Aerial work safety belt state recognition method and system

Abstract

The application discloses a kind of high-altitude operation safety belt state identification method and system, belong to operation risk detection technical field, above-mentioned method includes, obtain the human posture data and hook stress data of high-altitude operation personnel current period, and construct load posture coupling feature vector after input to built-in preset risk identification model, identify whether safety belt state is in virtual hanging state;Wherein, preset risk identification model is trained to obtain with historical load posture coupling feature vector as input, with safety belt state identification result as output;In the case where the safety belt state is identified to be in virtual hanging state, carry out hook virtual hanging early warning.Through implementation of the application, it can solve the problem that only hall switch detection buckle opening and closing state is used in prior art to judge whether safety belt is worn standard, cannot realize the detection and early warning of virtual hanging phenomenon.

Description

Technical Field

[0001] This invention relates to the field of operational risk detection technology, and in particular to a method and system for identifying the status of safety belts in high-altitude operations. Background Technology

[0002] Working at heights is a high-risk activity in industries such as power construction, building installation, and equipment maintenance. Safety belts, as the last line of defense for workers, are crucial for personal safety, and their proper use is directly related to safety. With the development of the Internet of Things and sensor technology, smart safety belts are increasingly being used on construction sites. These belts use sensors to monitor the wearing status and hook fastening of the safety belts, providing early warnings of any violations by workers.

[0003] Traditional safety belt monitoring solutions typically use Hall effect switches to detect the buckle's open / closed state and determine whether the safety belt is worn correctly through simple logic gate combinations. However, this method can only detect the physical open / closed state of the hook and cannot identify the phenomenon of "puppy hooking"—where the hook is attached to the structure but does not actually bear the weight of the worker. If the worker loses their footing, they will fall directly. Therefore, existing technology has the problem of not being able to detect and warn of puppy hooking. Summary of the Invention

[0004] This invention provides a method and system for identifying the status of safety belts for high-altitude operations, which can solve the problem that the existing technology only uses Hall switches to detect the opening and closing status of the buckle to determine whether the safety belt is worn correctly, and cannot detect and warn of the phenomenon of loose attachment.

[0005] An embodiment of the present invention provides a method for identifying the status of safety belts for high-altitude operations, applicable to edge computing gateways, comprising: The system acquires human posture data and hook force data of the high-altitude worker during the current time period; wherein, the human posture data is collected by the wearable terminal and transmitted to the edge computing gateway, and the hook force data is collected by the hook terminal and transmitted to the edge computing gateway. Based on the human posture data and hook force data, a load posture coupling feature vector is constructed. The load attitude coupling feature vector is input into the built-in preset risk identification model to identify whether the safety belt of the high-altitude worker is in a loose state; wherein, the preset risk identification model is trained by taking the historical load attitude coupling feature vector constructed based on historical human posture data and historical hook force data under several historical operation scenarios as input and the safety belt status identification result as output. If the seat belt is found to be in a state of loose attachment, a warning of loose attachment will be issued.

[0006] Furthermore, the construction of the load-posture coupling feature vector based on the human posture data and hook force data includes: Based on the hook force data, the resultant force amplitude, the rate of change of the resultant force amplitude, the hook force direction, the hook force direction angle, the rate of change of the hook force direction angle, the mean resultant force, the variance of the resultant force, and the peak resultant force of the current time period are calculated for each moment in the current time period. Based on the human posture data, the rate of change of human posture angle, human posture angle and human gravity direction at each moment within the current time period are calculated. Based on the rate of change of the human body posture angle, determine the first moment when the rate of change of the human body posture angle exceeds the first preset rate of change threshold. Based on the rate of change of the combined force amplitude, a second moment when the rate of change of the combined force amplitude exceeds a preset second rate of change threshold is determined; The attitude load response delay time is calculated based on the first time point and the second time point; Based on the combined force amplitude and the human body posture angle, the change in the combined force amplitude and the change in the human body posture angle are calculated. The ratio of attitude load response amplitude is calculated based on the change in the magnitude of the combined force and the change in the human body posture angle. Based on the direction of force on the hook and the direction of gravity of the person, the consistency deviation of the force direction angle and the projection matching degree are calculated; Based on the mean of the combined force, the variance of the combined force, the peak value of the combined force, the rate of change of the amplitude of the combined force, the rate of change of the force direction angle of the hook, the human body posture angle, the posture load response delay time, the change in the amplitude of the combined force, the change in the human body posture angle, the ratio of the amplitude of the posture load response, the projection matching degree, and the consistency deviation of the force direction angle, the load posture coupling feature vector is constructed.

[0007] Furthermore, the preset risk identification model is trained using a historical load-posture coupling feature vector constructed based on historical human posture data and historical hook force data from several historical work scenarios as input, and the seat belt status identification result as output, including: Add a real label to each historical load attitude coupling feature vector; wherein, the real label is used to characterize the real seat belt state corresponding to the historical load attitude coupling feature vector; The historical load attitude coupling feature vector with real labels is transmitted to the cloud platform, so that the cloud platform can construct decision trees based on the historical load attitude coupling feature vector with real labels, the preset minimum number of samples for node splitting, the preset maximum depth, and the preset number of decision trees, and obtain several completed decision trees; based on all the completed decision trees, the preset risk identification model is generated, and the preset risk identification model is stored and then transmitted to the edge computing gateway.

[0008] Furthermore, after issuing a warning about false hook-ups, it also includes: Acquire the first human posture data, the first hook force data, and the first timestamp within a preset time period before the warning time; Acquire the second human posture data, the second hook force data, and the second timestamp within a preset time period after the warning time; Based on the current human posture data, the current hook force data, the first human posture data, the first hook force data, the second human posture data, and the second hook force data, calculate and generate the composite force change curve, the force direction angle change curve, and the human posture angle change curve. The first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve are transmitted to the cloud platform so that the cloud platform can incrementally optimize the currently stored preset risk identification model based on the first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve, and feed the optimized preset risk identification model back to the edge computing gateway. The current preset risk identification model is updated based on the optimized preset risk identification model.

[0009] Furthermore, it also includes: Based on the hook force data, a dual-rope force balance index is calculated to characterize the degree of difference in load between the main hook and the auxiliary hook. Based on human posture data, the variance of the composite acceleration for the current time period is calculated; If the variance of the combined acceleration is greater than a preset movement threshold and the dual-rope force balance index exceeds a preset balance threshold, an early warning of uneven force distribution on the dual ropes will be issued.

[0010] Based on the above-described method embodiments, the present invention provides a high-altitude work safety belt status recognition system, comprising: Wearable devices, hook devices, and edge computing gateways; The wearable terminal and the hook terminal are respectively communicatively connected to the edge computing gateway; The wearable terminal is used to collect human posture data of high-altitude workers at the current time and transmit it to the edge computing gateway. The hook terminal is used to collect the hook force data of the high-altitude workers during the current time period and transmit it to the edge computing gateway; The edge computing gateway is used to acquire the human posture data and hook force data of the high-altitude worker during the current time period; based on the human posture data and hook force data, a load posture coupling feature vector is constructed; the load posture coupling feature vector is input to a built-in preset risk identification model to identify whether the safety belt of the high-altitude worker is in a loose state; wherein, the preset risk identification model is trained by taking historical load posture coupling feature vectors constructed based on historical human posture data and historical hook force data in several historical operation scenarios as input and safety belt status identification results as output; when the safety belt is identified as being in a loose state, a hook loose state warning is issued.

[0011] Furthermore, the construction of the load-posture coupling feature vector based on the human posture data and hook force data includes: Based on the hook force data, the resultant force amplitude, the rate of change of the resultant force amplitude, the hook force direction, the hook force direction angle, the rate of change of the hook force direction angle, the mean resultant force, the variance of the resultant force, and the peak resultant force of the current time period are calculated for each moment in the current time period. Based on the human posture data, the rate of change of human posture angle, human posture angle and human gravity direction at each moment within the current time period are calculated. Based on the rate of change of the human body posture angle, determine the first moment when the rate of change of the human body posture angle exceeds the first preset rate of change threshold. Based on the rate of change of the combined force amplitude, a second moment when the rate of change of the combined force amplitude exceeds a preset second rate of change threshold is determined; The attitude load response delay time is calculated based on the first time point and the second time point; Based on the combined force amplitude and the human body posture angle, the change in the combined force amplitude and the change in the human body posture angle are calculated. The ratio of attitude load response amplitude is calculated based on the change in the magnitude of the combined force and the change in the human body posture angle. Based on the direction of force on the hook and the direction of gravity of the person, the consistency deviation of the force direction angle and the projection matching degree are calculated; Based on the mean of the combined force, the variance of the combined force, the peak value of the combined force, the rate of change of the amplitude of the combined force, the rate of change of the force direction angle of the hook, the human body posture angle, the posture load response delay time, the change in the amplitude of the combined force, the change in the human body posture angle, the ratio of the amplitude of the posture load response, the projection matching degree, and the consistency deviation of the force direction angle, the load posture coupling feature vector is constructed.

[0012] Furthermore, this also includes: cloud platforms; The edge computing gateway is used to construct a historical load posture coupling feature vector based on historical human posture data and historical hook force data under several historical operation scenarios; add a real label to each historical load posture coupling feature vector; wherein, the real label is used to characterize the real seat belt state corresponding to the historical load posture coupling feature vector; and transmit the historical load posture coupling feature vector with the real label to the cloud platform. The cloud platform constructs decision trees based on historical load attitude coupling feature vectors with real labels, a preset minimum number of samples for node splitting, a preset maximum depth, and a preset number of decision trees, resulting in several completed decision trees. Based on all the completed decision trees, the preset risk identification model is generated, and the preset risk identification model is stored and then transmitted to the edge computing gateway.

[0013] Furthermore, the edge computing gateway is also used for: After issuing a false hook-attachment warning, the system acquires the first human posture data, the first hook force data, and the first timestamp within a preset time period before the warning time. Acquire the second human posture data, the second hook force data, and the second timestamp within a preset time period after the warning time; Based on the current human posture data, the current hook force data, the first human posture data, the first hook force data, the second human posture data, and the second hook force data, calculate and generate the composite force change curve, the force direction angle change curve, and the human posture angle change curve. The first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve are transmitted to the cloud platform. The cloud platform is used to incrementally optimize the current preset risk identification model based on the first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve, and to feed the optimized preset risk identification model back to the edge computing gateway. The edge computing gateway is also used to update the built-in current preset risk identification model according to the optimized preset risk identification model.

[0014] Furthermore, the edge computing gateway is also used for: Based on the hook force data, a dual-rope force balance index is calculated to characterize the degree of difference in load between the main hook and the auxiliary hook. Based on human posture data, the variance of the composite acceleration for the current time period is calculated; If the variance of the combined acceleration is greater than a preset movement threshold and the dual-rope force balance index exceeds a preset balance threshold, an early warning of uneven force distribution on the dual ropes will be issued.

[0015] The embodiments of the present invention have the following beneficial effects: This invention provides a method, device, terminal equipment, and storage medium for identifying the status of safety belts used in high-altitude operations. The method includes: acquiring human posture data and hook force data of the worker during the current time period; wherein the human posture data is collected by a wearable terminal and transmitted to an edge computing gateway, and the hook force data is collected by a hook terminal and transmitted to the edge computing gateway; constructing a load posture coupling feature vector based on the human posture data and hook force data; inputting the load posture coupling feature vector into a built-in preset risk identification model to identify whether the safety belt of the worker is in a loose attachment state; wherein the preset risk identification model is trained using historical load posture coupling feature vectors constructed based on historical human posture data and historical hook force data from several historical work scenarios as input and safety belt status identification results as output; and issuing a hook loose attachment warning when the safety belt is identified as being in a loose attachment state. Therefore, in this invention, by inputting the load posture coupling feature vector constructed based on human posture data and hook force data into a preset risk identification model, the system identifies whether the safety belt of the high-altitude worker is in a false attachment state, and issues an early warning when it is determined to be in a false attachment state, thus solving the problem that the prior art cannot detect the false attachment state. Attached Figure Description

[0016] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a method for identifying the status of safety belts for high-altitude operations, provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of a high-altitude work safety belt status recognition system provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0021] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0022] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment 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 refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0023] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0024] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0025] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0026] See Figure 1 To address the problem that existing technologies, which rely solely on Hall effect switches to detect the opening and closing of buckles to determine whether safety belts are worn correctly, cannot detect and warn of loose attachments, this invention provides a method for identifying the status of safety belts for high-altitude operations, applicable to edge computing gateways, comprising: Step S101: Obtain the human posture data and hook force data of the high-altitude worker during the current time period; wherein, the human posture data is collected by the wearable terminal and transmitted to the edge computing gateway, and the hook force data is collected by the hook terminal and transmitted to the edge computing gateway; Specifically, the aforementioned human posture data includes: the angular velocity and acceleration of the human body. The hook force data includes: the force components of the main hook along three mutually orthogonal axes, and the force components of the secondary hook along three mutually orthogonal axes.

[0027] Specifically, the wearable terminal is installed on the worker's safety belt to collect human posture data. In this invention, the wearable terminal includes a posture sensor, and the posture sensor adopts a six-axis inertial measurement unit to collect acceleration and angular velocity data of human movement.

[0028] Specifically, the hook terminal is set on the main hook and the auxiliary hook of the safety rope, and each collects the corresponding three-dimensional force data. In this invention, the hook terminal includes a three-dimensional force sensor and a wireless communication module 202. The three-dimensional force sensor is set in the force-bearing area at the root of the hook and is used to collect the axial tension, radial swing force and torque data of the hook, and output the force components of three mutually orthogonal axes.

[0029] Step S102: Based on the human posture data and hook force data, construct the load posture coupling feature vector; Specifically, before constructing the load-attitude coupling feature vector based on human posture data and hook force data, the sampling frequencies of the wearable terminal and the hook terminal may differ. For example, the posture sensor may have a sampling frequency of 100Hz, while the three-dimensional force sensor may have a sampling frequency of 50Hz. Therefore, the edge computing gateway first uses the timestamp information of the data and linear interpolation or cubic spline interpolation algorithms to align the sensor data with different sampling frequencies to the same time reference, ensuring the temporal consistency of subsequent feature extraction before constructing the feature vector.

[0030] Specifically, the load-attitude coupling feature vector mentioned above includes the time-domain statistical features of the composite force, the rate of change of the force's direction angle, the human body's posture angle, and the load-attitude related features. These features are integrated to obtain the load-attitude coupling feature vector.

[0031] In a preferred embodiment, the step of constructing a load-posture coupling feature vector based on the human posture data and hook force data includes: Based on the hook force data, the resultant force amplitude, the rate of change of the resultant force amplitude, the hook force direction, the hook force direction angle, the rate of change of the hook force direction angle, the mean resultant force, the variance of the resultant force, and the peak resultant force of the current time period are calculated for each moment in the current time period. Specifically, the combined force amplitude of the main hook and the auxiliary hook can be calculated using the following method: In the formula, Indicates the magnitude of the resultant force. Represents the force components along the x-axis. This represents the force component along the y-axis. This represents the force component along the z-axis.

[0032] Specifically, the direction of the force on the hook is calculated based on inverse trigonometric functions and the various force components. This is existing technology and will not be elaborated further here. The angle of the force direction on the hook is the angle between the direction of the force on the hook and the direction of gravity.

[0033] Specifically, each time period in this invention can be set to 2 seconds. The mean of the combined force is the arithmetic mean calculated based on the amplitude of all combined forces within 2 seconds, the variance of the combined force is the variance calculated based on the amplitude of all combined forces within 2 seconds, and the peak value of the combined force is the peak value extracted from the amplitude of all combined forces within 2 seconds.

[0034] Based on the human posture data, the rate of change of human posture angle, human posture angle and human gravity direction at each moment within the current time period are calculated. Specifically, based on the angular velocity and acceleration in the human posture data, the pitch and roll angles of the human body can be calculated. These pitch and roll angles are the human posture angles. Based on these human posture angles, the rate of change of the human posture angles can be further calculated. Subsequently, based on the human posture angles, the projection of gravity in the sensor coordinate system can be calculated, thus obtaining the direction of human gravity. This is existing technology and will not be elaborated further here.

[0035] Based on the rate of change of the human body posture angle, determine the first moment when the rate of change of the human body posture angle exceeds the first preset rate of change threshold. Specifically, the moment when the rate of change of the human body posture angle exceeds the first preset rate of change threshold within the current 2 seconds is recorded as the first moment. The first preset rate of change threshold can be set to 30° / s.

[0036] Based on the rate of change of the combined force amplitude, a second moment when the rate of change of the combined force amplitude exceeds a preset second rate of change threshold is determined; Specifically, the moment when the rate of change of the resultant force amplitude exceeds the second preset rate of change threshold within the current 2 seconds is recorded as the second moment. The second preset rate of change threshold can be set to 10% / s.

[0037] The attitude load response delay time is calculated based on the first time point and the second time point; Based on the combined force amplitude and the human body posture angle, the change in the combined force amplitude and the change in the human body posture angle are calculated. The ratio of attitude load response amplitude is calculated based on the change in the magnitude of the combined force and the change in the human body posture angle. Specifically, the absolute value of the ratio of the change in the magnitude of the resultant force to the change in the human body posture angle is taken as the posture load response amplitude ratio.

[0038] Based on the direction of force on the hook and the direction of gravity of the person, the consistency deviation of the force direction angle and the projection matching degree are calculated; Specifically, the standard deviation or maximum value of the angle deviation between the direction of force applied to the hook and the direction of human weight is taken as the consistency deviation of the force direction angle. The projection matching degree can be calculated based on the following formula: In the formula, Indicates the projection matching degree. It represents the weight of the human body.

[0039] Based on the mean of the combined force, the variance of the combined force, the peak value of the combined force, the rate of change of the amplitude of the combined force, the rate of change of the force direction angle of the hook, the human body posture angle, the posture load response delay time, the change in the amplitude of the combined force, the change in the human body posture angle, the ratio of the amplitude of the posture load response, the projection matching degree, and the consistency deviation of the force direction angle, the load posture coupling feature vector is constructed.

[0040] Specifically, the aforementioned load attitude coupling feature vectors will be used as input to the preset risk identification model. During the training process, the model will automatically learn the complex mapping relationship between these features and states such as virtual loading and normal loading, without the need for manual setting of rule thresholds.

[0041] In this preferred embodiment, a load posture coupling feature vector is constructed using human posture data and hook force data.

[0042] Step S103: Input the load attitude coupling feature vector into the built-in preset risk identification model to identify whether the safety belt of the high-altitude worker is in a loose state; wherein, the preset risk identification model is trained by taking the historical load attitude coupling feature vector constructed based on historical human posture data and historical hook force data under several historical operation scenarios as input and the safety belt status identification result as output. Specifically, the preset risk identification model in this invention is a model built based on random forest and trained using supervised learning on historical operation datasets from historical operation scenarios. Its input is a load attitude coupling feature vector, and its output is the safety belt status of the worker operating at height.

[0043] Specifically, the edge computing gateway couples the load attitude feature vector and inputs it into the pre-deployed pre-defined risk identification model. The model directly outputs the identification result of whether the seat belt is in a false attachment state.

[0044] Specifically, the model has learned through the training phase the complex nonlinear relationship between multidimensional features such as projection matching degree, response delay time, response amplitude ratio, and force direction angle consistency deviation and different recognition results. Compared with single threshold judgment, it has stronger anti-interference ability and adaptability.

[0045] For example, a low projection matching degree usually indicates a risk of false hooking, but if other features indicate that the low matching degree is caused by the worker turning around or moving laterally, the model may determine that there is no risk of false hooking, thereby reducing the false alarm rate.

[0046] In a preferred embodiment, the preset risk identification model is trained using a historical load-posture coupling feature vector constructed based on historical human posture data and historical hook force data under several historical work scenarios as input, and the seat belt status identification result as output, including: Add a real label to each historical load attitude coupling feature vector; wherein, the real label is used to characterize the real seat belt state corresponding to the historical load attitude coupling feature vector; Specifically, it collects raw sensor data from wearable terminals and hook terminals under different workers and different work scenarios (such as poles, iron towers, platforms, scaffolding, etc.) throughout history, including raw historical human posture data and raw historical hook three-dimensional force data.

[0047] Simultaneously, high-definition cameras will be used to record video of the operation process, which will serve as the basis for data annotation. A total of no fewer than 5,000 valid operation samples will be collected, with each sample containing the complete process from the start of attachment to the removal point, and the sampling time covering 5 to 30 seconds.

[0048] Subsequently, the raw sensor data is filtered and denoised (e.g., using a low-pass filter with a cutoff frequency of 20Hz), and then the data from different sampling frequencies are aligned to a unified time base (e.g., 100Hz) based on timestamp information using cubic spline interpolation.

[0049] Then, historical load attitude coupling feature vectors are extracted for each time window (e.g., a sliding window of 2 seconds with a step size of 0.5 seconds).

[0050] Then, based on the acquired video, the state of each historical load attitude coupling feature vector is labeled. The above state labels are divided into two categories: normal mounting and virtual mounting.

[0051] The historical load attitude coupling feature vector with real labels is transmitted to the cloud platform, so that the cloud platform can construct decision trees based on the historical load attitude coupling feature vector with real labels, the preset minimum number of samples for node splitting, the preset maximum depth, and the preset number of decision trees, and obtain several completed decision trees; based on all the completed decision trees, the preset risk identification model is generated, and the preset risk identification model is stored and then transmitted to the edge computing gateway.

[0052] Specifically, within the cloud platform, the labeled historical load attitude coupling feature vector is used as input, and the corresponding state category is used as the true label. The Random Forest algorithm is then used for training.

[0053] Specifically, the hyperparameters of the random forest are set as follows: 100 decision trees, 10 maximum depth, and 5 minimum samples for node splits.

[0054] During training, 5-fold cross-validation was used to evaluate model performance, with accuracy and recall as the main metrics.

[0055] After training, the preset risk identification model achieved an accuracy rate of over 95% in identifying spoofed states on the validation set.

[0056] Preferably, the aforementioned preset risk identification model can also be trained using the Lightweight Gradient Boosting Machine (LightGBM) algorithm.

[0057] Specifically, during the construction of each decision tree, for the node to be split, the following operations are performed: First, for the root node, the sample subset is all training samples obtained through Bootstrap sampling.

[0058] For other nodes, the sample subset is derived from the split of the parent node. Each sample in the subset contains a specific feature vector value and a corresponding true label.

[0059] Subsequently, a subset of all features from the current node's sample subset is randomly selected as candidate features.

[0060] For each candidate feature, sort the specific values ​​of that feature in all samples of the current subset, and use the midpoint of adjacent values ​​as the candidate split threshold.

[0061] For each candidate splitting threshold, the specific value of each sample on that feature is compared with the threshold. Samples with values ​​less than the threshold and their true labels are assigned to the left subset; otherwise, they are assigned to the right subset.

[0062] After the partitioning is completed, based on the true labels of the samples in the left and right subsets, the number of samples in each category is counted, and the Gini coefficient formula is used to calculate the purity index of the left and right subsets: In the formula, This represents the Gini coefficient corresponding to the node sample set D. This represents the proportion of the k-th type of samples (normal mounting or virtual mounting state) in the node sample set D.

[0063] Then, the purity indices of the left and right subsets are weighted and summed according to the proportion of their respective sample counts to the total number of samples in the current node, yielding the weighted Gini coefficient of the candidate feature and threshold combination. This value represents the quantification of the overall data disorder after splitting according to this rule.

[0064] Then, for all candidate features and all candidate thresholds, repeat the process of dividing left and right subsets and calculating the weighted Gini coefficient. Find the features and thresholds that minimize the weighted Gini coefficient and determine them as the optimal splitting rule for the current node. In the formula, This represents the weighted Gini coefficient. Indicates the left subset, Indicates the right subset. The purity index of the left subset. This represents the purity index of the right subset.

[0065] Specifically, the optimal splitting rule consists of selected features and specific thresholds (e.g., "projection matching degree M < 0.72").

[0066] Using the obtained optimal splitting rule, traverse each sample in the current node's sample subset and make a judgment based on its specific value on that feature: Samples that meet the criteria, along with their labels, are assigned to the left child node; those that do not meet the criteria are assigned to the right child node. Thus, the samples in the current node are divided in two.

[0067] Finally, the newly generated left and right child nodes are each treated as new "current nodes to be split", and the aforementioned sample partitioning steps are repeated.

[0068] This process continues on each branch until the number of samples contained in a node is less than the preset minimum number of split samples and the tree's growth depth reaches the preset maximum depth. The decision tree is complete when all branches' nodes no longer split.

[0069] In this preferred embodiment, the model was trained based on historical human posture data and historical hook force data from several historical work scenarios.

[0070] Step S104: If the seat belt is found to be in a state of loose attachment, issue a warning for loose attachment.

[0071] Specifically, when the model detects that the seat belt is not properly attached, it triggers an audible and visual alarm to remind workers to pay attention to safety.

[0072] In a preferred embodiment, after issuing a false hook-hanging warning, the method further includes: Acquire the first human posture data, the first hook force data, and the first timestamp within a preset time period before the warning time; Acquire the second human posture data, the second hook force data, and the second timestamp within a preset time period after the warning time; Based on the current human posture data, the current hook force data, the first human posture data, the first hook force data, the second human posture data, and the second hook force data, calculate and generate the composite force change curve, the force direction angle change curve, and the human posture angle change curve. The first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve are transmitted to the cloud platform so that the cloud platform can incrementally optimize the currently stored preset risk identification model based on the first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve, and feed the optimized preset risk identification model back to the edge computing gateway. The current preset risk identification model is updated based on the optimized preset risk identification model.

[0073] Specifically, after receiving the composite force change curve, the force direction angle change curve, and the human posture angle change curve, the cloud platform first retrieves the corresponding on-site surveillance video clips based on the first and second timestamps for verification, marking the current sample as either a genuine violation or a false alarm. Subsequently, the cloud platform performs an incremental optimization process based on these three curves.

[0074] First, extract the rise time of the composite force from its initial value to its peak value, the mean and variance of the steady-state phase (before and after the warning), the number of zero crossings of the curve, and the peak interval from the composite force change curve. Extract from the force direction angle variation curve: the smoothness of the direction angle variation (standard deviation within the sliding window), and the maximum delay time and correlation coefficient of the cross-correlation function between the direction angle and the human posture angle variation curve; Extract the following from the human posture angle change curve: the amplitude of the posture angle change and the main frequency component obtained by fast Fourier transform.

[0075] Subsequently, the extracted dynamic features are concatenated with the corresponding features in the original load attitude coupling feature vector (such as attitude load response delay time, projection matching degree, etc.) to form an enhanced feature vector, which is then combined with the real labels as incremental training samples and stored in the cloud incremental sample library.

[0076] Then, the preset risk identification model is optimized using any one or a combination of the following methods: Method 1 (Threshold Adaptive): For all leaf nodes of decision trees in the random forest, statistically analyze the distribution of key features (such as projection matching degree and response delay time) of samples belonging to "real violations", and adjust the judgment threshold of each leaf node to the median or quantile of this distribution. Method 2 (Incremental Training): When the number of incremental samples accumulated in the incremental sample library reaches a preset threshold (e.g., 100), the new samples are merged with the original training set, one or more new decision trees are retrained using the same hyperparameters, and the new trees are added to the original random forest to form an incremental random forest. Method 3 (Curve Template Matching): Three curves from typical "real traffic violation" scenarios are stored as positive sample templates in the cloud. During inference, the edge computing gateway can perform dynamic time-warped similarity matching between the real-time curves and the templates, and the results are used as auxiliary feature inputs for model decision-making.

[0077] Finally, the optimized model parameters (or newly added decision trees, threshold configurations, curve templates) are serialized and transmitted to the edge computing gateway via 4G / 5G network. The edge computing gateway hot-loads the new parameters or replaces the old model without interrupting its operation, thus completing the model update.

[0078] Specifically, through this physical-visual two-way verification mechanism, the accuracy and adaptability of the pre-set risk identification model are continuously improved.

[0079] To illustrate, in a high-altitude work scenario, workers climb poles, and wearable terminals collect human posture data in real time. Hook terminals on the main hook and auxiliary hook collect three-dimensional force data, and the edge computing gateway continuously performs data alignment and feature extraction.

[0080] After the workers climbed to the work position, they hung the main hook on the crossbeam, but it was not fully fastened. The hook was only loosely hung on the edge of the crossbeam and did not actually bear the weight of the human body.

[0081] At this point, the preset risk identification model detected that the human body was standing, but the three-dimensional force data of the hook showed an abnormal force direction (biased towards the horizontal direction) and a small amplitude.

[0082] At this point, the risk identification model, based on the load attitude coupling feature vector within the current time window (where the projection matching degree M is only 0.3 and the other features also deviate from the normal range), comprehensively determines that the hook is in a false hanging state, and the edge computing gateway immediately triggers a voice alarm: "The hook is not under load, please re-hook it."

[0083] After the task is completed, the edge computing network will upload the mechanical feature segments (i.e., the composite force change curve, the force direction angle change curve, and the human posture angle change curve) corresponding to the warning triggered in this task to the cloud platform.

[0084] The cloud platform retrieves the video from the corresponding time period for review. Once the violation is confirmed to be genuine, the corresponding sample is marked as a genuine violation sample for subsequent model optimization.

[0085] In this preferred embodiment, after issuing an early warning to workers at height, the model is further incrementally optimized based on human posture data and hook force data in adjacent time periods.

[0086] In another preferred embodiment, it further includes: Based on the hook force data, a dual-rope force balance index is calculated to characterize the degree of difference in load between the main hook and the auxiliary hook. Specifically, the above-mentioned double-rope force balance index is calculated based on the following formula: In the formula, E represents the force balance index of the two ropes. This indicates the magnitude of the resultant force of the main hook. This indicates the magnitude of the resultant force of the secondary hook.

[0087] Specifically, the value range of the above-mentioned double rope force balance index is 0-1. When the value of E is 0, it means that the double ropes are completely balanced. When the value of E is 1, it means that the hook is completely supported by a single rope.

[0088] Based on human posture data, the variance of the composite acceleration for the current time period is calculated; Specifically, the composite acceleration is first calculated based on the acceleration data in the human posture data, and then the variance of the composite acceleration is calculated based on all the composite accelerations in the current time period.

[0089] The combined acceleration is calculated based on the following formula: In the formula, Indicates the resultant acceleration. This represents acceleration along the x-axis. This represents acceleration in the y-axis direction. This represents the acceleration along the z-axis.

[0090] If the variance of the combined acceleration is greater than a preset movement threshold and the dual-rope force balance index exceeds a preset balance threshold, an early warning of uneven force distribution on the dual ropes will be issued.

[0091] Specifically, when the variance of the composite acceleration is greater than the preset movement threshold, it indicates that the worker is in a high-altitude movement state (such as climbing, walking, or changing attachment points), and conversely, it indicates that the worker is in a stationary or stable standing state.

[0092] Based on this, the system combines the dual-rope force balance index to determine the uneven force distribution between the two ropes. Specifically, if the variance of the combined acceleration is greater than the preset movement threshold and the dual-rope force balance index exceeds the preset balance threshold, the system determines that the system is in a state of uneven force distribution between the two ropes and issues an early warning for uneven force distribution between the two ropes.

[0093] Preferably, the aforementioned preset movement threshold and preset equalization threshold are calibrated using measured data, and the preset movement threshold ranges from 0.5 to 1.5, with units of m / s. 2In this invention, the preset movement threshold can be set to 1.0 and the preset equilibrium threshold can be set to 0.8.

[0094] As an illustration, when the worker moves on the crossarm, with the main hook hanging high and the auxiliary hook dragging below, the calculated double rope force balance index is 0.9, exceeding the preset balance threshold of 0.8. Furthermore, the variance of the synthesized acceleration is 1.2, which also exceeds the preset movement threshold of 1.0. Therefore, it is determined that the worker is in a high-altitude movement state, and a voice warning is issued: "Uneven force on the double ropes, please adjust the hanging point height."

[0095] In this preferred embodiment, the determination of the force balance of the two ropes is achieved based on the hook force data and human posture data.

[0096] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0097] like Figure 2 As shown, an embodiment of the present invention provides a high-altitude work safety belt status recognition system, comprising: Wearable devices, hook devices, and edge computing gateways; The wearable terminal and the hook terminal are respectively communicatively connected to the edge computing gateway; The wearable terminal is used to collect human posture data of high-altitude workers at the current time and transmit it to the edge computing gateway. The hook terminal is used to collect the hook force data of the high-altitude workers during the current time period and transmit it to the edge computing gateway; The edge computing gateway is used to acquire the human posture data and hook force data of the high-altitude worker during the current time period; based on the human posture data and hook force data, a load posture coupling feature vector is constructed; the load posture coupling feature vector is input to a built-in preset risk identification model to identify whether the safety belt of the high-altitude worker is in a loose state; wherein, the preset risk identification model is trained by taking historical load posture coupling feature vectors constructed based on historical human posture data and historical hook force data in several historical operation scenarios as input and safety belt status identification results as output; when the safety belt is identified as being in a loose state, a hook loose state warning is issued.

[0098] Specifically, the edge computing gateway communicates with the wearable terminal and the hook terminal respectively, adopts an ARM architecture embedded processor, and has built-in time-series data processing algorithms and preset risk identification models to execute the seat belt status identification method of the present invention. The edge computing gateway has local storage and real-time computing capabilities and can work independently in a network-free environment.

[0099] Specifically, the aforementioned human posture data includes: the angular velocity and acceleration of the human body. The hook force data includes: the force components of the main hook along three mutually orthogonal axes, and the force components of the secondary hook along three mutually orthogonal axes.

[0100] Specifically, the wearable terminal is installed on the worker's safety belt to collect human posture data. In this invention, the wearable terminal includes a posture sensor, and the posture sensor adopts a six-axis inertial measurement unit to collect acceleration and angular velocity data of human movement.

[0101] Specifically, the hook terminal is set on the main hook and the auxiliary hook of the safety rope, and each collects the corresponding three-dimensional force data. In this invention, the hook terminal includes a three-dimensional force sensor and a wireless communication module 202. The three-dimensional force sensor is set in the force-bearing area at the root of the hook and is used to collect the axial tension, radial swing force and torque data of the hook, and output the force components of three mutually orthogonal axes.

[0102] Specifically, before constructing the load-attitude coupling feature vector based on human posture data and hook force data, the sampling frequencies of the wearable terminal and the hook terminal may differ. For example, the posture sensor may have a sampling frequency of 100Hz, while the three-dimensional force sensor may have a sampling frequency of 50Hz. Therefore, the edge computing gateway first uses the timestamp information of the data and linear interpolation or cubic spline interpolation algorithms to align the sensor data with different sampling frequencies to the same time reference, ensuring the temporal consistency of subsequent feature extraction before constructing the feature vector.

[0103] Specifically, the load-attitude coupling feature vector mentioned above includes the time-domain statistical features of the composite force, the rate of change of the force's direction angle, the human body's posture angle, and the load-attitude related features. These features are integrated to obtain the load-attitude coupling feature vector.

[0104] In a preferred embodiment, the step of constructing a load-posture coupling feature vector based on the human posture data and hook force data includes: Based on the hook force data, the resultant force amplitude, the rate of change of the resultant force amplitude, the hook force direction, the hook force direction angle, the rate of change of the hook force direction angle, the mean resultant force, the variance of the resultant force, and the peak resultant force of the current time period are calculated for each moment in the current time period. Specifically, the combined force amplitude of the main hook and the auxiliary hook can be calculated using the following method: In the formula, Indicates the magnitude of the resultant force. Represents the force components along the x-axis. This represents the force component along the y-axis. This represents the force component along the z-axis.

[0105] Specifically, the direction of the force on the hook is calculated based on inverse trigonometric functions and the various force components. This is existing technology and will not be elaborated further here. The angle of the force direction on the hook is the angle between the direction of the force on the hook and the direction of gravity.

[0106] Specifically, each time period in this invention can be set to 2 seconds. The mean of the combined force is the arithmetic mean calculated based on the amplitude of all combined forces within 2 seconds, the variance of the combined force is the variance calculated based on the amplitude of all combined forces within 2 seconds, and the peak value of the combined force is the peak value extracted from the amplitude of all combined forces within 2 seconds.

[0107] Based on the human posture data, the rate of change of human posture angle, human posture angle and human gravity direction at each moment within the current time period are calculated. Specifically, based on the angular velocity and acceleration in the human posture data, the pitch and roll angles of the human body can be calculated. These pitch and roll angles are the human posture angles. Based on these human posture angles, the rate of change of the human posture angles can be further calculated. Subsequently, based on the human posture angles, the projection of gravity in the sensor coordinate system can be calculated, thus obtaining the direction of human gravity. This is existing technology and will not be elaborated further here.

[0108] Based on the rate of change of the human body posture angle, determine the first moment when the rate of change of the human body posture angle exceeds the first preset rate of change threshold. Specifically, the moment when the rate of change of the human body posture angle exceeds the first preset rate of change threshold within the current 2 seconds is recorded as the first moment. The first preset rate of change threshold can be set to 30° / s.

[0109] Based on the rate of change of the combined force amplitude, a second moment when the rate of change of the combined force amplitude exceeds a preset second rate of change threshold is determined; Specifically, the moment when the rate of change of the resultant force amplitude exceeds the second preset rate of change threshold within the current 2 seconds is recorded as the second moment. The second preset rate of change threshold can be set to 10% / s.

[0110] The attitude load response delay time is calculated based on the first time point and the second time point; Based on the combined force amplitude and the human body posture angle, the change in the combined force amplitude and the change in the human body posture angle are calculated. The ratio of attitude load response amplitude is calculated based on the change in the magnitude of the combined force and the change in the human body posture angle. Based on the direction of force on the hook and the direction of gravity of the person, the consistency deviation of the force direction angle and the projection matching degree are calculated; Specifically, the standard deviation or maximum value of the angle deviation between the direction of force applied to the hook and the direction of human weight is taken as the consistency deviation of the force direction angle. The projection matching degree can be calculated based on the following formula: In the formula, Indicates the projection matching degree. It represents the weight of the human body.

[0111] Based on the mean of the combined force, the variance of the combined force, the peak value of the combined force, the rate of change of the amplitude of the combined force, the rate of change of the force direction angle of the hook, the human body posture angle, the posture load response delay time, the change in the amplitude of the combined force, the change in the human body posture angle, the ratio of the amplitude of the posture load response, the projection matching degree, and the consistency deviation of the force direction angle, the load posture coupling feature vector is constructed.

[0112] Specifically, the aforementioned load attitude coupling feature vectors will be used as input to the preset risk identification model. During the training process, the model will automatically learn the complex mapping relationship between these features and states such as virtual loading and normal loading, without the need for manual setting of rule thresholds.

[0113] In another preferred embodiment, it further includes: a cloud platform; The edge computing gateway is used to construct a historical load posture coupling feature vector based on historical human posture data and historical hook force data under several historical operation scenarios; add a real label to each historical load posture coupling feature vector; wherein, the real label is used to characterize the real seat belt state corresponding to the historical load posture coupling feature vector; and transmit the historical load posture coupling feature vector with the real label to the cloud platform. Specifically, the edge computing gateway collects raw sensor data from wearable terminals and hook terminals under different workers and different work scenarios (such as poles, iron towers, platforms, scaffolding, etc.) throughout history, including raw historical human posture data and raw historical hook three-dimensional force data.

[0114] Meanwhile, the edge computing gateway will simultaneously record video of the operation process using a high-definition camera, which will serve as the basis for data annotation. At least 5,000 valid operation samples will be collected, each sample containing the complete process from the start of attachment to the detachment point, with a sampling duration ranging from 5 to 30 seconds.

[0115] Subsequently, the edge computing gateway filters and denoises the raw sensor data (e.g., using a low-pass filter with a cutoff frequency of 20Hz), and then aligns the data from different sampling frequencies to a unified time base (e.g., 100Hz) based on timestamp information using cubic spline interpolation.

[0116] Then, the edge computing gateway extracts historical load attitude coupling feature vectors for each time window (e.g., a sliding window of 2 seconds with a step size of 0.5 seconds), and then, based on the acquired video, performs state labeling on each historical load attitude coupling feature vector. The above state labeling is divided into two categories: normal mounting and virtual mounting.

[0117] The cloud platform constructs decision trees based on historical load attitude coupling feature vectors with real labels, a preset minimum number of samples for node splitting, a preset maximum depth, and a preset number of decision trees, resulting in several completed decision trees. Based on all the completed decision trees, the preset risk identification model is generated, and the preset risk identification model is stored and then transmitted to the edge computing gateway.

[0118] Specifically, the cloud platform and the edge computing gateway are connected via 4G / 5G network communication. The cloud platform receives relevant feature data transmitted by the edge computing gateway and verifies and updates model parameters based on the video collected by the on-site surveillance cameras.

[0119] Specifically, within the cloud platform, the labeled historical load attitude coupling feature vector is used as input, and the corresponding state category is used as the true label. The Random Forest algorithm is then used for training.

[0120] Specifically, the hyperparameters of the random forest are set as follows: 100 decision trees, 10 maximum depth, and 5 minimum samples for node splits.

[0121] During training, 5-fold cross-validation was used to evaluate model performance, with accuracy and recall as the main metrics. After training, the model achieved an accuracy of over 95% in identifying dummy states on the validation set.

[0122] Preferably, the above model can also be trained using the Lightweight Gradient Boosting Machine (LightGBM) algorithm.

[0123] In another preferred embodiment, the edge computing gateway is further configured to: After issuing a false hook-attachment warning, the system acquires the first human posture data, the first hook force data, and the first timestamp within a preset time period before the warning time. Acquire the second human posture data, the second hook force data, and the second timestamp within a preset time period after the warning time; Based on the current human posture data, the current hook force data, the first human posture data, the first hook force data, the second human posture data, and the second hook force data, calculate and generate the composite force change curve, the force direction angle change curve, and the human posture angle change curve. The first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve are transmitted to the cloud platform. The cloud platform is used to incrementally optimize the current preset risk identification model based on the first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve, and to feed the optimized preset risk identification model back to the edge computing gateway. The edge computing gateway is also used to update the built-in current preset risk identification model according to the optimized preset risk identification model.

[0124] In another preferred embodiment, the edge computing gateway is further configured to: Based on the hook force data, a dual-rope force balance index is calculated to characterize the degree of difference in load between the main hook and the auxiliary hook. Based on human posture data, the variance of the composite acceleration for the current time period is calculated; If the variance of the combined acceleration is greater than a preset movement threshold and the dual-rope force balance index exceeds a preset balance threshold, an early warning of uneven force distribution on the dual ropes will be issued.

[0125] It should be noted that the device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort. The above schematic diagrams are merely examples of a high-altitude work safety belt status recognition system and do not constitute a limitation on a high-altitude work safety belt status recognition system. It may include more or fewer components than illustrated, or combine certain components, or use different components.

[0126] The above are preferred embodiments of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying the status of safety belts used in high-altitude operations, applicable to edge computing gateways, characterized in that, include: The system acquires human posture data and hook force data of the high-altitude worker during the current time period; wherein, the human posture data is collected by the wearable terminal and transmitted to the edge computing gateway, and the hook force data is collected by the hook terminal and transmitted to the edge computing gateway. Based on the human posture data and hook force data, a load posture coupling feature vector is constructed. The load attitude coupling feature vector is input into the built-in preset risk identification model to identify whether the safety belt of the high-altitude worker is in a loose state; wherein, the preset risk identification model is trained by taking the historical load attitude coupling feature vector constructed based on historical human posture data and historical hook force data under several historical operation scenarios as input and the safety belt status identification result as output. If the seat belt is found to be in a state of loose attachment, a warning of loose attachment will be issued.

2. The method for identifying the status of a safety belt for high-altitude operations according to claim 1, characterized in that, The load-posture coupling feature vector is constructed based on the human posture data and hook force data, including: Based on the hook force data, the resultant force amplitude, the rate of change of the resultant force amplitude, the hook force direction, the hook force direction angle, the rate of change of the hook force direction angle, the mean resultant force, the variance of the resultant force, and the peak resultant force of the current time period are calculated for each moment in the current time period. Based on the human posture data, the rate of change of human posture angle, human posture angle and human gravity direction at each moment within the current time period are calculated. Based on the rate of change of the human body posture angle, determine the first moment when the rate of change of the human body posture angle exceeds the first preset rate of change threshold. Based on the rate of change of the combined force amplitude, a second moment when the rate of change of the combined force amplitude exceeds a preset second rate of change threshold is determined; The attitude load response delay time is calculated based on the first time point and the second time point; Based on the combined force amplitude and the human body posture angle, the change in the combined force amplitude and the change in the human body posture angle are calculated. The ratio of attitude load response amplitude is calculated based on the change in the magnitude of the combined force and the change in the human body posture angle. Based on the direction of force on the hook and the direction of gravity of the person, the consistency deviation of the force direction angle and the projection matching degree are calculated; Based on the mean of the combined force, the variance of the combined force, the peak value of the combined force, the rate of change of the amplitude of the combined force, the rate of change of the force direction angle of the hook, the human body posture angle, the posture load response delay time, the change in the amplitude of the combined force, the change in the human body posture angle, the ratio of the amplitude of the posture load response, the projection matching degree, and the consistency deviation of the force direction angle, the load posture coupling feature vector is constructed.

3. The method for identifying the status of a safety belt for high-altitude operations according to claim 2, characterized in that, The preset risk identification model is trained by taking a historical load-posture coupling feature vector, constructed based on historical human posture data and historical hook force data from several historical work scenarios, as input and the seat belt status identification result as output, including: Add a real label to each historical load attitude coupling feature vector; wherein, the real label is used to characterize the real seat belt state corresponding to the historical load attitude coupling feature vector; The historical load attitude coupling feature vector with real labels is transmitted to the cloud platform, so that the cloud platform can construct decision trees based on the historical load attitude coupling feature vector with real labels, the preset minimum number of samples for node splitting, the preset maximum depth, and the preset number of decision trees, and obtain several completed decision trees; based on all the completed decision trees, the preset risk identification model is generated, and the preset risk identification model is stored and then transmitted to the edge computing gateway.

4. The method for identifying the status of a safety belt for high-altitude operations according to claim 3, characterized in that, After issuing a warning about a false hook-up, the following is also included: Acquire the first human posture data, the first hook force data, and the first timestamp within a preset time period before the warning time; Acquire the second human posture data, the second hook force data, and the second timestamp within a preset time period after the warning time; Based on the current human posture data, the current hook force data, the first human posture data, the first hook force data, the second human posture data, and the second hook force data, calculate and generate the composite force change curve, the force direction angle change curve, and the human posture angle change curve. The first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve are transmitted to the cloud platform so that the cloud platform can incrementally optimize the currently stored preset risk identification model based on the first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve, and feed the optimized preset risk identification model back to the edge computing gateway. The current preset risk identification model is updated based on the optimized preset risk identification model.

5. The method for identifying the status of a safety belt for high-altitude operations according to claim 4, characterized in that, Also includes: Based on the hook force data, a dual-rope force balance index is calculated to characterize the degree of difference in load between the main hook and the auxiliary hook. Based on human posture data, the variance of the composite acceleration for the current time period is calculated; If the variance of the combined acceleration is greater than a preset movement threshold and the dual-rope force balance index exceeds a preset balance threshold, an early warning of uneven force distribution on the dual ropes will be issued.

6. A safety belt status recognition system for high-altitude operations, characterized in that, include: Wearable devices, hook devices, and edge computing gateways; The wearable terminal and the hook terminal are respectively communicatively connected to the edge computing gateway; The wearable terminal is used to collect human posture data of high-altitude workers at the current time and transmit it to the edge computing gateway. The hook terminal is used to collect the hook force data of the high-altitude workers during the current time period and transmit it to the edge computing gateway; The edge computing gateway is used to acquire the human posture data and hook force data of the high-altitude worker in the current time period; based on the human posture data and hook force data, a load posture coupling feature vector is constructed; the load posture coupling feature vector is input to the built-in preset risk identification model to identify whether the safety belt of the high-altitude worker is in a loose state; wherein, the preset risk identification model is trained by taking the historical load posture coupling feature vector constructed based on historical human posture data and historical hook force data in several historical operation scenarios as input and the safety belt status identification result as output; If the seat belt is found to be in a state of loose attachment, a warning of loose attachment will be issued.

7. A high-altitude work safety belt status recognition system according to claim 6, characterized in that, The load-posture coupling feature vector is constructed based on the human posture data and hook force data, including: Based on the hook force data, the resultant force amplitude, the rate of change of the resultant force amplitude, the hook force direction, the hook force direction angle, the rate of change of the hook force direction angle, the mean resultant force, the variance of the resultant force, and the peak resultant force of the current time period are calculated for each moment in the current time period. Based on the human posture data, the rate of change of human posture angle, human posture angle and human gravity direction at each moment within the current time period are calculated. Based on the rate of change of the human body posture angle, determine the first moment when the rate of change of the human body posture angle exceeds the first preset rate of change threshold. Based on the rate of change of the combined force amplitude, a second moment when the rate of change of the combined force amplitude exceeds a preset second rate of change threshold is determined; The attitude load response delay time is calculated based on the first time point and the second time point; Based on the combined force amplitude and the human body posture angle, the change in the combined force amplitude and the change in the human body posture angle are calculated. The ratio of attitude load response amplitude is calculated based on the change in the magnitude of the combined force and the change in the human body posture angle. Based on the direction of force on the hook and the direction of gravity of the person, the consistency deviation of the force direction angle and the projection matching degree are calculated; Based on the mean of the combined force, the variance of the combined force, the peak value of the combined force, the rate of change of the amplitude of the combined force, the rate of change of the force direction angle of the hook, the human body posture angle, the posture load response delay time, the change in the amplitude of the combined force, the change in the human body posture angle, the ratio of the amplitude of the posture load response, the projection matching degree, and the consistency deviation of the force direction angle, the load posture coupling feature vector is constructed.

8. A high-altitude work safety belt status recognition system according to claim 7, characterized in that, Also includes: Cloud platform; The edge computing gateway is used to construct a historical load posture coupling feature vector based on historical human posture data and historical hook force data under several historical operation scenarios; add a real label to each historical load posture coupling feature vector; wherein, the real label is used to characterize the real seat belt state corresponding to the historical load posture coupling feature vector; and transmit the historical load posture coupling feature vector with the real label to the cloud platform. The cloud platform constructs decision trees based on historical load attitude coupling feature vectors with real labels, a preset minimum number of samples for node splitting, a preset maximum depth, and a preset number of decision trees, resulting in several completed decision trees. Based on all the completed decision trees, the preset risk identification model is generated, and the preset risk identification model is stored and then transmitted to the edge computing gateway.

9. A high-altitude work safety belt status recognition system according to claim 8, characterized in that, The edge computing gateway is also used for: After issuing a false hook-attachment warning, the system acquires the first human posture data, the first hook force data, and the first timestamp within a preset time period before the warning time. Acquire the second human posture data, the second hook force data, and the second timestamp within a preset time period after the warning time; Based on the current human posture data, the current hook force data, the first human posture data, the first hook force data, the second human posture data, and the second hook force data, calculate and generate the composite force change curve, the force direction angle change curve, and the human posture angle change curve. The first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve are transmitted to the cloud platform. The cloud platform is used to incrementally optimize the current preset risk identification model based on the first timestamp, the second timestamp, the composite force change curve, the force direction angle change curve, and the human posture angle change curve, and to feed the optimized preset risk identification model back to the edge computing gateway. The edge computing gateway is also used to update the built-in current preset risk identification model according to the optimized preset risk identification model.

10. A high-altitude work safety belt status recognition system according to claim 9, characterized in that, The edge computing gateway is also used for: Based on the hook force data, a dual-rope force balance index is calculated to characterize the degree of difference in load between the main hook and the auxiliary hook. Based on human posture data, the variance of the composite acceleration for the current time period is calculated; If the variance of the combined acceleration is greater than a preset movement threshold and the dual-rope force balance index exceeds a preset balance threshold, an early warning of uneven force distribution on the dual ropes will be issued.

Images