A dangerous judgment and active guidance method and system for power construction safety supervision

Abstract

The application relates to a danger judgment and active guidance method and system for power construction safety supervision, and belongs to the technical field of power construction safety. The method comprises the following steps: collecting multi-modal data of a construction scene through intelligent glasses; analyzing the data at an edge computing end to obtain a scene semantic category and an interactive behavior of a worker; according to the scene semantic category, dynamically scheduling and hot loading a matched risk identification model set from a model library to identify a risk factor and a three-dimensional position thereof; acquiring a three-dimensional voxelized static danger level base map; mapping the risk factor into a dynamic risk observation point cloud in a voxel space; taking the static base map as a priori and taking the dynamic point cloud as observation evidence, performing recursive Bayesian filtering through a dynamic Bayesian network to generate a dynamic danger level three-dimensional distribution; generating and rendering graded AR early warning information; when a violation trend occurs, generating and rendering a three-dimensional safety path of a risk avoidance area. The method realizes adaptive, accurate and active construction safety supervision.

Description

Technical Field

[0001] This application relates to the field of power construction safety technology, and more specifically, to a method and system for hazard assessment and proactive guidance in power construction safety supervision. Background Technology

[0002] In the field of power construction safety supervision, existing technologies mostly employ fixed-mode video surveillance combined with specific AI algorithms for risk identification, such as using a single model to identify safety helmets and safety belts. This approach has significant limitations: First, fixed models struggle to adapt to diverse and complex work scenarios, such as working at heights, grounding, and equipment maintenance, resulting in narrow identification ranges, low accuracy, or wasted computational resources. Second, existing risk assessments are mostly discrete event alarms based on single-frame images, lacking continuous modeling of inherent spatial risks and their accumulation and diffusion over time, leading to fluctuating warning results and an inability to provide an overall risk situation. Finally, existing warning methods are mostly passive, relying on on-screen audio-visual prompts or simple AR tag overlays, failing to provide spatially directional proactive safety guidance when workers are about to violate regulations. Therefore, there is an urgent need for an intelligent safety supervision solution that can adapt to multiple scenarios, achieve continuous spatiotemporal risk assessment, and proactively intervene. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes a method and system for hazard assessment and proactive guidance in power construction safety supervision.

[0004] The technical solution of this invention is as follows: This invention proposes a method for hazard assessment and proactive guidance in power construction safety supervision, comprising the following steps: The smart glasses worn by construction workers collect first-person multimodal data of the construction scene; At the edge computing end, multimodal data is parsed to obtain scene semantic categories and construction worker interaction behaviors; Based on the semantic category of the scene, a set of matching risk identification models is dynamically scheduled and hot-loaded from the model library to identify risk factors and their three-dimensional locations and confidence levels; Obtain a three-dimensional voxelized static hazard level base map of the construction scene, with each voxel pre-marked with the basic risk probability; The identified risk factors are mapped as a dynamic risk observation point cloud in voxel space; Using a static hazard level base map as a priori and a dynamic risk observation point cloud as observation evidence, a dynamic hazard level three-dimensional distribution is generated through recursive Bayesian filtering via a dynamic Bayesian network. Based on the dynamic 3D distribution of hazard levels, graded AR early warning information is generated and rendered to smart glasses devices; When a trend of impending violation is observed, a 3D safety navigation path to avoid high-risk areas is generated and rendered in real time based on the dynamic 3D distribution of hazard levels.

[0005] Preferably, the step of dynamically scheduling and hot-loading a set of matching risk identification models from the model library based on the semantic category of the scene specifically includes: Based on the pre-set scene-model correlation map, determine the candidate risk identification model that is associated with the semantic category of the current scene; By combining the real-time computing power and memory resource constraints of the edge computing terminal, a target model set is selected from the candidate risk identification models; The model hot replacement engine loads the target model set into runtime memory without interrupting the multimodal data parsing process, so as to replace or expand the previously loaded model set.

[0006] Preferably, the method further includes a model co-evolution step: When the edge computing terminal detects a high-risk event or the confidence level of the identified risk factor is lower than a preset threshold, it triggers the capture and uploading of de-identified multimodal data segments within the relevant time period. Receive model differential update packages generated from training based on multi-party data and sent from the cloud; The model hot replacement engine performs uninterrupted version upgrades of the corresponding risk identification models in the model library based on differential update packages.

[0007] Preferably, the recursive Bayesian filtering via a dynamic Bayesian network recursively calculates the posterior probability of the danger level of each voxel at time t using the following formula: ; In the formula: Let be the posterior probability of the danger level of each voxel at time t; Let t be the observation likelihood probability with the dynamic risk observation point cloud as input; Let be the risk state transition probability between time t and time t-1; This is a normalization constant; Let be the posterior probability of the voxel's hazard level at time t-1; The true danger state of the voxel at time t; This provides observational evidence at time t.

[0008] Preferably, the generation of graded AR early warning information includes: configuring the visual attribute parameters of the superimposed AR visual elements differently according to the hazard level values ​​corresponding to different spatial locations in the three-dimensional distribution of dynamic hazard levels; the visual attribute parameters include at least one of color, transparency, flashing frequency, and boundary highlight intensity.

[0009] Preferably, the real-time generation and rendering of a 3D safe navigation path to avoid high-risk areas specifically involves: Based on the dynamic hazard level 3D distribution and work space topology, a path planning algorithm is used to dynamically calculate a 3D spatial path from the current pose to the safe target point, avoiding all areas where the hazard level exceeds the preset threshold. The three-dimensional spatial path is converted into a strip-shaped virtual sign with visual guidance effect, which is then overlaid and rendered onto the real-world scene presented by the smart glasses device.

[0010] Preferably, the process of parsing the multimodal data to obtain the scene semantic category and the construction worker's interaction behavior includes: Perform multi-task joint inference on RGB and depth images to output instance segmentation results and attribute labels for construction elements; Based on the position and category combination relationship of various instances in the instance segmentation results, the semantic category label of the current construction scene is inferred; Using construction element instances and construction workers as nodes, and their spatial relationships and movement intentions as edges, a spatiotemporal graph is constructed. Through graph neural network analysis, the work behavior classification labels and violation tendency probabilities are output as the interactive behaviors.

[0011] On the other hand, the present invention also provides a hazard assessment and proactive guidance system for power construction safety supervision, comprising: The risk identification module dynamically schedules and hot-loads a set of matching risk identification models from the model library based on the semantic category of the scene in order to identify risk factors and their three-dimensional location and confidence level. The base map acquisition module acquires a three-dimensional voxelized static hazard level base map of the construction scene, with each voxel pre-marked with the basic risk probability. The risk mapping module maps the identified risk factors into a dynamic risk observation point cloud in voxel space; The joint judgment module uses a static hazard level base map as a priori and a dynamic risk observation point cloud as observation evidence. It then uses a dynamic Bayesian network to perform recursive Bayesian filtering to generate a dynamic hazard level three-dimensional distribution. The early warning feedback module generates graded AR early warning information based on the dynamic three-dimensional distribution of hazard levels and renders it to the smart glasses device. The proactive guidance module generates and renders a 3D safety navigation path to avoid high-risk areas in real time based on the dynamic 3D distribution of hazard levels when a trend of impending violation is observed.

[0012] In another aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements a method for hazard assessment and proactive guidance for power construction safety supervision as described in any embodiment of the present invention.

[0013] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for hazard assessment and proactive guidance in power construction safety supervision as described in any embodiment of the present invention.

[0014] The present invention has the following beneficial effects: 1. By using a dynamic model scheduling and hot-replacement mechanism based on scene semantics, the system can intelligently load the most relevant lightweight identification model set according to the current operation scenario, which significantly improves the targeting, coverage and real-time performance of risk factor identification under limited edge computing power, and overcomes the problem of poor adaptability of fixed models.

[0015] 2. By introducing a three-dimensional voxelized static hazard level base map as prior knowledge and combining it with the dynamically identified risk observation point cloud in real time, recursive reasoning is performed within a dynamic Bayesian network framework to generate a smooth and continuous three-dimensional distribution of dynamic hazard levels. This method elevates risk assessment from discrete event alarms to continuous perception of the spatiotemporal evolution of risks, effectively suppressing transient false detection interference, and resulting in more accurate and reliable assessment results. 3. This invention not only provides graded and visualized AR early warning rendering based on dynamic risk distribution, but also generates and overlays a three-dimensional safety navigation path to avoid high-risk areas in real time when analyzing the trend of impending violations. This represents a fundamental advancement from "risk notification" to "safety guidance," enabling direct intervention in operational behavior, effectively preventing accidents, and significantly improving the initiative and effectiveness of safety supervision.

[0016] 4. Through a cloud-based collaborative evolution mechanism, the system can automatically collect data from high-risk, low-confidence scenarios, optimize models, and distribute updates, enabling the entire system to continuously learn and adapt to new environments and risks, thus ensuring the reliability of long-term applications. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0020] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0021] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0022] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0023] Example 1: To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments of the present application and with reference to the accompanying drawings.

[0024] To address the problems in existing technologies, this embodiment provides a method for hazard assessment and proactive guidance in power construction safety supervision, comprising the following steps: The smart glasses worn by construction workers collect first-person multimodal data of the construction scene; In this embodiment, multimodal data is collected using an RGB camera, a depth camera, and an IMU sensor integrated into the smart glasses worn by the construction worker. The collected multimodal data undergoes time-series alignment processing, with frame-level matching of the RGB images, depth images, and IMU pose data based on a unified time reference. Simultaneously, distortion correction and noise filtering are performed on the raw data to remove camera lens distortion and random noise from the IMU sensor. A Kalman filter algorithm is used to smooth the raw IMU data to obtain stable pose information; the state update formula for the smoothing process is: ; In the formula: for The estimated system state at time 10:00; This is the state transition matrix; for The estimated system state at time 10:00; To control the input matrix; for The control input vector at each time step; for Kalman gain at time step; for IMU measurement at time; This is the observation matrix.

[0025] The processed RGB image, depth image, and IMU pose data are packaged and uploaded to the edge computing terminal in real time via a low-latency wireless communication link.

[0026] At the edge computing end, multimodal data is parsed to obtain scene semantic categories and construction worker interaction behaviors; In a preferred embodiment of this practice, the step of parsing the multimodal data to obtain the scene semantic category and the construction worker's interaction behavior includes: Perform multi-task joint inference on RGB and depth images to output instance segmentation results and attribute labels for construction elements; Based on the position and category combination relationship of various instances in the instance segmentation results, the semantic category label of the current construction scene is inferred; Using construction element instances and construction workers as nodes, and their spatial relationships and movement intentions as edges, a spatiotemporal graph is constructed. Through graph neural network analysis, the work behavior classification labels and violation tendency probabilities are output as the interactive behaviors.

[0027] In this embodiment, scene semantic parsing is performed through multi-task joint reasoning on RGB and depth images. Using a shared backbone neural network, it simultaneously outputs instance segmentation results and attribute labels such as equipment voltage levels, the type and status of safety tools, and the boundary coordinates of hazardous areas for construction elements such as construction workers, electrical equipment, safety tools, and hazardous area boundaries. Furthermore, it infers scene semantic category labels such as working at heights, grounding wire installation, crossing fences, equipment maintenance, and power line erection by integrating the spatial relationships of all instances.

[0028] Interaction behavior analysis uses identified construction element instances and construction workers as graph nodes, and their spatial proximity, line-of-sight relationships, and action intentions analyzed based on IMU data as graph edges to construct a scene spatiotemporal graph. A graph convolutional network (GCN) is then run on this graph to analyze interactions between nodes and output work behavior classification labels and their probability of violation.

[0029] Based on the semantic category of the scene, a set of matching risk identification models is dynamically scheduled and hot-loaded from the model library to identify risk factors and their three-dimensional locations and confidence levels; In this embodiment, the scene adaptive model scheduling controller within the edge computing terminal queries its internally maintained scene-model correlation graph based on the received scene semantic category labels. This graph defines the probability of occurrence of each risk factor and the model performance under different scenarios. The controller, combined with current computing power and memory constraints, calculates and generates a loading list of a target risk identification model set; for example, in the "high-altitude operation" scenario, the list prioritizes "safety helmet detection model" and "safety belt detection model".

[0030] The model hot-replacement engine, based on the loading list, employs incremental memory mapping technology to dynamically load the target model from the model library into runtime memory, seamlessly replacing unnecessary models from the previous moment. By reserving dual model memory spaces, the engine switches model pointers atomically after the new model is loaded and verified in the background, achieving seamless replacement of models during runtime and ensuring uninterrupted inference flow. Subsequently, the currently active model set is used to perform parallel scene identification, outputting the existence confidence of various risk factors, such as "not wearing a safety helmet," and their position coordinates in three-dimensional space.

[0031] Obtain a three-dimensional voxelized static hazard level base map of the construction scene, with each voxel pre-marked with the basic risk probability; In this embodiment, a static hazard level base map matching the current work location is loaded from a pre-stored map database. This base map has divided the work environment into a fine grid using three-dimensional voxelization. Each voxel unit has a pre-defined basic risk probability value based on its associated equipment voltage level, historical accident data, safety procedures, and other information; for example, voxels near high-voltage busbars have a basic risk value close to 1.0.

[0032] The identified risk factors are mapped as a dynamic risk observation point cloud in voxel space; In this embodiment, the three-dimensional spatial location of each identified risk factor is transformed into a voxel coordinate system to determine its corresponding voxel index. Then, based on the existence confidence level of the risk factor, corresponding real-time risk weights are assigned to the voxel containing it and its adjacent voxels within its influence range. All weighted voxels together constitute the dynamic risk observation point cloud at the current moment, serving as real-time observation evidence.

[0033] Using a static hazard level base map as a priori and a dynamic risk observation point cloud as observation evidence, a dynamic hazard level three-dimensional distribution is generated through recursive Bayesian filtering via a dynamic Bayesian network. In this embodiment, a dynamic Bayesian network (DBN) is constructed, where the risk state of each voxel is a latent variable. The basic risk probability of each voxel in the static hazard level base map is used as the prior probability. The risk weights provided by the dynamic risk observation point cloud are transformed into observation likelihood probabilities. It also presets state transition probabilities that reflect the evolution of risk over time. .

[0034] At each time t, a recursive Bayesian filter is performed, fusing historical information with current observations to calculate the latest posterior probability of the hazard level for each voxel. The core update formula is: ; In the formula: Let be the posterior probability of the danger level of each voxel at time t; Let t be the observation likelihood probability with the dynamic risk observation point cloud as input; The risk state transition probability between time t and time t-1 is determined empirically based on the physical laws of risk diffusion in typical work processes or learned from historical data. This is a normalization constant; Let be the posterior probability of the voxel's hazard level at time t-1; The true danger state of the voxel at time t; This represents the observational evidence at time t. Through recursive calculation using this formula, instantaneous false detection noise is effectively smoothed, outputting a spatiotemporally continuous and stable three-dimensional dynamic hazard level distribution.

[0035] Based on the dynamic 3D distribution of hazard levels, graded AR early warning information is generated and rendered to smart glasses devices; As a preferred embodiment of this example, the generation of graded AR early warning information includes: configuring the visual attribute parameters of the superimposed AR visual elements differently according to the hazard level values ​​corresponding to different spatial locations in the three-dimensional distribution of dynamic hazard levels; the visual attribute parameters include at least one of color, transparency, flashing frequency, and boundary highlight intensity.

[0036] In this embodiment, a hierarchical rendering strategy based on visual saliency is executed according to the risk value corresponding to each spatial location in the three-dimensional distribution of dynamic hazard levels: For areas of fatal risk, generate high-frequency flashing, semi-transparent red warning blocks with parameters of high transparency and high flashing frequency.

[0037] For high-risk areas, a semi-transparent yellow bounding box and directional arrows are generated.

[0038] For areas of concern, generate a low-transparency blue outline.

[0039] The attribute parameters of the aforementioned visual elements are sent to the smart glasses, which then overlay and render them onto the real field of vision using their optical display unit.

[0040] When a trend of impending violation is observed, a 3D safety navigation path to avoid high-risk areas is generated and rendered in real time based on the dynamic 3D distribution of hazard levels.

[0041] In this embodiment, when the behavior analysis results indicate a clear trend of violation (such as a construction worker walking towards a dangerous area), the A-path planning algorithm is used to calculate in real time an optimal path from the construction worker's current location to the safe target point, avoiding all medium and high-risk voxel areas, based on the current dynamic hazard level 3D distribution and environmental topology map. This path is converted into a green strip-shaped virtual marker, which is sent out along with the warning information and rendered onto the real ground in the smart glasses' field of view, achieving proactive safety guidance.

[0042] As a preferred embodiment of this invention, a model co-evolution step is also included: When the edge computing terminal detects a high-risk event or the confidence level of the identified risk factor is lower than a preset threshold, it triggers the capture and uploading of de-identified multimodal data segments within the relevant time period; the cloud server aggregates data from multiple sources, performs incremental learning or adversarial training on a specific risk identification model, and generates an optimized model differential update package.

[0043] Receive model differential update packages generated from training based on multi-party data and sent from the cloud; The model hot replacement engine performs uninterrupted version upgrades of the corresponding risk identification models in the model library based on differential update packages, enabling the continuous evolution of the system's identification capabilities.

[0044] Example 2: This embodiment provides a hazard assessment and proactive guidance system for power construction safety supervision, including: The risk identification module dynamically schedules and hot-loads a set of matching risk identification models from the model library based on the semantic category of the scene in order to identify risk factors and their three-dimensional location and confidence level. The base map acquisition module acquires a three-dimensional voxelized static hazard level base map of the construction scene, with each voxel pre-marked with the basic risk probability. The risk mapping module maps the identified risk factors into a dynamic risk observation point cloud in voxel space; The joint judgment module uses a static hazard level base map as a priori and a dynamic risk observation point cloud as observation evidence. It then uses a dynamic Bayesian network to perform recursive Bayesian filtering to generate a dynamic hazard level three-dimensional distribution. The early warning feedback module generates graded AR early warning information based on the dynamic three-dimensional distribution of hazard levels and renders it to the smart glasses device. The proactive guidance module generates and renders a 3D safety navigation path to avoid high-risk areas in real time based on the dynamic 3D distribution of hazard levels when a trend of impending violation is observed.

[0045] Example 3: This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a method for hazard assessment and proactive guidance in power construction safety supervision as described in any embodiment of the present invention.

[0046] Example 4: This embodiment proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for hazard assessment and proactive guidance in power construction safety supervision as described in any embodiment of the present invention.

[0047] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0048] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. 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.

[0049] 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.

[0050] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a 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 part 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.

[0051] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for hazard assessment and proactive guidance in power construction safety supervision, characterized in that, Includes the following steps: The smart glasses worn by construction workers collect first-person multimodal data of the construction scene; At the edge computing end, multimodal data is parsed to obtain scene semantic categories and construction worker interaction behaviors; Based on the semantic category of the scene, a set of matching risk identification models is dynamically scheduled and hot-loaded from the model library to identify risk factors and their three-dimensional locations and confidence levels; Obtain a three-dimensional voxelized static hazard level base map of the construction scene, with each voxel pre-marked with the basic risk probability; The identified risk factors are mapped as a dynamic risk observation point cloud in voxel space; Using a static hazard level base map as a priori and a dynamic risk observation point cloud as observation evidence, a dynamic hazard level three-dimensional distribution is generated through recursive Bayesian filtering via a dynamic Bayesian network. Based on the dynamic 3D distribution of hazard levels, graded AR early warning information is generated and rendered to smart glasses devices; When a trend of impending violation is observed, a 3D safety navigation path to avoid high-risk areas is generated and rendered in real time based on the dynamic 3D distribution of hazard levels.

2. The method for hazard assessment and proactive guidance in power construction safety supervision according to claim 1, characterized in that: The step of dynamically scheduling and hot-loading a set of matching risk identification models from the model library based on the semantic category of the scene specifically includes: Based on the pre-set scene-model correlation map, determine the candidate risk identification model that is associated with the semantic category of the current scene; By combining the real-time computing power and memory resource constraints of the edge computing terminal, a target model set is selected from the candidate risk identification models; The model hot replacement engine loads the target model set into runtime memory without interrupting the multimodal data parsing process, so as to replace or expand the previously loaded model set.

3. The method for hazard assessment and proactive guidance in power construction safety supervision according to claim 2, characterized in that: The method also includes a model co-evolution step: When the edge computing terminal detects a high-risk event or the confidence level of the identified risk factor is lower than a preset threshold, it triggers the capture and uploading of de-identified multimodal data segments within the relevant time period. Receive model differential update packages generated from training based on multi-party data and sent from the cloud; The model hot replacement engine performs uninterrupted version upgrades of the corresponding risk identification models in the model library based on differential update packages.

4. The method for hazard assessment and proactive guidance in power construction safety supervision according to claim 1, characterized in that: The recursive Bayesian filtering using a dynamic Bayesian network recursively calculates the posterior probability of the danger level of each voxel at time t using the following formula: ； In the formula: Let be the posterior probability of the danger level of each voxel at time t; Let t be the observation likelihood probability with the dynamic risk observation point cloud as input; Let be the risk state transition probability between time t and time t-1; This is a normalization constant; Let be the posterior probability of the voxel's hazard level at time t-1; The true danger state of the voxel at time t; This provides observational evidence at time t.

5. The method for hazard assessment and proactive guidance in power construction safety supervision according to claim 1, characterized in that: The generation of graded AR early warning information includes: configuring the visual attribute parameters of the superimposed AR visual elements differently based on the hazard level values ​​corresponding to different spatial locations in the three-dimensional distribution of dynamic hazard levels; the visual attribute parameters include at least one of color, transparency, flashing frequency, and boundary highlight intensity.

6. The method for hazard assessment and proactive guidance in power construction safety supervision according to claim 1, characterized in that: The real-time generation and rendering of a 3D safe navigation path that avoids high-risk areas specifically involves: Based on the dynamic hazard level 3D distribution and work space topology, a path planning algorithm is used to dynamically calculate a 3D spatial path from the current pose to the safe target point, avoiding all areas where the hazard level exceeds the preset threshold. The three-dimensional spatial path is converted into a strip-shaped virtual sign with visual guidance effect, which is then overlaid and rendered onto the real-world scene presented by the smart glasses device.

7. The method for hazard assessment and proactive guidance in power construction safety supervision according to claim 1, characterized in that: The analysis of multimodal data yields scene semantic categories and construction worker interaction behaviors, including: Perform multi-task joint inference on RGB and depth images to output instance segmentation results and attribute labels for construction elements; Based on the position and category combination relationship of various instances in the instance segmentation results, the semantic category label of the current construction scene is inferred; Using construction element instances and construction workers as nodes, and their spatial relationships and movement intentions as edges, a spatiotemporal graph is constructed. Through graph neural network analysis, the work behavior classification labels and violation tendency probabilities are output as the interactive behaviors.

8. A hazard assessment and proactive guidance system for power construction safety supervision, characterized in that, include: The data acquisition module uses smart glasses worn by construction workers to collect first-person, multimodal data of the construction scene. The edge computing module parses multimodal data at the edge computing end to obtain scene semantic categories and construction worker interaction behaviors; The risk identification module dynamically schedules and hot-loads a set of matching risk identification models from the model library based on the semantic category of the scene in order to identify risk factors and their three-dimensional location and confidence level. The base map acquisition module acquires a three-dimensional voxelized static hazard level base map of the construction scene, with each voxel pre-marked with the basic risk probability. The risk mapping module maps the identified risk factors into a dynamic risk observation point cloud in voxel space; The joint judgment module uses a static hazard level base map as a priori and a dynamic risk observation point cloud as observation evidence. It then uses a dynamic Bayesian network to perform recursive Bayesian filtering to generate a dynamic hazard level three-dimensional distribution. The early warning feedback module generates graded AR early warning information based on the dynamic three-dimensional distribution of hazard levels and renders it to the smart glasses device. The proactive guidance module generates and renders a 3D safety navigation path to avoid high-risk areas in real time based on the dynamic 3D distribution of hazard levels when a trend of impending violation is observed.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a method for hazard assessment and proactive guidance in power construction safety supervision as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a method for hazard assessment and proactive guidance in power construction safety supervision as described in any one of claims 1 to 7.

Images