi-scale visual feature maps and point cloud geometric key points. The multi-scale visual feature maps include depth grayscale images, color feature maps and semantic feature maps.

[0024] It should be noted that the visual data includes RGB images and grayscale images of the inside of the power well, as well as the spatial position and angle at which each image was taken.

[0025] In this embodiment, the 3D point cloud data is 3D point cloud data after partition filtering. The specific filtering steps are as follows: The point cloud curvature distribution was calculated, dividing the point cloud into high curvature points and low curvature points; Use a small filter radius in high curvature point cloud regions to preserve details, and use a large filter radius in low curvature point cloud regions to effectively reduce noise.

[0026] In this embodiment, the specific method for extracting geometric key points of the point cloud is as follows: Create a 3D mesh on the entire point cloud, dividing the space into numerous small cubes; Within each small cube, compare the curvature values ​​of all points in the 3D point cloud data; Only the point with the largest curvature within the voxel is retained as the geometric key point of the point cloud.

[0027] It should be noted that point cloud geometric keypoints represent the semantics of a set of points in 3D point cloud data.

[0028] In this embodiment, the specific steps for extracting multi-scale visual feature maps are as follows: The original RGB image of the inside of the power well is subjected to illumination equalization, noise suppression, color enhancement and geometric correction. The accelerated segment test feature algorithm is used to locate all corner features in the image, and the corner points are used as color feature points in the color feature map; Convert the RGB image to the LAB color space and use K-means clustering to segment and extract color regions; Calculate the geometric properties of each color region; Based on the geometric and color attributes of color regions, a mathematical model is formed by encoding prior knowledge of the well environment, and semantic features are assigned to each color region to form a semantic feature map.

[0029] It should be noted that the color feature maps and semantic feature maps in the multi-scale visual feature maps are in one-to-one correspondence. Each set of color feature maps and semantic feature maps is extracted based on the same RGB image, and the spatial position and angle of each set of images are consistent when they are captured.

[0030] S2 constructs semantic color point cloud data based on multi-scale visual feature maps and 3D point cloud data.

[0031] In this embodiment, constructing semantic color point cloud data specifically involves: Spatial matching is performed between point cloud geometric keypoints and semantic feature maps to construct point cloud-semantic spatial relationships; Color key points are extracted based on color feature maps and combined with deep grayscale images to generate three-dimensional color feature points. Based on 3D color feature points and point cloud geometric key points, a point cloud registration transformation matrix is ​​constructed using the ICP algorithm; Based on the point cloud-semantic space relationship, color key points, and point cloud registration transformation matrix, semantic color point cloud data is generated.

[0032] In this embodiment, the construction of point cloud-semantic space relations specifically involves: Based on the shooting perspective information of RGB images, calculate the semantic projection set of key points on the semantic feature map of geometric key points in point cloud; The semantic projection set of key points is matched with the semantic feature map to extract the optimal semantic matching relationship; Based on the optimal semantic matching relationship, a semantic registration transformation matrix is ​​constructed between 3D point cloud data and multi-scale visual feature maps to obtain the point cloud-semantic space relationship.

[0033] In this embodiment, the specific steps for calculating the semantic projection set of the geometric key points of the point cloud onto the semantic feature map are as follows: Extract the shooting spatial location and angle of the semantic feature map, and construct the semantic two-dimensional plane of the semantic feature map in the three-dimensional space of the three-dimensional point cloud data; Calculate the vertical projection of the geometric key points of the point cloud onto the semantic two-dimensional plane to generate a semantic projection set of key points.

[0034] In this embodiment, the optimal semantic matching relationship is extracted, specifically as follows: Based on geometric consistency constraints, semantic consistency constraints, and spatial distribution constraints, the cosine similarity of the semantic projection set of key points on the semantic feature map is calculated as the semantic feature similarity. Establish the globally optimal semantic matching relationship based on the semantic feature relationship with the highest similarity.

[0035] It should be noted that the point cloud-semantic space relationship represents the semantics of each group of point clouds after the 3D point cloud data is divided according to the geometric key points of the point cloud.

[0036] In this embodiment, the point cloud registration transformation matrix is ​​constructed as follows: Based on color keypoints and depth grayscale images, calculate the keypoint color back projection set of the color feature map on point cloud data; The key point color back projection set and point cloud geometric key points are matched as 3D point cloud base points. All point cloud data are registered using the ICP algorithm, and a point cloud registration transformation matrix for all point cloud data is generated.

[0037] In this embodiment, the calculation of the key point color back projection set is specifically as follows: Extract the shooting spatial location and angle of the color feature map, and construct a semantic two-dimensional plane of the color feature map in the three-dimensional space of the three-dimensional point cloud data; By combining the deep grayscale image, the back projection points of the color feature points in the color feature map in the three-dimensional space of the three-dimensional point cloud data are calculated to form the key point color back projection set.

[0038] In this embodiment, the key point color back projection set and the point cloud geometric key points are matched as the base points of the 3D point cloud. Specifically, the key point color back projection set and the point cloud geometric key points are matched by the nearest neighbor search algorithm, and the matched points are used as the base points of the 3D point cloud.

[0039] It should be noted that the ICP algorithm uses the matching of the keypoint color backprojection set and the point cloud geometric keypoints as coarse registration to register all 3D point cloud data into the same space. The 3D point cloud base points are used as the initial pose to execute the ICP algorithm, iteratively calculating the optimal spatial transformation parameters to achieve high-precision alignment of the 3D coordinates of the cloud map.

[0040] In this embodiment, generating semantic color point cloud data specifically involves: Based on the point cloud-semantic space relationship, semantic information of geometric key points in the point cloud is extracted; Based on the key point color back projection set and point cloud geometric key points, extract the color information of point cloud geometric key points; Based on the semantic and color information of geometric key points in the point cloud, semantic color point cloud data is generated.

[0041] In this embodiment, the semantic color point cloud data specifically includes: Semantic color point cloud data {field1: Geometric coordinates, field2: Normal vector field3: Color field4: semantics}.

[0042] S3 generates high-precision semantic color 3D models based on semantic color point cloud data.

[0043] In this embodiment, generating a high-precision semantic color 3D model specifically involves: Based on equipment motion data, prior monitoring of motion state is performed to construct absolute scale information; Based on semantic color point cloud data and absolute scale information, a semantic color 3D model is constructed. A high-precision semantic color 3D model is generated by iterative correction based on semantic color 3D model and multi-scale visual feature map.

[0044] In this embodiment, the construction of absolute scale information specifically refers to: Attitude, velocity, and displacement calculations are performed based on equipment motion data. Based on the physical size calibration points and the calibration point cloud scale, absolute scale information is constructed. Spatial information correction is performed on 3D point cloud data in absolute scale space.

[0045] In this embodiment, a semantic color 3D model is constructed as follows: Octree spatial partitioning based on 3D point cloud data; Calculate the spatial normal vectors of the octree nodes to generate the orientation information of the model surface; Based on the Poisson reconstruction algorithm, 3D point cloud data and model surface orientation information are calculated to generate a continuous triangular mesh; Surface noise of continuous triangular networks is eliminated based on the Laplace smoothing algorithm. The color and semantic information in semantic color point cloud data are mapped onto the mesh surface to form a semantic color 3D model.

[0046] In this embodiment, the iterative correction based on the semantic color 3D model and multi-scale visual feature maps specifically involves: Render a semantic color 3D model from a two-dimensional perspective to generate a virtual two-dimensional perspective image set. Semantic extraction is performed on virtual two-dimensional viewpoint image groups to generate multi-scale rendering feature maps; Based on the differences between multi-scale rendered feature maps and multi-scale visual feature maps, the semantic color 3D model is adjusted to generate a high-precision semantic color 3D model.

[0047] In this embodiment, generating a virtual two-dimensional viewpoint image group specifically involves: Based on the shooting position and angle of each group of images in the multi-scale visual feature map, the plane where the multi-scale visual feature map is located is determined in the high-precision semantic color 3D model. Virtual depth grayscale images, virtual color images, and virtual depth grayscale images are generated by vertex color interpolation rendering based on a high-precision semantic color 3D model.

[0048] It should be noted that the step of generating multi-scale rendering feature maps is the same as the step of extracting multi-scale visual feature maps based on visual data inside power wells in S1. The multi-scale rendering feature maps include virtual depth grayscale images, virtual color feature maps, and virtual semantic feature maps.

[0049] In this embodiment, the difference between the multi-scale rendering feature map and the multi-scale visual feature map is specifically calculated by using a cosine similarity algorithm to determine the geometric feature similarity between the multi-scale rendering feature map and the multi-scale visual feature map.

[0050] In this embodiment, the semantic color 3D model is adjusted as follows: For regions with geometric feature similarity below a threshold, the difference type is classified. For semantically inconsistent parts, the semantics are redistributed; For regions with geometric shape deviations, vertex positions are corrected.

[0051] In this embodiment, the specific formula for vertex position correction is as follows:

[0052] In the formula, The direction and distance that the vertex needs to move. The geometric similarity of the vertex positions. This represents the depth value of the vertex at the corresponding position in the virtual depth grayscale image within the multi-scale rendered feature map. This represents the depth value of the vertex at the corresponding position in the depth grayscale image of the multi-scale visual feature map. Let be the unit normal vector at the vertex.

[0053] S4 uses a multi-objective adaptive algorithm to process the high-precision semantic color 3D model and generate the optimal path for optical cable laying.

[0054] In this embodiment, the multi-objective adaptive path planning algorithm includes: Based on a high-precision semantic color 3D model, laying distance, path smoothness and safety redundancy distance are extracted, and a multi-objective cost function is constructed. Based on scene characteristics, a preset weight set is assigned to the multi-objective cost function, and the optimal path for optical cable laying is generated by using the multi-objective A* search algorithm.

[0055] It should be noted that the laying distance is specifically calculated as the Euclidean distance in three-dimensional space from the coordinate sequence of the path points. This is used to identify the required fiber optic cable length.

[0056] It should be noted that path smoothness, specifically, is the sum of the turning angles formed by consecutive path points. This is used to ensure that no mechanical damage occurs during fiber optic cable laying and to avoid excessive bending.

[0057] It should be noted that the safety redundancy distance is specifically the distance between the route point and nearby obstacles. This ensures that the laying path maintains a safe distance from hazardous areas, preventing intrusion into areas with standing water or high-temperature metal pipes.

[0058] In this embodiment, the multi-objective cost function is specifically:

[0059] In the formula, The cost of path P. Let P be the radiation distance of path P. Let P be the path smoothness. Let P be the safe redundancy distance. , , These are preset weighting coefficients.

[0060] In this embodiment, generating the optimal path for optical cable laying specifically involves: Scenario category analysis is conducted based on expert experience, and weight allocation groups for the corresponding scenario categories are selected. Convert high-precision semantic color 3D models into voxel meshes and assign semantic meaning; The starting point and key points of the optical cable laying are determined. In the voxel network, the weight allocation group is used as the search standard, and the multi-objective A* search algorithm is used to search for the path as the optimal path for optical cable laying.

[0061] S5 verifies the feasibility of laying the optimal optical cable path based on physical simulation and generates the final laying path.

[0062] In this embodiment, the step of generating the final laying step specifically involves collision detection based on real-time feedback data from the distance sensor. If a path conflict is detected, path adjustment information is generated and fed back to the path planning module for re-optimization of the path.

[0063] In this embodiment, the generated path adjustment information specifically refers to: Based on the physical properties of optical cables and the environment model of the manhole, a virtual laying scenario is constructed in the physics engine; The optimal path for laying optical cables, as planned, is imported into the scenario, and dynamic physical simulation is performed to reproduce the real behavior of optical cables under gravity, friction, and bending. During the simulation, it is detected in real time whether the optical cable collides geometrically with the model, whether its bending radius is lower than the minimum allowable value, or whether the laying tension exceeds the safety threshold. If any of the above checks fail, an adjustment instruction containing specific conflict information is generated and fed back to the path planning module.

[0064] In this embodiment, the feedback to the path planning module specifically refers to... Mark the path segments where collisions occur in the physics simulation; Add path turning radius constraints to path segments with insufficient bending radius; For suspended sections with excessive laying tension, add intermediate virtual support points and mark them in the generated laying path.

[0065] Example 2, Figure 2 The present invention provides a machine vision-based intelligent optical cable laying system for power wells, comprising: The data acquisition module is used to acquire visual data and 3D point cloud data inside the power well, and extract multi-scale visual feature maps and point cloud geometric key points. The multi-scale visual feature maps include color feature maps and semantic feature maps. The data processing module is used to construct semantic color point cloud data based on multi-scale visual feature maps and 3D point cloud data; The environment modeling module is used to generate high-precision semantic color 3D models based on semantic color point cloud data; The path planning module is used to combine a high-precision semantic color 3D model and generate the optimal path for optical cable laying through a multi-objective adaptive algorithm. The simulation verification module is used to verify the feasibility of laying the optimal optical cable path based on physical simulation and generate the final laying path.

[0066] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0067] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0068] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0069] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0070] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0071] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine vision-based intelligent laying method for optical cables in power manholes, characterized in that, Includes the following steps: Visual data and 3D point cloud data inside the power well are acquired, and multi-scale visual feature maps and point cloud geometric key points are extracted. The multi-scale visual feature maps include depth grayscale images, color feature maps and semantic feature maps. Semantic color point cloud data is constructed based on multi-scale visual feature maps and 3D point cloud data. Generate a high-precision semantic color 3D model based on semantic color point cloud data; The optimal path for optical cable laying is generated by processing a high-precision semantic color 3D model using a multi-objective adaptive algorithm. The feasibility of laying the optimal optical cable path is verified by physical simulation, and the final laying path is generated.

2. The intelligent laying method for optical cables in power wells based on machine vision according to claim 1, characterized in that, The construction of semantic color point cloud data specifically involves: Spatial matching is performed between point cloud geometric keypoints and semantic feature maps to construct point cloud-semantic spatial relationships; Color key points are extracted based on color feature maps and combined with deep grayscale images to generate three-dimensional color feature points. Based on 3D color feature points and point cloud geometric key points, a point cloud registration transformation matrix is ​​constructed using the ICP algorithm; Based on the point cloud-semantic space relationship, color key points, and point cloud registration transformation matrix, semantic color point cloud data is generated.

3. The intelligent laying method for optical cables in power wells based on machine vision according to claim 2, characterized in that, The construction of point cloud-semantic space relationships specifically involves: Based on the shooting perspective information of RGB images, calculate the semantic projection set of key points on the semantic feature map of geometric key points in point cloud; The semantic projection set of key points is matched with the semantic feature map to extract the optimal semantic matching relationship; Based on the optimal semantic matching relationship, a semantic registration transformation matrix is ​​constructed between 3D point cloud data and multi-scale visual feature maps to obtain the point cloud-semantic space relationship.

4. The intelligent laying method for optical cables in power wells based on machine vision according to claim 3, characterized in that, The construction of the point cloud registration transformation matrix is ​​specifically as follows: Based on color keypoints and depth grayscale images, calculate the keypoint color back projection set of the color feature map on point cloud data; The key point color back projection set and point cloud geometric key points are matched as 3D point cloud base points. All point cloud data are registered using the ICP algorithm, and a point cloud registration transformation matrix for all point cloud data is generated.

5. The intelligent laying method for optical cables in power wells based on machine vision according to claim 4, characterized in that, The generation of semantic color point cloud data specifically includes: Based on the point cloud-semantic space relationship, semantic information of geometric key points in the point cloud is extracted; Based on the key point color back projection set and point cloud geometric key points, extract the color information of point cloud geometric key points; Based on the semantic and color information of geometric key points in the point cloud, semantic color point cloud data is generated.

6. The intelligent laying method for optical cables in power wells based on machine vision according to claim 5, characterized in that, The generation of the high-precision semantic color 3D model specifically involves: Based on equipment motion data, prior monitoring of motion state is performed to construct absolute scale information; Based on semantic color point cloud data and absolute scale information, a semantic color 3D model is constructed. A high-precision semantic color 3D model is generated by iterative correction based on semantic color 3D model and multi-scale visual feature map.

7. The intelligent laying method for optical cables in power wells based on machine vision according to claim 6, characterized in that, The semantic color 3D model and multi-scale visual feature maps are iteratively corrected, specifically as follows: Render a semantic color 3D model from a two-dimensional perspective to generate a virtual two-dimensional perspective image set. Semantic extraction is performed on virtual two-dimensional viewpoint image groups to generate multi-scale rendering feature maps; Based on the differences between multi-scale rendered feature maps and multi-scale visual feature maps, the semantic color 3D model is adjusted to generate a high-precision semantic color 3D model.

8. The intelligent laying method for optical cables in power wells based on machine vision according to claim 7, characterized in that, The multi-objective adaptive path planning algorithm includes: Based on a high-precision semantic color 3D model, laying distance, path smoothness and safety redundancy distance are extracted, and a multi-objective cost function is constructed. Based on scene characteristics, a preset weight set is assigned to the multi-objective cost function, and the optimal path for optical cable laying is generated by using the multi-objective A* search algorithm.

9. The intelligent laying method for optical cables in power wells based on machine vision according to claim 8, characterized in that, The final laying step involves collision detection based on real-time feedback data from distance sensors. If a path conflict is detected, path adjustment information is generated and fed back to the path planning module for path optimization.

10. A system using the machine vision-based intelligent laying method for optical cables in power wells as described in any one of claims 1-9, comprising: The data acquisition module is used to acquire visual data and 3D point cloud data inside the power well, and extract multi-scale visual feature maps and point cloud geometric key points. The multi-scale visual feature maps include color feature maps and semantic feature maps. The data processing module is used to construct semantic color point cloud data based on multi-scale visual feature maps and 3D point cloud data; The environment modeling module is used to generate high-precision semantic color 3D models based on semantic color point cloud data; The path planning module is used to process the high-precision semantic color 3D model through a multi-objective adaptive algorithm to generate the optimal path for optical cable laying. The simulation verification module is used to verify the feasibility of laying the optimal optical cable path based on physical simulation and generate the final laying path.

Images