Method, system and electronic device for identifying personal protective equipment wear compliance

Abstract

The application relates to a personal protective equipment wearing compliance identification method and system and electronic equipment. The method comprises the following steps: obtaining a two-dimensional image containing an operator wearing personal protective equipment; identifying the two-dimensional image to generate a human skeleton model and an equipment component model respectively; based on the human skeleton model and the equipment component model, a human-equipment semantic topology graph is constructed through hierarchical constraint solving; and based on a preset compliance constraint library, wearing compliance identification is performed on the human-equipment semantic topology graph to obtain an identification result. The method solves the technical problem that the prior art cannot accurately determine whether correct physical connection and structural topology relationship are formed between equipment components according to safety specifications, and can only identify the 'existence' and 'approximate position' of components, resulting in a high false alarm rate and poor reliability. The method achieves the technical effects of accurately restoring the real wearing structure of the equipment, effectively identifying hidden safety hazards and significantly reducing the false alarm rate and the missed detection rate.

Description

Technical Field

[0001] This application relates to the field of computer vision technology, and in particular to a method, system and electronic device for identifying compliance of personal protective equipment wearing. Background Technology

[0002] In high-risk industries such as high-altitude operations, power construction, and petrochemicals, correctly wearing personal protective equipment (PPE) is the last line of defense for protecting workers' lives. Taking a five-point harness as an example, its effectiveness depends not only on the presence of its components, but more importantly, on the correct structural topology between these components (such as D-rings, shoulder straps, leg straps, and chest straps). These components must be connected in a specific order and manner to form an effective load-bearing system. Incorrect wearing, such as twisted webbing, misaligned D-rings, or loose buckles, will render the harness ineffective in the event of a fall, posing a fatal risk.

[0003] Traditional PPE compliance checks primarily rely on visual inspections by on-site safety officers. These officers, based on personal experience and a sense of responsibility, observe to determine if the seatbelt components are in the correct position. This model has several inherent drawbacks: First, the inspection results are highly dependent on the inspector's subjective judgment, lacking unified and objective standards, and are prone to omissions or misjudgments due to factors such as experience and fatigue, leading to safety hazards. Second, manual inspections are sporadic and isolated, making it difficult to continuously monitor the entire work process and effectively cover all work areas and personnel, resulting in low management efficiency. Finally, inspection records are mostly verbal or paper-based, making data difficult to digitize, archive, and analyze, hindering the tracing of historical violations and preventing proactive risk prediction.

[0004] In recent years, with the development of computer vision and artificial intelligence technologies, some image-based automated PPE detection systems have emerged. These systems typically employ object detection or instance segmentation algorithms to automatically identify the presence and approximate location of PPE components such as helmets, seat belts, and reflective vests from surveillance videos or still images. This improves the objectivity and efficiency of detection to some extent. However, existing automated detection technologies still have significant limitations and cannot meet the need for accurate verification of the topological relationships of PPE wearable structures.

[0005] For example, most existing technologies remain at the level of determining whether PPE components "exist" and "are roughly located." These systems are often incapable of handling complex physical connections between components (such as whether webbing passes through the D-ring or whether buckles are locked), the order of wearing, and the integrity of the overall structure (such as whether a correct load-bearing path is formed). They can tell you "there is a D-ring" and "there is a shoulder strap," but they cannot determine "whether the shoulder strap is correctly connected to the D-ring." Furthermore, some methods attempting to resolve topological relationships may rely on expensive 3D sensors (such as depth cameras), multi-camera stereo vision systems, or complex environmental calibration. These solutions are costly in hardware, complex to deploy and maintain, and difficult to scale up in large industrial environments. Additionally, inferring connections solely based on simple geometric proximity (e.g., assuming two components are connected because they are close together) will generate numerous incorrect connection combinations in complex real-world scenarios (such as multi-person operations, occlusion, and changing lighting). This approach lacks domain knowledge guidance, leading to a "combinatorial explosion" problem, ultimately resulting in a very high false alarm rate and poor reliability.

[0006] There is currently no effective solution to the above problems. Summary of the Invention

[0007] This application provides a method, system, and electronic device for identifying the compliance of personal protective equipment (PPE) wear, in order to at least solve the technical problem that the prior art cannot accurately determine whether the various equipment components have formed the correct physical connection and structural topology relationship in accordance with safety specifications, and can only identify the "existence" and "approximate location" of the components, resulting in a high false alarm rate and poor reliability.

[0008] According to one aspect of the embodiments of this application, a method for identifying the compliance of personal protective equipment (PPE) wearing is provided, comprising: acquiring a two-dimensional image containing a worker wearing PPE; identifying the two-dimensional image to generate a human skeleton model and an equipment component model respectively; constructing a human-equipment semantic topology graph based on the human skeleton model and the equipment component model through hierarchical constraint solving; and performing PPE wearing compliance identification on the human-equipment semantic topology graph based on a preset compliance constraint library to obtain an identification result.

[0009] Optionally, recognizing the two-dimensional image and generating a human skeleton model and an equipment component model respectively includes: processing the two-dimensional image using a pose estimation algorithm to identify and locate key human skeletal points and generate the human skeleton model; and processing the two-dimensional image using a segmentation algorithm to identify and segment each independent key component of the personal protective equipment and generate the equipment component model containing the pixel mask of each component.

[0010] Optionally, based on the human skeleton model and the equipment component model, a human-equipment semantic topology graph is constructed through hierarchical constraint solving, including: using the human key point information extracted from the human skeleton model and the component information segmented from the equipment component model to determine semantic nodes carrying physical and functional attributes; filtering candidate edge sets based on a predefined connection compatibility matrix; the connection compatibility matrix is ​​used to determine whether a connection relationship is allowed between any two semantic nodes; based on the relative positional relationship provided by the human skeleton model and the component connection characteristics provided by the equipment component model, combined with a predefined global topology constraint library, performing graph structure optimization and solving on the candidate edge set; modeling the graph generation process as an integer linear programming problem, obtaining the combination of edges with the largest total weight under the premise of satisfying all global topology constraints, to obtain the human-equipment semantic topology graph.

[0011] Optionally, the connection compatibility matrix includes a mapping relationship between source node type, source connection point type, target node type, and target connection point type, and its element values ​​represent the validity of the connection; wherein, the source node type and target node type include human semantic nodes or equipment semantic nodes; the type of the human semantic node includes at least one of left shoulder, right shoulder, chest center, and hip center; the type of the equipment semantic node includes at least one of D-ring, shoulder strap, leg strap, chest strap, and buckle.

[0012] Optionally, the graph generation process can be modeled as an integer linear programming problem, including: decision variables being the selection state of each candidate edge; objective function being to maximize the total weight of the selected edges; and constraints transforming the rules in the global topological constraint library into linear equations or inequalities with respect to the decision variables.

[0013] Optionally, the rules in the global topology constraint library include at least one of degree constraints, path constraints, exclusivity constraints, and cyclic constraints; the degree constraint is used to specify the number of connections for a specific node; the path constraint is used to specify that there must be a continuous path from the first node to the second node; the exclusivity constraint is used to specify that direct connections are not allowed between two specific nodes; and the cyclic constraint is used to specify that the two endpoints of a single component cannot form a closed loop.

[0014] Optionally, wearability compliance identification is performed on the human-equipment semantic topology graph, including: performing topology compliance verification on the human-equipment semantic topology graph, checking whether the connection relationship between equipment components complies with safety specifications through graph traversal and pattern matching; and / or, performing alignment analysis between the human-equipment semantic topology graph and the human skeleton model, checking whether the spatial position of the equipment components relative to the human body is correct, and determining the final credibility score by combining the recognition confidence of key points of the human body.

[0015] Optionally, the final confidence score is obtained by multiplying the preliminary judgment result of the geometric position by the lower of the recognition confidence values ​​of the two human body key points used to define the safe area; wherein, the preliminary judgment result of the geometric position is used to indicate whether the center coordinates of the equipment component fall within the safe area determined by the two human body key points.

[0016] According to another aspect of the embodiments of this application, a personal protective equipment (PPE) wearing compliance identification system is provided, comprising: an acquisition module for acquiring a two-dimensional image containing a worker wearing PPE; a generation module for identifying the two-dimensional image and generating a human skeleton model and an equipment component model respectively; a construction module for constructing a human-equipment semantic topology graph based on the human skeleton model and the equipment component model through hierarchical constraint solving; and an identification module for performing wear compliance identification on the human-equipment semantic topology graph based on a preset compliance constraint library to obtain an identification result.

[0017] According to another aspect of the embodiments of this application, an electronic device is provided, including: a processor, and a memory storing a program, the program including instructions that, when executed by the processor, cause the processor to perform the methods described above.

[0018] In this embodiment, a two-dimensional image of a worker wearing personal protective equipment (PPE) is acquired; the two-dimensional image is identified to generate a human skeleton model and an equipment component model; a human-equipment semantic topology graph is constructed based on the human skeleton model and the equipment component model through hierarchical constraint solving; and wearing compliance identification is performed on the human-equipment semantic topology graph based on a preset compliance constraint library to obtain the identification result. This solves the technical problem that existing technologies cannot accurately determine whether the equipment components have formed the correct physical connection and structural topology relationship according to safety specifications, and can only identify the "existence" and "approximate location" of the components, resulting in high false alarm rate and poor reliability. This achieves the technical effect of accurately restoring the real wearing structure of personal protective equipment, effectively identifying hidden safety hazards such as twisted webbing and incorrect connections, and significantly reducing the false alarm rate and missed detection rate.

[0019] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0021] Figure 1 A flowchart illustrating the method for identifying compliance of personal protective equipment wearing as provided in this application embodiment;

[0022] Figure 2 A schematic diagram of a personal protective equipment (PPE) wearing compliance identification system provided in this application embodiment;

[0023] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0024] Embodiments of this application will now be described in more detail with reference to the accompanying drawings. While some embodiments of this application are shown in the drawings, it should be understood that embodiments of this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the embodiments of this application. It should be understood that the accompanying drawings and embodiments are for illustrative purposes only and are not intended to limit the scope of protection of this application.

[0025] As mentioned in the background section of this application, existing automated detection technologies generally face combinatorial explosion and high false alarm rates when constructing the wearable topology of complex equipment from two-dimensional images, especially under the challenge of uncalibrated, single-frame images. Based on this, this application proposes a method for identifying the compliance of personal protective equipment (PPE) wearability based on a "semantic node-connection compatibility matrix-global topological constraint" model, aiming to solve the problem of verifying the topological relationships of complex PPE wearable structures in single-frame, uncalibrated two-dimensional images. By introducing two levels of formal domain knowledge models—the "connection compatibility matrix" and the "global topological constraint"—the graph construction process is transformed from a crude "connect first, filter later" process into an accurate solution process guided by attributes and progressively layered. By transforming graph construction into a formally solvable optimization problem (such as integer linear programming), this application can efficiently and accurately identify the unique or optimal compliant topology from a massive number of candidate relationships, fundamentally solving the combinatorial explosion and high false alarm rate problems caused by existing technologies relying solely on geometric proximity judgments, and significantly improving the robustness and accuracy of constructing wearable structure topology relationships in complex scenarios.

[0026] According to one aspect of the embodiments of this application, a method for identifying the compliance of personal protective equipment wearing is provided. Figure 1 A flowchart of the method for identifying compliance of personal protective equipment wearing provided in the embodiments of this application is shown below. Figure 1 As shown, the method includes the following steps:

[0027] Step S102: Obtain a two-dimensional image containing a worker wearing personal protective equipment. The two-dimensional image is a single-frame, uncalibrated two-dimensional image.

[0028] Step S104: Recognize the two-dimensional image and generate a human skeleton model and equipment component model respectively.

[0029] Step S106: Based on the human skeleton model and equipment component model, a human-equipment semantic topology graph is constructed through hierarchical constraint solving. By constructing a "human-equipment semantic topology graph" carrying physical and functional attributes, and performing multi-level constraint solving based on this graph, a high-precision "human-equipment" compliance digital twin model is generated and verified.

[0030] Step S108: Based on the preset compliance constraint library, perform wear compliance recognition on the human body-equipment semantic topology graph to obtain the recognition result.

[0031] In this embodiment, a two-dimensional image of a worker wearing personal protective equipment (PPE) is acquired; the two-dimensional image is identified to generate a human skeleton model and an equipment component model; a human-equipment semantic topology graph is constructed based on the human skeleton model and the equipment component model through hierarchical constraint solving; and wearing compliance identification is performed on the human-equipment semantic topology graph based on a preset compliance constraint library to obtain the identification result. This solves the technical problem that existing technologies cannot accurately determine whether the equipment components have formed the correct physical connection and structural topology relationship according to safety specifications, and can only identify the "existence" and "approximate location" of the components, resulting in high false alarm rate and poor reliability. This achieves the technical effect of accurately restoring the real wearing structure of personal protective equipment, effectively identifying hidden safety hazards such as twisted webbing and incorrect connections, and significantly reducing the false alarm rate and missed detection rate.

[0032] It should be noted that the embodiments of this application have extremely high verification accuracy and reliability. Through the graph construction method of "semantic nodes-connection compatibility matrix-global topological constraints," the actual wearable structure can be accurately reconstructed, avoiding false judgments and thus discovering hidden security risks. Compared with traditional manual inspection, the false judgment rate can be significantly reduced; compared with existing automated methods, it has higher accuracy in verifying complex structural topological relationships.

[0033] Furthermore, the embodiments of this application have extremely low hardware costs and deployment requirements. They can be operated with just a regular 2D camera, without the need for expensive 3D sensors, multi-view cameras, or complex environmental calibration. This greatly lowers the threshold for the promotion and application of the technology and overcomes the dependence of some existing automation methods on high-cost hardware.

[0034] Optionally, after obtaining the recognition results, a structured verification report will be generated. This report will include at least the following:

[0035] Overall conclusion: Compliant, non-compliant, or partially uncertain.

[0036] List of Violations: A detailed list of all failed constraints and their descriptions.

[0037] Uncertainty List: Lists all "location compliance" checks that cannot be strongly judged due to low confidence levels of the relevant nodes.

[0038] Visual snapshot: An image frame marked with irregular or uncertain areas for visual visualization.

[0039] Furthermore, the verified Compliance Digital Twin (CDT) model and violation details are archived in a time-series format. Long Short-Term Memory (LSTM) networks can be used to model the historical violation sequences of specific workers to identify habitual violation patterns and generate predictive alerts. If non-compliance is verified, corrective guidance (such as virtual correct positions, adjustment arrows, etc.) can be rendered in the real-time video stream based on the CDT model, providing users with closed-loop correction guidance.

[0040] This application's embodiments achieve a leap from "passive response" to "proactive prediction." Through time-series analysis, it is possible to discern the patterns of risk evolution, identify "habitual violation patterns" of individuals or groups, and issue predictive alerts, thus enabling predictive and root-cause-based security management.

[0041] In some embodiments, recognizing a two-dimensional image and generating a human skeleton model and an equipment component model respectively includes: processing the two-dimensional image using a pose estimation algorithm to identify and locate key human skeletal points and generate a human skeleton model; and processing the two-dimensional image using a segmentation algorithm to identify and segment each independent key component of the personal protective equipment and generate an equipment component model containing pixel masks of each component.

[0042] The pose estimation algorithm can employ the OpenPose model based on deep learning. By detecting and associating human joints in the image, it constructs a human skeleton model containing 18 key skeletal points. These key skeletal points cover major movable joints such as the head, neck, shoulders, elbows, wrists, hips, knees, and ankles, accurately reflecting the human body's pose characteristics. The segmentation algorithm can use the Mask R-CNN model. Based on object detection, this model can generate pixel-level masks for each component of personal protective equipment, clearly distinguishing the regional range of different equipment components, such as helmets, safety shoes, protective gloves, and seat belts. The mask information for each component includes its position coordinates, size, and contour shape in the two-dimensional image, providing a precise component localization basis for subsequent equipment wearing compliance verification.

[0043] For example, pose estimation algorithms can be used to identify and locate key skeletal points of the human body from 2D images, generating a human skeleton sub-model. Instance segmentation algorithms can be used to identify and segment all independent key components of the PPE (such as D-rings, shoulder straps, leg straps, etc.) from 2D images, obtaining the pixel mask of each component.

[0044] In some embodiments, a human-equipment semantic topology graph is constructed based on a human skeleton model and an equipment component model through hierarchical constraint solving. This includes: determining semantic nodes carrying physical and functional attributes using key human point information extracted from the human skeleton model and component information segmented from the equipment component model; filtering candidate edge sets based on a predefined connection compatibility matrix; the connection compatibility matrix is ​​used to determine whether a connection relationship is allowed between any two semantic nodes. Each item of the matrix is ​​defined by a quadruple (source node type, source node connection point type, target node type, target node connection point type), and its value is a Boolean value, indicating whether the connection between a source node with a specified type and a specified connection point type and a target node with a specified type and a specified connection point type is valid under the correct PPE wearing specifications; optimizing and solving the graph structure of the candidate edge set based on the relative positional relationship provided by the human skeleton model and the component connection characteristics provided by the equipment component model, combined with a predefined global topology constraint library; and modeling the graph generation process as an integer linear programming problem, obtaining the combination of edges with the largest total weight under the premise of satisfying all global topology constraints to obtain the human-equipment semantic topology graph.

[0045] The global topology constraint library covers the hierarchical structure requirements, component spatial distribution rules, and functional coordination conditions that must be followed when wearing different types of PPE. For example, for a full-body seatbelt, the construction of its semantic topology graph must satisfy the priority connection relationship between the shoulder strap node and the back D-link node, and the lap belt node must be located within the preset height range of the hip key point. At the same time, the leg seatbelt node must form a closed loop structure with the lap belt node. In the graph structure optimization stage, by introducing an edge weight mechanism, the physical strength, functional necessity, and spatial adaptability of component connections are transformed into quantitative indicators. The weight value of key connection points (such as the main lock connection point of the seatbelt) is set to the highest to ensure that the core constraints are satisfied first during the solution process. The objective function of the integer linear programming problem is set to maximize the sum of edge weights. The constraints include node connectivity limits, component hierarchical relationships, and spatial location thresholds. By calling mature linear programming solvers (such as Gurobi and CPLEX), the optimal topology graph structure can be obtained quickly, providing an accurate semantic relationship foundation for subsequent wearability compliance verification.

[0046] HumanSemanticNode: These are key human body points extracted using pose estimation algorithms. They not only include two-dimensional coordinates (x, y) but are also assigned specific type identifiers (e.g., left shoulder LS, right shoulder RS, chest center TC, hip center HC) and confidence scores (machine vision recognition confidence; image quality and recognition algorithm jointly affect the calculation results, such as occlusion, reflection, or motion blur, which will reduce this value). These nodes constitute the baseline framework for equipment wear.

[0047] EquipmentSemanticNode: For each segmented equipment component, in addition to assigning its component type (e.g., back D-ring, chest strap, leg strap) and pixel mask, this application innovatively defines and extracts a "Connector" attribute. This attribute is predefined based on the component's physical shape and functional design; it encodes the specific region or point where the component functionally connects with other components, and is key to resolving connection ambiguity. Connection point extraction method:

[0048] 1) For linear components (such as webbing): The geometric centerline is extracted by skeletonizing the segmentation mask. The two endpoints of this centerline are defined as primary connection points. Additionally, equidistant sampling along the centerline can generate intermediate connection points to handle cross-connections such as those between a chest strap and a shoulder strap. Each connection point contains coordinates and a type (e.g., Endpoint, Midpoint).

[0049] 2) For regional components (such as D-rings and buckles): Define their functional connection areas based on the component's functional template (available from design drawings or standard samples). For example, the connection point of a D-ring is defined as its "ring-aperture" rather than a simple geometric centroid, which precisely expresses the functional constraint that "the webbing must pass through this area." The connection point of a buckle is defined as the "locking interface" of the male and female buckles.

[0050] Through this step, unstructured pixel information is transformed into a set of nodes containing rich prior knowledge and capable of logical reasoning.

[0051] To avoid inefficient geometric calculations on massive node pairs, a "connectivity compatibility matrix" is introduced as the first layer of constraint. This matrix is ​​a formalized domain knowledge base that encodes human experts' rules on the correct wearing of PPE in a structured and computable form, thereby enabling efficient pruning before combustible explosions occur.

[0052] Alternatively, the candidate edge generation process is as follows:

[0053] 1. Traverse all equipment semantic node pairs .

[0054] 2. Query the "Connectivity Compatibility Matrix" to determine... and Check if the types and their respective combinations of "connection points" are compatible. If not, skip this step; this step eliminates the vast majority of illogical connections.

[0055] 3. If compatible, calculate geometric proximity between the regions defined by their respective "connection point" attributes. The calculation method is adaptively selected based on the connection point type: point-to-point connections use Euclidean distance; point-to-region connections use the minimum distance from the point to the region boundary; region-to-region connections use the region centroid distance or Hausdorff distance.

[0056] 4. If the geometric proximity is less than the preset threshold This generates a weighted candidate edge. Weight Designed as a function that comprehensively reflects connection reliability:

[0057]

[0058] in, It is a decreasing function of distance. and It is a function of node identification confidence. This ensures that connections that are closer and more clearly identified have higher initial weights.

[0059] In this embodiment, by using predefined node attributes and a compatibility matrix, computational resources are focused on a few meaningful connections, which greatly improves the efficiency and accuracy of candidate edge generation.

[0060] In some embodiments, the connection compatibility matrix includes a mapping relationship between source node type, source connection point type, target node type, and target connection point type, and its element values ​​represent the validity of the connection; wherein, the source node type and target node type include human semantic nodes or equipment semantic nodes; the type of human semantic node includes at least one of left shoulder, right shoulder, chest center, and hip center; the type of equipment semantic node includes at least one of D-ring, shoulder strap, leg strap, chest strap, and buckle; the connection point type is predefined according to the component shape and function. For linear components, the connection point type includes endpoints or intermediate points; for regional components, the connection point type includes annular through-hole areas or locking interfaces.

[0061] Optionally, this matrix explicitly defines which node pairs have potential valid connections by combining and mapping different types of source nodes with target nodes and their corresponding connection point types. For example, when the source node type is "shoulder strap" in the equipment semantic node and its source connection point type is "endpoint," while the target node type is "left shoulder" in the human semantic node and its target connection point type is "endpoint," the corresponding element value in the matrix will be marked as valid, indicating that the two nodes have the possibility of being connected. Conversely, if the source node type is "buckle" and the target node type is "hip center," and their connection point types do not match, the matrix element value will be displayed as invalid, thereby directly excluding such meaningless node pair combinations and avoiding unnecessary subsequent geometric proximity calculations.

[0062] The connection compatibility matrix defines the validity of the connection between (source node type, source connection point type) and (target node type, target connection point type) (Boolean value: 1 / 0); the connection compatibility matrix is ​​shown in Table 1.

[0063] Table 1 Connection Compatibility Matrix

[0064]

[0065] In some embodiments, the graph generation process is modeled as an integer linear programming problem, including: decision variables being the selection state of each candidate edge; an objective function being to maximize the total weight of the selected edges; and constraints transforming rules in a global topological constraint library into linear equations or inequalities with respect to the decision variables.

[0066] This mathematical modeling approach optimizes candidate edge selection while ensuring the rationality of the topological structure. The weight of each candidate edge is determined by a combination of factors, including the verification results of the connectivity compatibility matrix, the geometric proximity between nodes, and equipment functional requirements. Rules in the global topology constraint library, such as hierarchical relationships between different types of equipment nodes and limitations on the number of key connection points, are transformed into strict linear constraints. This ensures that the final generated graph structure satisfies both the physical logic of equipment wear and the basic requirements of safety and compliance. When solving the integer linear programming problem, mature optimization algorithms, such as branch and bound or cutting plane methods, can be used to efficiently find the optimal solution to the objective function, thereby determining the final semantic node connection relationships of the equipment and providing an accurate topological foundation for subsequent wear compliance identification.

[0067] This application models the final graph generation process as an integer linear programming (ILP) problem, thereby transforming a fuzzy matching problem into a mathematical optimization problem with an exact solution.

[0068] 1. Decision variables: for each candidate edge Set a binary decision variable In this context, 1 indicates that the edge is selected, and 0 indicates that it is not selected.

[0069] 2. Objective Function: The objective is to find the subgraph with the largest total weight, i.e. ,in, Candidate edge The weight.

[0070] 3. Constraints: Transform all global topological constraints into constraints about... Linear equations or inequalities. For example, "the degree of the chest strap buckle node must be 2" can be transformed into... .

[0071] By using the ILP solver, the system can efficiently find the combination of edges with the largest total weight while satisfying all global constraints, thus obtaining a unique, globally optimal, and highly accurate "human-equipment semantic topology graph." This method fundamentally solves the problems of "combinatorial explosion" and "high false alarm rate" caused by relying solely on local geometric proximity judgments.

[0072] In some embodiments, the rules in the global topology constraint library include at least one of degree constraints, path constraints, exclusivity constraints, and cycle constraints; degree constraints are used to specify the number of connections for a specific node; path constraints are used to specify that there must be a continuous path from the first node to the second node; exclusivity constraints are used to specify that direct connections are not allowed between two specific nodes; and cycle constraints are used to specify that the two endpoints of a single component cannot form a closed loop.

[0073] During the traversal, the potential connection relationships of each pair of nodes are verified one by one according to these constraint types. For degree constraints, it checks whether the number of existing connections of a specific node meets the specified upper or lower limit; path constraints verify whether there is a continuous path between specified nodes that meets the conditions through a search algorithm; exclusivity constraints directly exclude node pairs that are explicitly prohibited from being connected; and cycle constraints focus on checking whether the two ends of a single component form an unreasonable closed loop structure to ensure that the topological connection relationships of each component of the equipment meet safety specifications and functional requirements.

[0074] After generating the candidate edge set, the graph structure remains ambiguous (e.g., a buckle may be geometrically adjacent to multiple webbing ends). To address this issue, this application introduces a "global topology constraint library" as a second layer of constraints for the final graph structure solution. These constraints are higher-dimensional rules describing the overall structural rationality than the "connectivity compatibility matrix," ensuring that the final generated graph is structurally complete, correct, and unique.

[0075] Example of constraint library content: These constraints are stored in graph schema or first-order logic expression form, derived from security standards and best practices.

[0076] Degree constraints: For example, "the total number of connections (degrees) of a chest strap buckle node must be equal to 2"; "the number of connections at the end of a webbing must be 1".

[0077] Path constraints: For example, "There must be a continuous path from the back D-ring to any leg strap through the main webbing and shoulder straps."

[0078] Exclusivity constraints: For example, "No direct connection is allowed between the left shoulder strap and the right leg strap".

[0079] Cyclic constraints: For example, "the two ends of any single webbing component cannot be directly or indirectly connected to form a closed loop."

[0080] In some embodiments, wear compliance identification of the human-equipment semantic topology graph includes: performing topology compliance verification on the human-equipment semantic topology graph, checking whether the connection relationship between equipment components complies with safety specifications through graph traversal and pattern matching; and / or, performing alignment analysis between the human-equipment semantic topology graph and the human skeleton model, checking whether the spatial position of the equipment components relative to the human body is correct, and determining the final credibility score by combining the recognition confidence of key points of the human body.

[0081] Optionally, for the topology compliance verification step, various constraint rules in the global topology constraint library will be called sequentially according to a preset order. Taking degree constraint as an example, when traversing to the chest strap buckle node, the current number of connected edges will be automatically counted. If the detection result is not equal to 2, the node connection relationship will be marked as abnormal, and the specific degree value deviation will be recorded.

[0082] For path constraint verification, a depth-first search algorithm is used. Starting from the D-node on the back, all possible path branches are explored step by step. It tracks in real time whether a continuous path passing through the main webbing and shoulder straps reaches each leg strap. If a path is interrupted or deviates from the preset component sequence, a path compliance alarm is triggered. Exclusivity constraint checks are implemented by constructing a prohibited connection matrix. When a direct connection edge is detected between the left shoulder strap node and the right leg strap node, the connection is judged as illegal and highlighted according to the rules in the constraint library. During the verification of cyclic constraints, a depth-first traversal is performed on the endpoints of each webbing component to check for closed-loop paths that start from one endpoint, pass through other nodes, and return to another endpoint. Once a closed loop is found, the node sequence and connection path forming the loop are recorded, providing detailed evidence for subsequent compliance assessments.

[0083] After topology compliance verification is completed, all abnormal connections, path issues, degree mismatches, etc., will be summarized to generate a preliminary topology compliance report. Next, the spatial location verification phase begins. The system first maps the coordinates of each equipment component node in the human-equipment semantic topology graph to the corresponding key points on the human skeleton model. For example, the 3D coordinates of the chest strap buckle node are compared with the coordinates of the human sternal manubrium key point, calculating the vertical distance and horizontal offset between them. If the Y-axis coordinate (height direction) of the chest strap buckle node is lower than a certain threshold of the human sternal manubrium key point coordinates, the chest strap height is determined to be non-compliant with safety regulations. For the verification of the back D-ring position, the relative position coordinates of the back D-ring point in the human back coordinate system are calculated using the seventh cervical vertebra and lumbar vertebra key points as reference benchmarks. If its X-axis coordinate (left-right direction) deviates from the human midline by more than a preset range, or its Z-axis coordinate (front-back direction) is too close to the human back surface, the D-ring position is determined to have a compliance issue.

[0084] During the spatial location verification process, the verification results are weighted by combining the confidence values ​​of key human body point detection. When the recognition confidence of a certain key human body part (such as the manubrium of the sternum or the seventh cervical vertebra) is lower than the set threshold, the confidence weight of the location verification result of the corresponding equipment part will be reduced accordingly. Finally, the overall wearability compliance confidence score is calculated by combining the topology compliance verification results and the spatial location verification results. When the score is higher than the set threshold, the wearability is judged to be compliant; otherwise, specific compliance issues and improvement suggestions are output.

[0085] Topology compliance verification is mainly used to check whether the connection relationship between equipment components complies with safety specifications. It directly performs graph traversal and pattern matching on the "equipment semantic topology graph".

[0086] 1) Example 1: Forced connection verification

[0087] The rule states: "The chest strap must be connected to both shoulder straps."

[0088] Verification logic: In the graph, verify whether the "chest strap buckle" node has a connection edge with both the "left shoulder strap" node and the "right shoulder strap" node.

[0089] 2) Example 2: Load-bearing path verification

[0090] The rule states: "There must be a complete load-bearing path from the D-ring at the back to each leg strap."

[0091] Verification logic: Starting from the "back D-ring" node, execute a graph search algorithm (such as DFS or BFS) to confirm whether there is a valid path through nodes such as "shoulder strap" and "main webbing" that can eventually reach the "left leg strap" and "right leg strap" nodes.

[0092] 3) Example 3: Disable connection verification

[0093] The rule states: "The safety rope hook must not be attached to the equipment loop on the chest."

[0094] Verification logic: Check the adjacent nodes of the "safety rope hook" node in the diagram to confirm that its type is not "equipment ring".

[0095] The results of such validations are usually explicit Boolean values ​​(pass / fail) because they are based on a defined graph structure.

[0096] Location compliance verification is primarily used to check whether the spatial position of equipment components relative to the human body is correct, and is crucial to ensuring the effectiveness of wearing the equipment. It performs alignment analysis between the "equipment semantic topology map" and the "human semantic node" framework, and introduces the confidence level of human nodes as the core criterion for judging reliability.

[0097] 1) Example 1: Verification of the position of the D-ring on the back

[0098] The rule states: "The D-ring on the back should be located between the shoulder blades, in the center."

[0099] Verification logic:

[0100] 1. Define the safe area: Based on the coordinates of the "left shoulder" and "right shoulder" human body nodes, define a "safe rectangular area" through geometric calculation.

[0101] 2. Geometric Position Determination: Determine whether the center coordinates of the "back D-ring" equipment node fall within this "safe rectangular area". Obtain preliminary geometric judgment results. .

[0102] 3. Introduce confidence adjustment: Obtain the confidence scores for the two human body nodes, "left shoulder" and "right shoulder". and .

[0103] 4. Calculate the final score: The final credibility score for this verification. The calculation is as follows:

[0104]

[0105] 5. Conclusion: Based on (range To make a judgment. For example, "High confidence level compliance"; "Low confidence level compliance"; It is then marked as "location uncertain" or "suspected violation", thus avoiding making erroneous strong judgments when the human posture is not accurately estimated.

[0106] 2) Example 2: Chest strap height verification

[0107] The rule states: "The chest band should be positioned in the middle of the sternum at an appropriate height."

[0108] Verification logic: Similarly, based on the Y coordinates of the "neck" and "abdomen" human body nodes, determine whether the Y coordinate of the "chest belt" node is within the allowable range, and give the final judgment by combining the confidence of the relevant human body nodes.

[0109] In some embodiments, the final confidence score is obtained by multiplying the preliminary judgment result of the geometric position by the lower of the recognition confidence values ​​of the two human body key points used to define the safe area; wherein the preliminary judgment result of the geometric position is used to indicate whether the center coordinates of the equipment component fall within the safe area determined by the two human body key points.

[0110] In the location compliance verification process, a method combining preliminary geometric location assessment and human keypoint recognition confidence level was used to calculate the final credibility score. Specifically, this method first performs a preliminary assessment of the geometric location (… This provides a basic compliance assessment in the form of true / false values. Then, it considers the recognition confidence of the two key human body points upon which the security region is defined. and (This will allow for further adjustments to the credibility of this initial assessment.)

[0111] Final credibility score ( ) is the result of a preliminary judgment of geometric position ( ) and the one with the lower confidence level among these two key points ( The result is calculated by multiplying the values ​​of the key points. This method ensures that even if the geometric position is roughly correct, the final compliance judgment will be less certain if one of the key points is not accurately identified (i.e., its confidence level is low). This avoids drawing overly confident conclusions when the human posture estimation is inaccurate, thus providing a more rigorous wearability compliance assessment.

[0112] According to another aspect of the embodiments of this application, a system for identifying compliance of personal protective equipment wearing is provided. Figure 2 A schematic diagram of a personal protective equipment (PPE) wearing compliance identification system provided in this application embodiment, such as... Figure 2 As shown, the personal protective equipment (PPE) wearing compliance identification system includes: an acquisition module 202, a generation module 204, a construction module 206, and an identification module 208. The following is a detailed description of this PPE wearing compliance identification system.

[0113] The acquisition module 202 is used to acquire a two-dimensional image containing a worker wearing personal protective equipment.

[0114] The generation module 204 is used to recognize two-dimensional images and generate human skeleton models and equipment component models respectively.

[0115] Module 206 is used to construct a human-equipment semantic topology graph based on the human skeleton model and equipment component model through hierarchical constraint solving.

[0116] The recognition module 208 is used to perform wear compliance recognition on the human-equipment semantic topology map based on a preset compliance constraint library, and obtain the recognition result.

[0117] In this embodiment, the system acquires a two-dimensional image of a worker wearing personal protective equipment (PPE); identifies the two-dimensional image to generate a human skeleton model and an equipment component model; constructs a human-equipment semantic topology graph based on the human skeleton model and equipment component model through hierarchical constraint solving; and performs wear compliance identification on the human-equipment semantic topology graph based on a preset compliance constraint library to obtain the identification result. This solves the technical problem of existing technologies being unable to accurately determine whether the equipment components have formed the correct physical connection and structural topology relationship according to safety specifications, and only being able to identify the "existence" and "approximate location" of the components, resulting in high false alarm rates and poor reliability. The system achieves the technical effect of accurately restoring the actual wearing structure of PPE, effectively identifying hidden safety hazards such as twisted webbing and incorrect connections, and significantly reducing false alarm and missed detection rates.

[0118] It should be noted that the above-mentioned acquisition module 202, generation module 204, construction module 206 and identification module 208 correspond to steps S102 to S108 in the method embodiment. The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above method embodiment.

[0119] According to another aspect of the embodiments of this application, an electronic device is provided, including: a processor, and a memory storing a program, the program including instructions that, when executed by the processor, cause the processor to perform the methods described above.

[0120] refer to Figure 3 The present invention describes a structural block diagram of an electronic device that can serve as a server or client in embodiments of this application, which is an example of a hardware device that can be applied to various aspects of this application. The electronic device is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the application described and / or claimed herein.

[0121] like Figure 3As shown, the electronic device includes a computing unit 301, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 302 or a computer program loaded from a storage unit 308 into a random access memory (RAM) 303. The RAM 303 may also store various programs and data required for the operation of the electronic device. The computing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0122] Multiple components in the electronic device are connected to I / O interface 305, including: input unit 306, output unit 307, storage unit 308, and communication unit 309. Input unit 306 can be any type of device capable of inputting information into the electronic device. Input unit 306 can receive input digital or character information and generate key signal inputs related to user settings and / or function control of the electronic device. Output unit 307 can be any type of device capable of presenting information and may include, but is not limited to, a display, speaker, video / audio output terminal, vibrator, and / or printer. Storage unit 308 may include, but is not limited to, disks and optical discs. Communication unit 309 allows the electronic device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks, and may include, but is not limited to, modems, network cards, infrared communication devices, and wireless communication transceivers, such as Bluetooth devices, WiFi devices, WiMax devices, cellular communication devices, and / or the like.

[0123] The computing unit 301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, CPUs, graphics processing units (GPUs), various special-purpose artificial intelligence (AI) computing units, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processors, controllers, microcontrollers, etc. The computing unit 301 performs the various methods and processes described above. For example, in some embodiments, the method embodiments of this application can be implemented as computer programs tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program can be loaded and / or installed on an electronic device via ROM 302 and / or communication unit 309. In some embodiments, the computing unit 301 can be configured to perform the methods described above by any other suitable means (e.g., by means of firmware).

[0124] It should be noted that the term "comprising" and its variations used in the embodiments of this application are open-ended, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; and the term "some embodiments" means "at least some embodiments". The modifications of "one" and "multiple" mentioned in the embodiments of this application are illustrative and not restrictive. Those skilled in the art should understand that, unless explicitly indicated otherwise in the context, they should be understood as "one or more".

[0125] The steps described in the method embodiments provided in this application can be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of protection of this application is not limited in this respect.

[0126] The term "embodiment" in this specification refers to a specific feature, structure, or characteristic described in connection with an embodiment that may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily imply the same embodiment, nor does it imply independence from or alternative to other embodiments. The various embodiments in this specification are described in a related manner, with reference to each other for similar or identical parts. In particular, for apparatus, device, and system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, and relevant details are referred to in the description of the method embodiments.

[0127] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of patent protection. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the appended claims.

Claims

1. A method for identifying compliance in the wearing of personal protective equipment, characterized in that, include: Acquire two-dimensional images containing workers wearing personal protective equipment; The two-dimensional image is identified, and a human skeleton model and an equipment component model are generated respectively. Based on the human skeleton model and the equipment component model, a human-equipment semantic topology graph is constructed through hierarchical constraint solving. This process includes: using key human point information extracted from the human skeleton model and component information segmented from the equipment component model to determine semantic nodes carrying physical and functional attributes; filtering candidate edge sets based on a predefined connection compatibility matrix; the connection compatibility matrix is ​​used to determine whether a connection relationship is allowed between any two semantic nodes; optimizing and solving the graph structure of the candidate edge set based on the relative positional relationships provided by the human skeleton model and the component connection characteristics provided by the equipment component model, combined with a predefined global topology constraint library; and modeling the graph generation process as an integer linear programming problem, obtaining the combination of edges with the largest total weight while satisfying all global topology constraints to obtain the human-equipment semantic topology graph. Based on a pre-defined compliance constraint library, wear compliance is identified in the human-equipment semantic topology graph, and the identification results are obtained.

2. The method according to claim 1, characterized in that, Recognize the two-dimensional image and generate a human skeleton model and an equipment component model, respectively, including: The two-dimensional image is processed using a pose estimation algorithm to identify and locate key human skeletal points, thereby generating the human skeleton model. The two-dimensional image is processed using a segmentation algorithm to identify and segment the individual key components of the personal protective equipment, generating a model of the equipment components containing pixel masks of each component.

3. The method according to claim 1, characterized in that, The connection compatibility matrix includes a mapping relationship between source node type, source connection point type, target node type, and target connection point type, and its element values ​​represent the validity of the connection; The source node type and target node type include human semantic nodes or equipment semantic nodes; The types of human semantic nodes include at least one of the following: left shoulder, right shoulder, chest center, and hip center; The types of equipment semantic nodes include at least one of D-rings, shoulder straps, leg straps, chest straps, and buckles.

4. The method according to claim 1, characterized in that, The graph generation process is modeled as an integer linear programming problem, including: The decision variable is the selection state of each candidate edge; The objective function is to maximize the total weight of the selected edges; Constraints transform the rules in the global topology constraint library into linear equations or inequalities with respect to decision variables.

5. The method according to claim 4, characterized in that, The rules in the global topology constraint library include at least one of degree constraints, path constraints, exclusive constraints, and cyclic constraints. The degree constraint is used to specify the number of connections for a particular node; The path constraint is used to specify that there must be a continuous path from the first node to the second node; The exclusivity constraint is used to specify that direct connections are not allowed between two specific nodes; The cyclic constraint is used to specify that the two endpoints of a single component cannot form a closed loop.

6. The method according to any one of claims 1 to 5, characterized in that, Perform wearability compliance identification on the human-equipment semantic topology graph, including: Topological compliance verification is performed on the human-equipment semantic topology graph. This involves checking whether the connections between equipment components comply with safety specifications through graph traversal and pattern matching; and / or, The semantic topology map of the human body and the human skeleton model are aligned and analyzed to check whether the spatial position of the equipment parts relative to the human body is correct, and the final credibility score is determined by combining the recognition confidence of the human body key points.

7. The method according to claim 6, characterized in that, The final confidence score is obtained by multiplying the preliminary judgment result of the geometric position by the lower of the recognition confidence values ​​of the two human body key points used to define the safe area; wherein, the preliminary judgment result of the geometric position is used to indicate whether the center coordinates of the equipment component fall within the safe area determined by the two human body key points.

8. A system for identifying compliance of personal protective equipment wearing, characterized in that, include: The acquisition module is used to acquire two-dimensional images containing workers wearing personal protective equipment; A generation module is used to recognize the two-dimensional image and generate a human skeleton model and an equipment component model, respectively. A construction module is used to construct a human-equipment semantic topology graph based on the human skeleton model and the equipment component model through hierarchical constraint solving. The construction of the human-equipment semantic topology graph based on the human skeleton model and the equipment component model through hierarchical constraint solving includes: determining semantic nodes carrying physical and functional attributes using human key point information extracted from the human skeleton model and component information segmented from the equipment component model; filtering a candidate edge set based on a predefined connection compatibility matrix; the connection compatibility matrix is ​​used to determine whether a connection relationship is allowed between any two semantic nodes; optimizing and solving the graph structure of the candidate edge set based on the relative positional relationships provided by the human skeleton model and the component connection characteristics provided by the equipment component model, combined with a predefined global topology constraint library; and modeling the graph generation process as an integer linear programming problem, obtaining the combination of edges with the largest total weight under the premise of satisfying all global topology constraints, thus obtaining the human-equipment semantic topology graph. The identification module is used to identify wearability compliance of the human-equipment semantic topology based on a preset compliance constraint library, and obtain the identification result.

9. An electronic device, comprising: A processor and a memory storing a program, characterized in that the program includes instructions that, when executed by the processor, cause the processor to perform the method of any one of claims 1 to 7.

Images