A power distribution network construction safety distance measurement and risk identification method and device

Abstract

The present application relates to the technical field of power operation safety monitoring, and discloses a power distribution network construction safety distance measurement and risk identification method and device. The method comprises: acquiring a double-view synchronous image sequence of a power distribution network construction site; extracting a pixel-level target mask with fine edges; inputting the double-view synchronous image sequence into a stereo matching model; reconstructing various independent operation object point clouds; calculating the three-dimensional spatial Euclidean distance of the interactive target pair, and decoupling the three-dimensional spatial Euclidean distance along the spatial coordinate axis into a horizontal distance component and a vertical distance component; constructing a spatial relationship topology graph; inputting the spatial relationship topology graph into EGNN for deep spatial logical reasoning, and extracting global high-order topological features; and performing double-track intelligent evaluation on the current power distribution network construction scene. The method greatly reduces the false positives and false negatives in safety monitoring.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology for power construction operations, and in particular to a method and device for measuring safety distances and identifying risks during power distribution network construction. Background Technology

[0002] As the core hub of the power system, the safety of the new construction, renovation, and maintenance of the power distribution network directly affects the stable operation of the power grid and the safety of frontline workers. At outdoor power distribution network construction sites, the working environment is often highly open, complex, and dynamically changing. The construction process involves not only the lifting and hoisting of large heavy machinery such as cranes, but also the coordinated efforts of multiple workers, and there are often nearby power lines or equipment within the work area. Therefore, accurately monitoring the spatial relationships between construction personnel, construction machinery, and electrical equipment is fundamental to effectively preventing serious accidents such as electric shock and mechanical collisions.

[0003] Safety supervision at traditional power distribution construction sites relies heavily on manual inspections and experience-based estimations. On-site safety officers visually assess the distances and conditions between various work entities. However, given the complexities of the work area, blind spots, and fatigue from prolonged monitoring, manual supervision is prone to oversights. Furthermore, the human eye's perception of three-dimensional distances is inherently subjective and cannot meet the demands of continuous, precise, and quantifiable dynamic safety management in modern power grids.

[0004] To overcome the shortcomings of manual supervision, some solutions have attempted to introduce binocular vision technology in recent years. However, in actual complex outdoor construction scenarios, existing technologies have revealed many insurmountable bottlenecks: First, the drastic changes in outdoor lighting, cluttered background textures, and the tendency for equipment surfaces to produce high levels of glare cause the disparity maps obtained by conventional binocular stereo matching algorithms to often contain a large number of holes and noise, resulting in highly unstable depth perception. Second, existing target detection and segmentation networks, when undergoing multi-task joint training, are often limited by the imbalance in the optimization of detection and segmentation tasks, leading to insufficiently refined edges of the output construction object masks. This pixel-level positioning deviation directly lowers the quality of the source data during subsequent conversion from 2D depth maps to 3D point clouds. More importantly, most existing visual safety early warning systems rely on the spacing of the target's 2D bounding boxes or simple 3D spatial linear Euclidean distance for risk assessment, severely lacking the ability to deeply analyze the 3D spatial topological distribution among multiple targets. In power distribution network construction, a scenario with strong interaction among multiple stakeholders, the spatial distribution characteristics often determine the nature of the risk. Under different hazardous conditions, the three-dimensional straight-line Euclidean distance between targets may be exactly the same, but their implied physical meaning and hazard level are completely different. Existing methods lack the ability to decompose three-dimensional spatial distances in multiple dimensions and combine them with graph structures for spatial geometric logic reasoning. They cannot automate and accurately achieve the task of safety monitoring at the work site or identify complex construction operation conditions, and are prone to false alarms and missed alarms in practical applications. Summary of the Invention

[0005] The main objective of this invention is to provide a method and apparatus for measuring safe distances and identifying risks during power distribution network construction, so as to solve the technical problems in the prior art.

[0006] To achieve the above objectives, the present invention provides a method for measuring and identifying the safety distance during power distribution network construction, the method comprising:

[0007] S10: Acquire a dual-view synchronous image sequence of the power distribution network construction site and establish a basic image sample library covering multiple categories of entity objects, including construction workers, construction machinery, and power equipment.

[0008] S20, the acquired left image is input into the joint object detection and instance segmentation network DI-MaskDINO, the balance token-aware optimization module is used to guide the initial feature update, and the pixel-level target mask with fine edge is extracted;

[0009] S30, the dual-view synchronized image sequence is input into the FoundationStereo stereo matching model to calculate the disparity map of the current complex operation scene, and based on the pre-calibrated binocular camera intrinsic and extrinsic parameter matrix, the disparity value D in the disparity map is converted into a depth map containing the depth value Z of each point in the scene.

[0010] S40, Spatial pixel alignment is performed between the depth map and the pixel-level target mask with fine edges to reconstruct the point cloud of various independent operation objects participating in the construction operation;

[0011] S50, for the point clouds of the various independent operation objects, the spatial nearest neighbor between specific interactive target pairs is found by traversing the K-dimensional tree spatial indexing algorithm, the three-dimensional spatial Euclidean distance between the interactive target pairs is calculated, and the three-dimensional spatial Euclidean distance is decoupled into horizontal distance components and vertical distance components along the spatial coordinate axis;

[0012] S60, construct a spatial relationship topology map based on the physical distribution of each working entity on site, and use the horizontal distance component and the vertical distance component as edge attribute features connecting the corresponding nodes;

[0013] S70, The spatial relationship topology graph is input into EGNN for deep spatial logic reasoning, and global high-order topological features are extracted;

[0014] S80, based on the global high-order topology features, performs dual-track intelligent evaluation of the current power distribution network construction scenario and outputs an evaluation report.

[0015] Optionally, step S20 includes the following steps:

[0016] S210, input the acquired left image into the joint object detection and instance segmentation network DI-MaskDINO;

[0017] S220, extracts balance-aware queries based on the residual dual-selection mechanism in the imbalance removal module;

[0018] S230, initial feature update is guided by the balance token perception optimization module;

[0019] S240 extracts pixel-level target masks with fine edges.

[0020] Optionally, step S30 includes the following steps:

[0021] S310, Input the dual-view synchronized image sequence into the FoundationStereo stereo matching model;

[0022] S320, which fuses monocular prior geometric features of the visual base model through a lateral fine-tuning adapter;

[0023] S330 uses an attention-based hybrid cost filtering module to aggregate features in the spatial and disparity dimensions and calculate the disparity map of the current complex work scenario;

[0024] S340, based on a pre-calibrated stereo camera intrinsic and extrinsic parameter matrix, converts the disparity values ​​D in the disparity map into a depth map containing the depth values ​​Z of each point in the scene, as shown in the following formula:

[0025] ;

[0026] In the formula: f is the camera focal length, and T is the distance between the horizontal baselines of the left and right cameras.

[0027] Optionally, step S40 includes the following steps:

[0028] S410, Spatial pixel alignment is performed between the depth map and the pixel-level target mask with fine edges;

[0029] S420 combines the two-dimensional pixel coordinate system point sets of each work subject with the corresponding depth values ​​and back-projects them to the three-dimensional world coordinate system to reconstruct the point cloud of various independent work objects participating in the construction operation.

[0030] Optionally, the formula for back-projecting the two-dimensional pixel coordinate system point set of each working entity, combined with the corresponding depth value, to the three-dimensional world coordinate system is as follows:

[0031] ;

[0032] In the formula: f x and f y , respectively, are the focal lengths of the camera in the x and y directions; u0 and v0 are the coordinates of the center of the imaging plane of the binocular camera; X w , Y w , Z w , respectively, are three-dimensional world coordinates; u and v are two-dimensional coordinates of the image plane.

[0033] Optionally, step S50 includes the following steps:

[0034] S510, for the point clouds of the various independent operation objects, the spatial nearest neighbor between specific interactive target pairs is found by traversing the K-dimensional tree spatial indexing algorithm.

[0035] S520, assuming the three-dimensional spatial coordinates of the closest points of interactive target A and target B are respectively P A Calculate the three-dimensional Euclidean distance d between the interactive target pair using (x1, y1, z1) and PB(x2, y2, z2);

[0036] S530, the three-dimensional Euclidean distance is decoupled into a horizontal distance component d along the spatial coordinate axis. h and vertical distance component d v .

[0037] Optionally, step S60 includes the following steps:

[0038] S610, construct a spatial relationship topology map based on the physical distribution of various on-site operating entities;

[0039] S620, assuming the three-dimensional spatial coordinates of the closest points of interactive target A and target B are respectively P A Calculate the three-dimensional Euclidean distance d between the interactive target pair using (x1, y1, z1) and PB(x2, y2, z2);

[0040] S630, the three-dimensional Euclidean distance is decoupled into a horizontal distance component d along the spatial coordinate axis. h and vertical distance component d v .

[0041] Furthermore, to achieve the above objectives, this application also provides a device for measuring and identifying safe distances during power distribution network construction, the device comprising:

[0042] The sequence acquisition module is used to acquire a dual-view synchronous image sequence of the power distribution network construction site and to establish a basic image sample library covering multiple categories of entity objects, including construction workers, construction machinery and power equipment.

[0043] The mask extraction module is used to input the acquired left image into the joint object detection and instance segmentation network DI-MaskDINO, use the balance token perception optimization module to guide the initial feature update, and extract a pixel-level target mask with fine edges.

[0044] The depth map conversion module is used to input the dual-view synchronized image sequence into the FoundationStereo stereo matching model, calculate the disparity map of the current complex operation scene, and convert the disparity value D in the disparity map into a depth map containing the depth value Z of each point in the scene based on the pre-calibrated binocular camera intrinsic and extrinsic parameter matrix.

[0045] The pixel alignment module is used to spatially align the depth map with the finely edged pixel-level target mask to reconstruct the point cloud of various independent work objects involved in the construction operation.

[0046] The decoupling module is used to traverse and find the spatial nearest neighbor between specific interactive target pairs based on the K-dimensional tree spatial indexing algorithm for the point clouds of the various independent operation objects, calculate the three-dimensional spatial Euclidean distance of the interactive target pairs, and decouple the three-dimensional spatial Euclidean distance into horizontal distance components and vertical distance components along the spatial coordinate axis.

[0047] The topology graph construction module is used to construct a spatial relationship topology graph based on the physical distribution of various on-site work entities, and uses the horizontal distance component and the vertical distance component as edge attribute features connecting corresponding nodes.

[0048] The feature extraction module is used to input the spatial relationship topology graph into the EGNN for deep spatial logic reasoning and to extract global high-order topological features.

[0049] The evaluation module is used to perform dual-track intelligent evaluation of the current power distribution network construction scenario based on the global high-order topology features and output an evaluation report.

[0050] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0051] The power distribution network construction safety distance measurement and risk identification method provided in this application, by introducing the DI-MaskDINO model and through residual dual selection and balance perception mechanism, fundamentally overcomes the technical bottleneck of mutual constraint between detection and segmentation feature extraction in traditional multi-task networks under complex outdoor occlusion scenarios. It achieves high-quality edge mask stripping for complex targets such as personnel with changing postures and slender mechanical booms, eliminating a large amount of two-dimensional noise interference for back-end three-dimensional mapping.

[0052] Furthermore, to address the binocular mismatch problem caused by metal reflection and drastic changes in lighting at outdoor power distribution network construction sites, this invention adopts the FoundationStereo framework, seamlessly injecting powerful monocular vision priors into the stereo matching cost volume. By leveraging attention-based hybrid cost filtering, long-range contextual reasoning is performed in the spatial and disparity domains. This allows for stable output of high-fidelity disparity maps without the need for tedious fine-tuning for specific scenarios, significantly enhancing the robustness of 3D depth perception and low-level ranging under complex working conditions.

[0053] This invention decouples the extreme distance of point clouds into horizontal and vertical features and uses them as edge attributes of a graph structure. Combined with the rigorous spatial equivariance reasoning capabilities of EGNN, the system can not only perceive distance but also deeply analyze the complex and fundamentally different three-dimensional physical topological logic between multiple targets. This breakthrough enables the system to accurately warn of dynamic safety hazards such as electric shock and collisions between humans and machines, and to automatically and accurately identify complex construction behaviors involving multiple categories and high-level semantics. This significantly reduces false alarms and missed alarms in safety monitoring, and achieves refined and intelligent joint management and control of safety precautions and work progress at outdoor construction sites of power distribution networks. Attached Figure Description

[0054] Figure 1 A flowchart illustrating the method for measuring and identifying safety distances during power distribution network construction provided in this application embodiment;

[0055] Figure 2 This is a structural block diagram of the power distribution network construction safety distance measurement and risk identification device provided in the embodiments of this application;

[0056] Figure 3 This is a schematic diagram of the spatial topology provided in an embodiment of this application. Detailed Implementation

[0057] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings.

[0058] It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of the application. Rather, these embodiments are provided to make the disclosure more thorough and complete, and to fully convey the scope of the disclosure to those skilled in the art.

[0059] To address the aforementioned technical problems, embodiments of this application provide a method for measuring safety distances and identifying risks during power distribution network construction. This method can be executed by a computer, such as... Figure 1 As shown, the method may include the following steps:

[0060] S10: Use binocular vision acquisition equipment to acquire a dual-view synchronous image sequence of the power distribution network construction site, and establish a basic image sample library covering multiple categories of physical objects including construction workers, construction machinery and power equipment.

[0061] Specifically, the binocular vision acquisition device is a binocular camera, and the dual-view synchronous image sequence refers to the image pair of the same scene captured by two cameras.

[0062] S20, the acquired left image is input into the joint object detection and instance segmentation network DI-MaskDINO, the initial feature update is guided by the balanced token-aware optimization module, and a pixel-level object mask with fine edge details is extracted.

[0063] Among them, the binocular vision acquisition device (binocular camera) has two cameras, and the left image refers to the image captured by the left camera;

[0064] In an exemplary embodiment, step S20 may specifically include the following steps:

[0065] S210, input the acquired left image into the joint object detection and instance segmentation network DI-MaskDINO;

[0066] S220, extracts balance-aware queries based on the residual dual-selection mechanism in the imbalance removal module;

[0067] S230, initial feature update is guided by the balance token perception optimization module;

[0068] S240 extracts pixel-level target masks with fine edges.

[0069] Specifically, both the imbalance removal module and the balance token perception optimization module are internal structures of the DI-MaskDINO model.

[0070] In step S220, the steps for extracting the balance-aware query are as follows:

[0071] A joint object detection and instance segmentation network DI-MaskDINO, including an imbalance removal module and a balance token-aware optimization module, is constructed. The acquired left image is input into DI-MaskDINO. The imbalance removal module extracts a balance-aware query, and the balance token-aware optimization module uses the query to guide the initial feature update. Finally, a pixel-level target mask with fine edges is extracted based on the fusion of the refined query and multi-scale features.

[0072] S30, the dual-view synchronized image sequence is input into the FoundationStereo stereo matching model to calculate the disparity map of the current complex operation scene, and based on the pre-calibrated binocular camera intrinsic and extrinsic parameter matrix, the disparity value D in the disparity map is converted into a depth map containing the depth value Z of each point in the scene.

[0073] FoundationStereo is a stereo matching model with strong zero-shot generalization ability. In this embodiment, the network weights pre-trained on a massive general dataset are directly loaded for inference. This eliminates the need for remodeling or tedious model fine-tuning for specific construction scenarios of outdoor power distribution networks with varying lighting and complex backgrounds, and can stably output high-fidelity disparity maps.

[0074] In an exemplary embodiment, step S30 may specifically include the following steps:

[0075] S310, Input the dual-view synchronized image sequence into the FoundationStereo stereo matching model;

[0076] S320, which fuses monocular prior geometric features of the visual base model through a lateral fine-tuning adapter;

[0077] S330 uses an attention-hybrid cost filtering module to aggregate features in the spatial and disparity dimensions and calculate the disparity map of the current complex operation scene. The attention-hybrid cost filtering module is a core network component built into the FoundationStereo stereo matching model.

[0078] S340, based on a pre-calibrated stereo camera intrinsic and extrinsic parameter matrix, converts the disparity values ​​D in the disparity map into a depth map containing the depth values ​​Z of each point in the scene, as shown in the following formula:

[0079] ;

[0080] In the formula: f is the camera focal length, and T is the distance between the horizontal baselines of the left and right cameras.

[0081] S40, Spatial pixel alignment is performed between the depth map and the pixel-level target mask with fine edges to reconstruct the point cloud of various independent operation objects participating in the construction operation.

[0082] In an exemplary embodiment, step S40 may specifically include the following steps:

[0083] S410, Spatial pixel alignment is performed between the depth map and the pixel-level target mask with fine edges;

[0084] S420 combines the two-dimensional pixel coordinate system point sets of each work subject with the corresponding depth values ​​and back-projects them to the three-dimensional world coordinate system to reconstruct the point cloud of various independent work objects participating in the construction operation.

[0085] Optionally, the formula for back-projecting the two-dimensional pixel coordinate system point set of each working entity, combined with the corresponding depth value, to the three-dimensional world coordinate system is as follows:

[0086] ;

[0087] In the formula: f x and f y , respectively, are the focal lengths of the camera in the x and y directions; u0 and v0 are the coordinates of the center of the imaging plane of the binocular camera; X w , Y w , Z w , respectively, are three-dimensional world coordinates; u and v are two-dimensional coordinates of the image plane.

[0088] S50, for the point clouds of the various independent operation objects, the spatial nearest neighbor between specific interactive target pairs is found by traversing the K-dimensional tree spatial indexing algorithm, the three-dimensional spatial Euclidean distance between the interactive target pairs is calculated, and the three-dimensional spatial Euclidean distance is decoupled into horizontal distance components and vertical distance components along the spatial coordinate axis.

[0089] In an exemplary embodiment, step S50 may specifically include the following steps:

[0090] S510, for the point clouds of the various independent operation objects, the spatial nearest neighbor between specific interactive target pairs is found by traversing the K-dimensional tree spatial indexing algorithm.

[0091] S520, assuming the three-dimensional spatial coordinates of the closest points of interactive target A and target B are respectively P ACalculate the three-dimensional Euclidean distance d between the interactive target pair using (x1, y1, z1) and PB(x2, y2, z2);

[0092] S530, the three-dimensional Euclidean distance is decoupled into a horizontal distance component d along the spatial coordinate axis. h and vertical distance component d v .

[0093] In this exemplary embodiment, the K-tree is a classic computer science spatial partitioning data structure, commonly used for searching key data in multidimensional space. The purpose of using the K-tree is not to innovate the algorithm itself, but to address the pain point of excessively large point cloud computing loads. This is because calculating the distance between the point clouds of target A and target B pairwise is too slow and cannot meet real-time requirements. Specifically, a 3D KD-tree is built for the point cloud of target B, and then the points of target A are used to quickly search for nearest neighbors.

[0094] S60, construct a spatial relationship topology map based on the physical distribution of each working entity on site, and use the horizontal distance component and the vertical distance component as edge attribute features connecting the corresponding nodes.

[0095] In an exemplary embodiment, step S60 may specifically include the following steps:

[0096] S610, construct a spatial relationship topology map based on the physical distribution of various on-site operating entities;

[0097] S620, assuming the three-dimensional spatial coordinates of the closest points of interactive target A and target B are respectively P A Calculate the three-dimensional Euclidean distance d between the interactive target pair using (x1, y1, z1) and PB(x2, y2, z2);

[0098] S630, the three-dimensional Euclidean distance is decoupled into a horizontal distance component d along the spatial coordinate axis. h and vertical distance component d v .

[0099] Specifically, constructing a topology graph is to transform the relative positional relationships in physical space into a data format that graph neural networks can process. In detail: the nodes of the topology graph are the targets identified on-site, such as personnel, machinery, and electrical equipment; the edges are the lines connecting each of them; and the attributes of the edges are the horizontal and vertical distances decoupled earlier. Figure 3 An example diagram of the spatial topology is given.

[0100] S70, the spatial relationship topology graph is input into EGNN (equivariant graph neural network) for deep spatial logic reasoning and global high-order topological features are extracted.

[0101] Specifically, when performing node information aggregation and feature transfer, the equivariant graph neural network synchronously and equivariantly updates the latent feature embedding and 3D coordinate embedding of the nodes.

[0102] S80, based on the global high-order topology features, performs dual-track intelligent evaluation of the current power distribution network construction scenario and outputs an evaluation report.

[0103] Specifically, the assessment report includes the following:

[0104] On the one hand, it outputs safety risk warning status, including human-machine collision risk, human-machine electric shock risk, and suspended load intrusion risk (i.e., personnel are too close to or below the object being hoisted); on the other hand, it outputs construction operation status with high-level semantics, covering idle status, hook operation, manual assisted installation operation, and climbing operation, so as to realize the joint identification of safety and progress at the work site.

[0105] The power distribution network construction safety distance measurement and risk identification method provided in this application, by introducing the DI-MaskDINO model and through residual dual selection and balance perception mechanism, fundamentally overcomes the technical bottleneck of mutual constraint between detection and segmentation feature extraction in traditional multi-task networks under complex outdoor occlusion scenarios. It achieves high-quality edge mask stripping for complex targets such as personnel with changing postures and slender mechanical booms, eliminating a large amount of two-dimensional noise interference for back-end three-dimensional mapping.

[0106] Furthermore, to address the binocular mismatch problem caused by metal reflection and drastic changes in lighting at outdoor power distribution network construction sites, this invention adopts the FoundationStereo framework, seamlessly injecting powerful monocular vision priors into the stereo matching cost volume. By leveraging attention-based hybrid cost filtering, long-range contextual reasoning is performed in the spatial and disparity domains. This allows for stable output of high-fidelity disparity maps without the need for tedious fine-tuning for specific scenarios, significantly enhancing the robustness of 3D depth perception and low-level ranging under complex working conditions.

[0107] This invention decouples the extreme distance of point clouds into horizontal and vertical features and uses them as edge attributes of a graph structure. Combined with the rigorous spatial equivariance reasoning capabilities of EGNN, the system can not only perceive distance but also deeply analyze the complex and fundamentally different three-dimensional physical topological logic between multiple targets. This breakthrough enables the system to accurately warn of dynamic safety hazards such as electric shock and collisions between humans and machines, and to automatically and accurately identify complex construction behaviors involving multiple categories and high-level semantics. This significantly reduces false alarms and missed alarms in safety monitoring, and achieves refined and intelligent joint management and control of safety precautions and work progress at outdoor construction sites of power distribution networks.

[0108] Based on the above embodiments, refer to Figure 2Another embodiment of this application also provides a power distribution network construction safety distance measurement and risk identification device. The power distribution network construction safety distance measurement and risk identification device 200 may include the following modules:

[0109] The sequence acquisition module 210 is used to acquire a dual-view synchronous image sequence of the power distribution network construction site and establish a basic image sample library covering multiple categories of entity objects, including construction workers, construction machinery and power equipment.

[0110] The mask extraction module 220 is used to input the acquired left image into the joint object detection and instance segmentation network DI-MaskDINO, use the balance token perception optimization module to guide the initial feature update, and extract a pixel-level target mask with fine edges.

[0111] The depth map conversion module 230 is used to input the dual-view synchronized image sequence into the FoundationStereo stereo matching model, calculate the disparity map of the current complex operation scene, and convert the disparity value D in the disparity map into a depth map containing the depth value Z of each point in the scene based on the pre-calibrated binocular camera intrinsic and extrinsic parameter matrix.

[0112] The pixel alignment module 240 is used to spatially align the depth map with the finely edged pixel-level target mask to reconstruct the point cloud of various independent operation objects participating in the construction operation.

[0113] The decoupling module 250 is used to search for the spatial nearest neighbor between specific interactive target pairs based on the K-dimensional tree spatial indexing algorithm for the point clouds of the various independent operation objects, calculate the three-dimensional spatial Euclidean distance between the interactive target pairs, and decouple the three-dimensional spatial Euclidean distance into horizontal distance components and vertical distance components along the spatial coordinate axis.

[0114] The topology graph construction module 260 is used to construct a spatial relationship topology graph based on the physical distribution of each working entity on site, and uses the horizontal distance component and the vertical distance component as edge attribute features connecting the corresponding nodes.

[0115] The feature extraction module 270 is used to input the spatial relationship topology graph into the EGNN for deep spatial logic reasoning and to extract global high-order topological features.

[0116] The evaluation module 280 is used to perform dual-track intelligent evaluation of the current power distribution network construction scenario based on the global high-order topology features and output an evaluation report.

[0117] In the description of this application, it should be noted that the terms "first", "second", and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0118] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0119] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interface; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0120] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0121] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0122] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The protection scope of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the technical scope disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.

[0124] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

Claims

1. A method for measuring safe distances and identifying risks during power distribution network construction, characterized in that, The method includes the following steps: S10: Acquire a dual-view synchronous image sequence of the power distribution network construction site and establish a basic image sample library covering multiple categories of entity objects, including construction workers, construction machinery, and power equipment. S20, the acquired left image is input into the joint object detection and instance segmentation network DI-MaskDINO, the balance token-aware optimization module is used to guide the initial feature update, and the pixel-level target mask with fine edge is extracted; S30, the dual-view synchronized image sequence is input into the FoundationStereo stereo matching model to calculate the disparity map of the current complex operation scene, and based on the pre-calibrated binocular camera intrinsic and extrinsic parameter matrix, the disparity value D in the disparity map is converted into a depth map containing the depth value Z of each point in the scene. S40, Spatial pixel alignment is performed between the depth map and the pixel-level target mask with fine edges to reconstruct the point cloud of various independent operation objects participating in the construction operation; S50, for the point clouds of the various independent operation objects, the spatial nearest neighbor between specific interactive target pairs is found by traversing the K-dimensional tree spatial indexing algorithm, the three-dimensional spatial Euclidean distance between the interactive target pairs is calculated, and the three-dimensional spatial Euclidean distance is decoupled into horizontal distance components and vertical distance components along the spatial coordinate axis; S60, construct a spatial relationship topology map based on the physical distribution of each working entity on site, and use the horizontal distance component and the vertical distance component as edge attribute features connecting the corresponding nodes; S70, The spatial relationship topology graph is input into EGNN for deep spatial logic reasoning, and global high-order topological features are extracted; S80, based on the global high-order topology features, performs dual-track intelligent evaluation of the current power distribution network construction scenario and outputs an evaluation report.

2. The method for measuring and identifying the safe distance during power distribution network construction according to claim 1, characterized in that, Step S20 includes the following steps: S210, input the acquired left image into the joint object detection and instance segmentation network DI-MaskDINO; S220, extracts balance-aware queries based on the residual dual-selection mechanism in the imbalance removal module; S230, initial feature update is guided by the balance token perception optimization module; S240 extracts pixel-level target masks with fine edges.

3. The method for measuring and identifying the safe distance during power distribution network construction according to claim 1, characterized in that, Step S30 includes the following steps: S310, Input the dual-view synchronized image sequence into the FoundationStereo stereo matching model; S320, which fuses monocular prior geometric features of the visual base model through a lateral fine-tuning adapter; S330 uses an attention-based hybrid cost filtering module to aggregate features in the spatial and disparity dimensions and calculate the disparity map of the current complex work scenario; S340, based on a pre-calibrated stereo camera intrinsic and extrinsic parameter matrix, converts the disparity values ​​D in the disparity map into a depth map containing the depth values ​​Z of each point in the scene, as shown in the following formula: ； In the formula: f is the camera focal length, and T is the distance between the horizontal baselines of the left and right cameras.

4. The method for measuring and identifying the safety distance during power distribution network construction according to claim 1, characterized in that, Step S40 includes the following steps: S410, Spatial pixel alignment is performed between the depth map and the pixel-level target mask with fine edges; S420 combines the two-dimensional pixel coordinate system point sets of each work subject with the corresponding depth values ​​and back-projects them to the three-dimensional world coordinate system to reconstruct the point cloud of various independent work objects participating in the construction operation.

5. The method for measuring and identifying the safe distance during power distribution network construction according to claim 4, characterized in that, The formula for back-projecting the two-dimensional pixel coordinate system point set of each working entity, combined with the corresponding depth value, to the three-dimensional world coordinate system is as follows: ； In the formula: f x and f y , respectively, are the focal lengths of the camera in the x and y directions; u0 and v0 are the coordinates of the center of the imaging plane of the binocular camera; X w , Y w , Z w , respectively, are three-dimensional world coordinates; u and v are two-dimensional coordinates of the image plane.

6. The method for measuring and identifying the safe distance during power distribution network construction according to claim 1, characterized in that, Step S50 includes the following steps: S510, for the point clouds of the various independent operation objects, the spatial nearest neighbor between specific interactive target pairs is found by traversing the K-dimensional tree spatial indexing algorithm. S520, obtain the three-dimensional spatial coordinates of the interactive target A as P. A (x1, y1, z1), the three-dimensional spatial coordinates of target B are P. B (x2, y2, z2), calculate the three-dimensional Euclidean distance d between the interactive target pairs; S530, the three-dimensional Euclidean distance is decoupled into a horizontal distance component d along the spatial coordinate axis. h and vertical distance component d v .

7. A device for measuring safe distances and identifying risks during power distribution network construction, characterized in that, include: The sequence acquisition module is used to acquire a dual-view synchronous image sequence of the power distribution network construction site and to establish a basic image sample library covering multiple categories of entity objects, including construction workers, construction machinery and power equipment. The mask extraction module is used to input the acquired left image into the joint object detection and instance segmentation network DI-MaskDINO, use the balance token perception optimization module to guide the initial feature update, and extract a pixel-level target mask with fine edges. The depth map conversion module is used to input the dual-view synchronized image sequence into the FoundationStereo stereo matching model, calculate the disparity map of the current complex operation scene, and convert the disparity value D in the disparity map into a depth map containing the depth value Z of each point in the scene based on the pre-calibrated binocular camera intrinsic and extrinsic parameter matrix. The pixel alignment module is used to spatially align the depth map with the finely edged pixel-level target mask to reconstruct the point cloud of various independent work objects involved in the construction operation. The decoupling module is used to traverse and find the spatial nearest neighbor between specific interactive target pairs based on the K-dimensional tree spatial indexing algorithm for the point clouds of the various independent operation objects, calculate the three-dimensional spatial Euclidean distance of the interactive target pairs, and decouple the three-dimensional spatial Euclidean distance into horizontal distance components and vertical distance components along the spatial coordinate axis. The topology graph construction module is used to construct a spatial relationship topology graph based on the physical distribution of various on-site work entities, and uses the horizontal distance component and the vertical distance component as edge attribute features connecting corresponding nodes. The feature extraction module is used to input the spatial relationship topology graph into the EGNN for deep spatial logic reasoning and to extract global high-order topological features. The evaluation module is used to perform dual-track intelligent evaluation of the current power distribution network construction scenario based on the global high-order topology features and output an evaluation report.

Images