A visual monitoring and early warning system and method for preventing falling at height

Abstract

The present invention relates to the field of aerial work technology, and in particular to a visual monitoring and early warning system and method for preventing falls during aerial work. The method comprises the following steps: collecting physiological status data of workers to generate worker fatigue characteristics and an attention distraction index; collecting behavioral posture data to analyze movement regularity; collecting work environment parameters and evaluating the work platform stability index; cross-correlating worker fatigue characteristics, attention distraction index, movement regularity, and work platform stability index to obtain worker-environment interaction state data; extracting risk precursor features from the worker-environment interaction state data to form a multi-dimensional risk precursor feature set. The present invention constructs individual baseline parameters and behavioral pattern features based on historical safe work data, customizes exclusive risk assessment standards for each worker, and significantly improves the accuracy of early warnings.

Description

Technical Field

[0001] The present invention relates to the technical field of high-altitude operations, and in particular to a high-altitude operation anti-fall visual monitoring and early warning system and method. Background Art

[0002] High-altitude work refers to work performed at heights where there is a risk of falling, 2 meters or more above the fall height reference plane. This carries a high risk and requires strict safety management and protective measures. High-altitude work fall prevention measures generally ensure worker safety through multiple means, including personal protection (safety belts, helmets, fall arresters, etc.), collective protection (edge ​​guardrails, opening protection, etc.), and technical protection (anchor systems, horizontal lifeline systems). High-altitude work fall prevention visual monitoring is a system that uses modern technologies (such as 5G, the Internet of Things, artificial intelligence, and sensors) to provide real-time monitoring and early warning of the safety status of workers working at heights.

[0003] However, traditional visual monitoring and early warning methods for fall prevention during high-altitude work often suffer from the following issues: Existing systems generally use uniform thresholds for risk assessment, failing to account for individual differences among workers, such as work habits, physiological characteristics, and professional skill levels. Most fall prevention systems are only activated passively when an accident occurs, lacking the ability to identify and warn of hazardous factors in advance, making them incapable of predictive risk prevention and control. Summary of the Invention

[0004] Based on this, it is necessary for the present invention to provide a visual monitoring and early warning system and method for preventing falls during high-altitude operations to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above-mentioned purpose, a visual monitoring and early warning method for preventing falls during high-altitude operations is provided, comprising the following steps:

[0006] Step S1: Collect the worker's physiological state data to generate the worker's fatigue characteristics and attention distraction index; collect behavioral posture data to analyze the regularity of the movement; collect work environment parameters and evaluate the work platform stability index;

[0007] Step S2: Interactively correlate worker fatigue characteristics, distraction index, movement regularity, and work platform stability index to obtain worker-environment interaction status data; extract risk precursor features from the worker-environment interaction status data to form a multidimensional risk precursor feature set;

[0008] Step S3: Establish individual baseline parameters based on pre-acquired historical safe operation data, and record typical operation postures to construct behavioral pattern characteristics; calculate the operation difficulty coefficient based on the operation type complexity and environmental condition change rate in the historical safe operation data;

[0009] Step S4: Generate a personalized risk threshold matrix based on individual baseline parameters, behavioral pattern characteristics, and task difficulty coefficient; perform real-time risk deviation calculation on the personalized risk threshold matrix and multi-dimensional risk precursor feature set, predict the risk development trajectory, and assess the potential fall risk probability to form a risk prediction report;

[0010] Step S5: Generate graded warnings and visual feedback based on the risk prediction report, form a high-altitude operation risk situation map, and send an alarm to the operator through the call system.

[0011] This aspect also provides a visual monitoring and early warning system for preventing falls during high-altitude operations, which is used to implement the above-mentioned visual monitoring and early warning method for preventing falls during high-altitude operations. The visual monitoring and early warning system for preventing falls during high-altitude operations includes:

[0012] The multi-source data acquisition module is used to collect workers' physiological status data to generate worker fatigue characteristics and attention distraction index; collect behavioral posture data to analyze movement regularity; collect work environment parameters and evaluate the work platform stability index;

[0013] The interactive feature extraction module is used to interactively correlate worker fatigue characteristics, attention distraction index, movement regularity, and work platform stability index to obtain worker-environment interaction status data; and to extract risk precursor features from the worker-environment interaction status data to form a multi-dimensional risk precursor feature set;

[0014] The baseline parameter and behavior pattern module is used to establish individual baseline parameters based on pre-acquired historical safety operation data, and record typical operation postures to construct behavior pattern characteristics; the operation difficulty coefficient is calculated based on the complexity of the operation type and the rate of change of environmental conditions in the historical safety operation data;

[0015] The risk assessment and prediction module is used to generate a personalized risk threshold matrix based on individual baseline parameters, behavioral pattern characteristics, and task difficulty coefficients. It calculates the risk deviation degree in real time based on the personalized risk threshold matrix and multi-dimensional risk precursor feature set, predicts the risk development trajectory, and assesses the potential fall risk probability to form a risk prediction report.

[0016] The early warning and visual feedback module is used to generate graded early warnings and visual feedback based on the risk prediction report, form a high-altitude operation risk situation map, and issue alarms to operators through the call system.

[0017] This invention addresses the issues of traditional fall prevention systems, which often lack consideration for individual differences and experience delayed response, by constructing a multi-dimensional, intelligent, and predictable safety assurance system for high-altitude work. Wearable intelligent sensing devices collect real-time information on multiple physiological and behavioral parameters of workers (such as heart rate, body temperature, gravitational acceleration, and posture angle), enabling dynamic perception of the worker's current state. This processing step can reflect the worker's concentration, fatigue level, and potential abnormal behavior, providing the foundation for accurate risk identification and preventing high-altitude fall accidents caused by physiological discomfort or abnormal conditions. The system deploys environmental sensors (such as wind speed, temperature and humidity, light intensity, and air pressure detectors), combined with video surveillance and edge computing devices, to continuously collect and analyze environmental parameters at the work site. This step can quickly identify environmental risks unsuitable for high-altitude work, such as high winds, high temperatures, and low visibility, providing real-time decision-making support for on-site managers and effectively avoiding environmentally induced safety accidents. By modeling the historical work data of different workers and incorporating parameters such as their professional skill level, work habits, and physiological response characteristics, a personalized risk assessment model is generated. This step transcends the limitations of traditional, standardized risk thresholds. It allows for flexible adjustment of warning thresholds based on individual characteristics, enabling customized safety risk assessments and improving the accuracy and practicality of the system's response. Through smart terminals or large-screen platforms, real-time visualization of worker location, physiological status, surrounding environmental data, and current risk level is displayed, assisting management with a comprehensive understanding of the safety situation in the work area. This processing step not only enhances the intuitiveness and timeliness of on-site management but also serves as a crucial basis for retrospective analysis of historical data and accident responsibility determination. The system incorporates artificial intelligence algorithms to analyze collected data, including worker behavior trajectories, posture patterns, and abnormal events, in real time. It predictively identifies high-risk behavioral trends, such as falls, slips, and imbalances, and sends preemptive warnings to management and worker terminals. This mechanism shifts from post-intervention to pre-emptive prevention and control, significantly enhancing the proactive nature of height-based work safety assurance. Once the system identifies a high-risk event or an abnormal fall, it immediately activates audible and visual alarms, video focus cameras, automatic capture systems, and emergency broadcast systems to immediately notify on-site management and record comprehensive data. This step ensures the efficiency and traceability of emergency response, significantly improves the efficiency of accident handling, and reduces the degree of injury. The visual monitoring and early warning system and method for preventing falls from high altitude provided by the present invention breaks the limitations of the traditional protection mechanism of "unified standards and post-event alarms". Through the five-in-one approach of individualized modeling, multi-source data fusion, AI intelligent prediction, visual display and linkage disposal, a full-process safety assurance system covering before, during and after operations is constructed. It not only significantly improves the accuracy and real-time nature of risk identification, but also realizes the "predictive control" of the risk of falling from heights, and has important application value and promotion prospects in improving the level of digital management of construction sites. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Other features, objects and advantages of the present invention will become more apparent upon reading the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0019] Figure 1 This is a schematic diagram of the steps of a visual monitoring and early warning method for preventing falls during high-altitude operations according to the present invention;

[0020] Figure 2 for Figure 1 Detailed step flow diagram of step S2;

[0021] Figure 3 for Figure 1 Detailed step flow chart of step S3 in FIG. DETAILED DESCRIPTION

[0022] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. It is obvious that the embodiments described are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts are within the scope of protection of the present invention.

[0023] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor and / or microcontroller approaches.

[0024] It should be understood that although the terms "first," "second," and the like may be used herein to describe various elements, these elements should not be limited by these terms. These terms are used solely to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the exemplary embodiments. The term "and / or" as used herein includes any and all combinations of one or more of the listed associated items.

[0025] To achieve this, please refer to Figures 1 to 3 The present invention provides a visual monitoring and early warning method for preventing falling during high-altitude operations, the method comprising the following steps:

[0026] Step S1: Collect the worker's physiological state data to generate the worker's fatigue characteristics and attention distraction index; collect behavioral posture data to analyze the regularity of the movement; collect work environment parameters and evaluate the work platform stability index;

[0027] In the embodiment of the present invention, wearable physiological sensors (such as heart rate belts, skin electrical sensors, brain wave sensing headbands, etc.) are used to continuously collect data from workers at high-altitude work sites, recording their heart rate changes, heart rate variability, skin conductivity, and brain wave signals (such as Wave and Wave ratio value), the above indicators are input into the fatigue calculation model, and the fatigue characteristics of workers are calculated by comparing the real-time data with the baseline values ​​of the individual in the resting state. For example, if the heart rate variability decreases by more than 25% and the change in skin conductivity is lower than the normal range, it is judged as a fatigue accumulation trend; at the same time, the specific band ratio in the EEG signal (such as The attention distraction index is calculated when the ratio exceeds a threshold. Behavioral posture data is collected using IMUs (inertial measurement units) installed on safety helmets and work clothes. Acceleration and angular velocity sequences are extracted, and the DTW (dynamic time warping) algorithm under a sliding time window is used to evaluate the similarity between the movements and the standard templates to determine the regularity of the work movements. For example, if the acceleration trajectory of the hand during a scissors operation deviates from the standard sample by more than 30%, it is considered an abnormal movement. Environmental parameters are collected using gyroscopes, anemometers, vibration monitors, and other devices on the aerial work platform to monitor the platform's tilt angle, wind level, and micro-seismic frequency in real time. The platform's stability index is calculated. An instability flag is triggered when the platform's lateral tilt angle exceeds 5 degrees or the wind speed continuously exceeds 10 meters per second. The structural stability risk of the work surface is assessed in combination with the vibration amplitude change trend, ultimately forming a structured basic data set of worker status and environmental parameters.

[0028] Step S2: Interactively correlate worker fatigue characteristics, distraction index, movement regularity, and work platform stability index to obtain worker-environment interaction status data; extract risk precursor features from the worker-environment interaction status data to form a multidimensional risk precursor feature set;

[0029] In this embodiment of the present invention, the fatigue features, distraction index, movement regularity index, and platform stability index obtained in S1 are interactively analyzed using a multimodal feature fusion method. For example, a structure based on a graph convolutional network (GCN) is used to construct nodes as feature items, and edge weights represent the correlation between them. Through model learning, a set of worker-environment interaction state data is obtained, such as the strong correlation between increased fatigue and increased frequency of movement instability. Subsequently, a time series model based on a long short-term memory network (LSTM) is used to extract risk precursor features from the worker-environment interaction state data sequence. Signal patterns such as the three minutes before the onset of a combination of "high fatigue + irregular movement + increased platform vibration" are identified. Such combinations are classified as high-risk precursor labels, and a multidimensional risk precursor feature set is constructed, including categories such as fatigue superposition, movement variation, and environmental instability. In practical applications, for example, after 120 minutes of continuous work, a worker experiences increased EEG theta waves, frequent slight body tilts, and wind speed increases from 8 m / s to 12 m / s. The model detects that the feature combination of this period is 92% consistent with the historical accident precursor pattern, and marks this period as a state with a high potential fall risk.

[0030] Step S3: Establish individual baseline parameters based on pre-acquired historical safe operation data, and record typical operation postures to construct behavioral pattern characteristics; calculate the operation difficulty coefficient based on the operation type complexity and environmental condition change rate in the historical safe operation data;

[0031] This embodiment of the present invention retrieves historical work data from workers over the past month and uses a clustering algorithm (such as DBSCAN) to categorize their physiological states and behavioral postures in safe conditions. Parameters such as resting heart rate, standard movement trajectory, and maximum sustained focus time are extracted to construct an individualized baseline model, which serves as a reference range for the worker's physiological and behavioral safety. Furthermore, typical movement and posture characteristics for each type of work are recorded, such as the "lift-docking-torque operation" action sequence in steel structure installation. Time series matching techniques are used to develop a standard work behavior template. The work difficulty coefficient is calculated based on the historical accident rate, platform complexity level, and frequency of external environmental changes corresponding to each work task. For example, the accident frequency for steel beam welding in a force 10 wind environment is 2.3 times the standard level. The system then calculates the work difficulty coefficient for the current task based on factors such as the average completion time and the number of changes in the work path. A higher value indicates a greater physical and mental strain on the worker.

[0032] Step S4: Generate a personalized risk threshold matrix based on individual baseline parameters, behavioral pattern characteristics, and task difficulty coefficient; perform real-time risk deviation calculation on the personalized risk threshold matrix and multi-dimensional risk precursor feature set, predict the risk development trajectory, and assess the potential fall risk probability to form a risk prediction report;

[0033] This embodiment of the present invention utilizes the individual baseline parameters, behavioral pattern characteristics, and task difficulty coefficients constructed in step S3 as input into a personalized risk threshold generation model. Based on fuzzy logic reasoning, this model maps each feature item to the risk level to form a risk tolerance boundary matrix. For example, the fatigue tolerance threshold is set between 0.35 and 0.45, and the motion regularity deviation tolerance threshold is set to within 15%. The multidimensional risk precursor features extracted in step S2 are compared with this threshold matrix in real time, and a sliding window is used to assess risk deviation. If the actual fatigue value exceeds the baseline by 25%, the posture deviation exceeds the standard template by 20%, or the platform vibration value continuously exceeds the threshold, a comprehensive deviation score is calculated for this state (e.g., reaching 0.78). The model then uses a dynamic Bayesian network to predict the risk development trajectory and simulate the risk level change trend within 30 minutes without intervention. Ultimately, a fall risk prediction report with time series analysis is generated, including information such as a potential risk value curve and the time window where the risk point is most likely to occur.

[0034] Step S5: Generate graded warnings and visual feedback based on the risk prediction report, form a high-altitude operation risk situation map, and send an alarm to the operator through the call system.

[0035] The system of the embodiment of the present invention performs hierarchical management of risk probabilities based on the risk prediction report generated in step S4. For example, a risk probability exceeding 0.7 is defined as a red warning, 0.5 to 0.7 is defined as an orange warning, and 0.3 to 0.5 is defined as a yellow warning. In response to different levels of risk, the system generates alarm information at different levels to form a high-altitude operation risk situation map that is visualized in time and space dimensions. For example, dynamic elements such as the risk level of the current operator, the corresponding fatigue status, and the unstable platform area are marked on the 3D platform model. The warning information is visually fed back through terminals such as smart helmets and AR glasses worn by the operators. When a red warning appears, a red frame flashes at the edge of the field of vision and a voice prompt "Please stop the current operation and retreat immediately" is given. At the same time, a call mechanism linked to the dispatching system is activated to synchronously send the alarm information to the command center and the on-site management terminal so that a dedicated person can be assigned to confirm the situation and implement intervention, forming a closed-loop warning response process.

[0036] Preferably, the step S1 of collecting the worker's physiological state data to generate fatigue characteristics and attention distraction index includes:

[0037] Collect workers' heart rate variability parameters, including RR interval standard deviation and low-frequency / high-frequency power ratio;

[0038] Collect skin electrical response data, including 0.01-1.0 Skin conductance levels in the S range and spontaneous fluctuation frequency of 0–12 beats per minute;

[0039] Measure the power density changes in the 20-200 Hz frequency band and the muscle fatigue index of 10-60% to obtain the electromyographic activity signal;

[0040] Extract heart rate irregularity index based on heart rate variability parameters and obtain cardiogenic fatigue factor;

[0041] The sympathetic nerve activity level was calculated based on the galvanic skin response data;

[0042] Fatigue analysis is performed on cardiogenic fatigue factors, sympathetic nerve activity levels, and myoelectric activity signals to generate a comprehensive fatigue score;

[0043] The fatigue accumulation rate is calculated based on the time trend of the comprehensive fatigue score to form the worker fatigue characteristics;

[0044] Analyze the mutation characteristics of electromyographic activity signals, identify small jitters and uncoordinated patterns, and calculate the rate of decrease in movement accuracy;

[0045] The stability of the autonomic nervous system was assessed based on the consistency of fluctuations between heart rate variability parameters and galvanic skin response data;

[0046] Cognitive load index was calculated based on the rate of decline in motor accuracy and the stability of the autonomic nervous system;

[0047] Based on the changing pattern of cognitive load indicators, an attention distraction index is generated.

[0048] In an embodiment of the present invention, workers wear an ECG acquisition device (such as a wearable ECG patch or chest strap) to record electrocardiogram (ECG) signals in real time. The sampling frequency is generally set to 250 Hz to ensure R-wave detection accuracy. The system extracts consecutive RR intervals (i.e., the time interval between two R waves) based on a QRS wave detection algorithm (such as the Pan-Tompkins algorithm). Within a 5-minute sliding time window of continuous recording, the standard deviation of the RR intervals (SDNN) is calculated as a time-domain indicator of heart rate variability. A fast Fourier transform (FFT) is simultaneously performed on the RR interval sequence to extract frequency-domain features. The low-frequency power (LF) in the 0.04-0.15 Hz range and the high-frequency power (HF) in the 0.15-0.4 Hz range are statistically analyzed, and the ratio (LF / HF) is calculated as an indicator of the balance between sympathetic and parasympathetic nerve activity, reflecting the load status of the autonomic nervous system. In an aerial work environment, if the SDNN drops below 30 ms and the LF / HF ratio rises above 2.5, it is initially indicated that the worker is showing a trend towards stress-related fatigue. Galvanic skin response (EDA) data is collected using fingertip-attached galvanic skin sensors with a signal sampling frequency of 10 Hz. These sensors record minute changes in skin conductance per unit time, with typical resting values ​​ranging from 0.01 to 1.0 microsiemens. The system analyzes spontaneous fluctuations (i.e., increases in skin conductance not accompanied by external stimulation) every 60 seconds, using zero-crossing or peak detection methods to identify spikes. Detecting more than seven spontaneous fluctuations within a minute indicates sympathetic nerve activity. Furthermore, the baseline skin conductance value is smoothed using a sliding average and compared to the baseline for subsequent estimation of sympathetic nerve activity. During high-altitude inspections, a persistent increase in the baseline skin conductance value and a frequency of spontaneous fluctuations exceeding 10 per minute indicate a worker experiencing mental stress or persistent alertness. Surface electromyography (sEMG) sensors are attached to key muscle groups in the worker's upper arm or shoulder (such as the deltoid or biceps) to collect muscle electrical activity signals in real time. A sampling frequency of 1000Hz is recommended to ensure frequency domain accuracy. The system uses short-time Fourier transforms to perform power density analysis in the 20-200Hz frequency range to assess muscle contraction strength and control stability, extracting root mean square (RMS) values ​​and average power. The system also calculates the decrease in median frequency and the increase in EMG amplitude to estimate a muscle fatigue index (e.g., a decrease in median frequency of more than 15% and an increase in RMS of more than 20% indicates moderate fatigue). The fatigue index is normalized to a scale between 0 and 1, creating a muscle fatigue index ranging from 10% to 60%. During high-altitude handling operations, if the muscle fatigue index approaches 50% after a task lasts for more than 45 minutes, a rotation of personnel is indicated.Based on the RR interval time series, the ratio of the mean square error to the mean (i.e., coefficient of variation) and the standard deviation of the differences between adjacent RR intervals are further calculated to measure rhythm irregularity. Furthermore, sudden interval shortening or prolongation events are identified. For example, if the frequency of fluctuations exceeding ±20% of the mean interval exceeds 10%, these are considered rhythmic disturbances. A cardiogenic fatigue factor scoring function with a 0-1 scale is developed based on these indicators. Higher values ​​indicate a greater degree of imbalance in the autonomic nervous system's control of the heart. During continuous high-altitude grinding operations, a cardiogenic fatigue factor score exceeding 0.7 continuously indicates abnormal circulatory system overload. After preprocessing, EDA signals are characterized by distinguishing between background conductance (basal value) and rapid response components (peak fluctuations). The frequency of spontaneous fluctuations per unit time (e.g., 8 per minute), the mean amplitude of the fluctuations (e.g., 0.2 microsiemens per rise), and the rate of rise are calculated. A linear weighted model is then developed to map this to a sympathetic nerve activity score, which typically ranges from 0 to 10. A score exceeding 6 indicates a state of high sympathetic activation. When performing wind turbine tower maintenance tasks in harsh weather, if the score reaches 8 or above, the risk of fatigue accumulation and distraction caused by sympathetic activity should be considered. The three indicators are standardized to a range of 0-1 and assigned weights (e.g., 40% for the cardiogenic fatigue factor, 30% for sympathetic nerve activity, and 30% for the electromyographic fatigue index). A weighted linear model is used to calculate a comprehensive fatigue score, which reflects the worker's overall physiological fatigue state during the current period. The score generally ranges from 0 to 1, with a value above 0.7 considered a high fatigue state. During welding operations in high-temperature environments, if the comprehensive score rises from 0.4 to 0.75 within 40 minutes, the system will automatically trigger a high-risk physiological fatigue alert. The system records time series data for comprehensive fatigue scores and calculates the score growth rate within a 5-minute window: the difference between the current score and the previous time point divided by the time interval. A growth rate greater than 0.05 / minute indicates rapid fatigue accumulation. Furthermore, the system compares this rate with the average rate over the previous hour to assess any abnormal trends. Ultimately, it generates a "fatigue accumulation curve" and a "fatigue warning threshold" to assess the worker's fatigue progression. In high-altitude glass replacement operations, if the cumulative rate remains at 0.06 / minute for 20 consecutive minutes, the system deems the work intensity to exceed the individual's recovery capacity. sEMG signals are analyzed in both the time and frequency domains to identify sudden, short-periodic high-frequency components (such as jitter), non-periodic oscillation patterns, and unstable frequency band shifts in the spectrum. By measuring the proportion of unstable signal segments within each minute of movement, areas of decreased muscle control accuracy are identified. For example, standard work movements should exhibit a stable 20-100 Hz power distribution; any asymmetric rise or excessively rapid switching is considered uncoordinated. The ratio of decreased movement accuracy, that is, the ratio of the duration of abnormal segments to the total movement duration, was calculated as an important indicator reflecting the decreased coordination of hands or limbs.During cable splicing operations requiring precise manipulation, if this ratio rises above 20%, the worker's status should be paused and reassessed. Cross-series analysis was performed on the SDNN, LF / HF ratio, and the electrodermal baseline and fluctuation frequency in the EDA. The degree of co-fluctuation between the two was calculated using the Pearson correlation coefficient or mutual information method. High consistency indicates good autonomic nervous system control; a synchronization difference greater than 0.5 indicates an imbalance in neural control. Synchrony scores were standardized to a stability score ranging from 0 to 1 for subsequent cognitive load modeling. During early morning high-altitude work, a sharp decrease in SDNN and no significant change in EDA fluctuations indicate sluggish neural feedback, resulting in a stability score below 0.4. The movement accuracy decline ratio and the autonomic nervous system stability index were entered as negative and positive variables, respectively, into a multifactor cognitive load model. A cognitive load score was calculated using a weighted regression function. The score ranges from 0 to 1. Higher scores indicate greater neural resources and perceptual control required to perform the task in the current environment, making it more likely to experience distraction or movement disorder. When performing high-altitude securing work on steel structures in strong winds, a cognitive load score above 0.75 indicates a high cognitive risk. The continuous time series of cognitive load scores is trend-fitted and, combined with the presence of sudden changes (e.g., a score increase from 0.4 to 0.8 within 10 minutes), a time series neural network model like LSTM is used to predict the short-term fluctuation range. If the predicted value fluctuates significantly from the actual value and the fluctuations are frequent, the system identifies this as a distraction phase and generates a distraction index between 0 and 1, with a higher index indicating less focused attention. During high-precision welding work at the work site for more than 60 minutes, if the index exceeds 0.65, the system identifies a distraction risk and recommends suspending the work.

[0049] Preferably, the collected behavior posture data and analysis of movement regularity in step S1 include:

[0050] Collect the worker's trunk tilt parameters, including forward tilt angle 0-90°, side tilt angle ±45°, and torsion angle ±60°;

[0051] Measure the displacement of ±25cm on the X axis and ±20cm on the Y axis relative to the standard standing position to obtain the center of gravity offset data

[0052] Collect key joint motion information, including the range of motion and six-axis motion trajectory of the shoulder, wrist, knee, and ankle joints, to form raw posture data;

[0053] Perform time series processing on the sampling frequency of the original attitude data and extract the attitude stability index within every 30 seconds;

[0054] Compare the trunk tilt parameters with the preset standard posture template of the work task and calculate the posture deviation;

[0055] A steady-state analysis of the center of gravity offset data was performed in a 3-second time window to extract acceleration mutation points of ±2 cm / s², identify body sway patterns with a frequency of 0.5-5 Hz, and obtain the body balance index.

[0056] Based on the posture stability index, the trajectory deviation of ±3 cm within the 95% confidence interval within the action cycle was extracted to obtain the joint trajectory repeatability;

[0057] The regularity of the movement is calculated based on the posture deviation, body balance index, and joint trajectory repeatability with a weight ratio of 0.3:0.4:0.3.

[0058] In an embodiment of the present invention, a fatigue assessment system for an automotive assembly line utilizes a wearable inertial measurement unit (IMU) positioned at the chest and waist of the worker to acquire three-dimensional torso posture angle data, specifically including forward tilt angles ranging from 0 to 90 degrees, roll angles ranging from ±45 degrees, and twist angles ranging from ±60 degrees. The data acquisition frequency is set to 50Hz, and three-axis rotation angles are extracted by converting quaternions into Euler angles. The system calculates the real-time torso angles of the three axes once per second and performs sliding window smoothing to improve data continuity and stability, ensuring that the angle measurements are not affected by short-term jitter. Ultimately, this data is used to determine the torso tilt state during work. In the same application environment, a two-dimensional displacement monitoring system is constructed using a ground reaction force measurement pad and IMU foot sensors. Using a standard standing position as the reference origin, the system records the center of gravity shift during the worker's actual movements. By real-time tracking of plantar pressure center and IMU inertial displacement data, the system measures a maximum displacement range of ±25 cm in the X-axis and ±20 cm in the Y-axis. Median filtering is used to eliminate the effects of vibration. The resulting trajectory and distribution of center of gravity movement are used to identify changes in balance during task execution. To obtain comprehensive information on joint motion, six-axis IMU sensors are attached to the worker's shoulder, wrist, knee, and ankle joints. Data includes the rotation angle (pitch, roll, and yaw) and acceleration vector for each joint in space. The data sampling frequency is set to 100 Hz, and data is continuously recorded over a standard work cycle. Using a skeleton modeling algorithm, a three-dimensional trajectory model is constructed for each joint, generating raw posture data. This provides the foundation for subsequent assessments of posture stability and movement repeatability. This raw posture data undergoes time-series resampling, with a sampling period of 30 seconds per analysis window. Within each window, the mean and standard deviation of posture change angles are first analyzed, and the root mean square (RMS) value of the posture curve's rate of change is calculated to characterize the degree of posture fluctuation within that period. By comparing this with a predefined stability threshold—for example, a standard deviation of less than 5 degrees for the forward tilt angle is considered stable—the system can then delineate periods of stability and instability throughout the work period, providing a basis for subsequent assessment of movement regularity. The system incorporates a set of standard posture templates corresponding to different work tasks. For example, standing tasks in assembly positions require a torso tilt of no more than 30 degrees, a left-right tilt of no more than 10 degrees, and a shoulder range of motion of no more than 60 degrees. The collected torso tilt angle data is then compared item by item with the template data. A frame-by-frame difference accumulation method is used to calculate the total deviation, which is then quantified as a posture deviation degree based on the ratio of the total deviation to the template's total limit. For example, a torso tilt deviation of 50% of the limit indicates a moderate deviation, reflecting the worker's adherence to the standard for performing the task.Center of gravity offset data is dynamically analyzed every three seconds as a time window. The rate of change of acceleration is calculated within each window, and points with a sudden change greater than ±2 cm / s² are screened. These sudden change points are further analyzed through frequency domain transformation to extract sway signals with a frequency range of 0.5 to 5 Hz. This allows identification of unstable motion patterns such as slight imbalance and body sway. A body balance index is constructed using sway amplitude and frequency as indicators, representing the worker's standing stability during different task phases. Based on the motion data analysis, the repeatability of joint trajectories within the motion cycle is extracted. First, the three-dimensional trajectory points of each joint in different cycles are aligned, and the Euclidean distance deviation between trajectory points is calculated. Within a 95% confidence level, the maximum deviation contained in the most repetitive trajectories is screened. For example, if the deviation of most trajectory points from the reference trajectory falls within ±3 cm, the joint motion is considered to have good trajectory repeatability, and the size of the deviation range is used as the basis for the repeatability score. The motion regularity index is calculated by combining the three indicators of posture deviation, body balance index, and joint trajectory repeatability, with weights of 0.3, 0.4, and 0.3. The weighted approach uses normalized scores followed by linear combination. Posture deviation is scored inversely (smaller deviations result in higher scores), while balance index and trajectory repeatability are scored positively based on stability. The final score represents a worker's ability to maintain regular movements throughout the task cycle, serving as a comprehensive reflection of fatigue status and posture control ability, and can be used for further behavioral prediction and risk intervention.

[0059] Preferably, the collection of working environment parameters and evaluation of the working platform stability index in step S1:

[0060] Collect vibration frequency data of the work platform, measuring the amplitude distribution and main frequency components in the 1-100 Hz frequency range, and recording the peak acceleration of 0.1-5.0g;

[0061] Collect environmental wind parameters, including wind speed of 0-25m / s and wind direction changes of 0-360°;

[0062] Measure the surface friction coefficient of the work platform, ranging from 0.15-0.95 dynamic friction value and 0.2-1.0 static friction value;

[0063] Monitor the load distribution on the platform, record local pressure data from 0-2000kg / m², and measure the tilt of the working platform;

[0064] Extract the energy ratio of human-sensitive frequencies within the range of 5-25Hz from the vibration frequency data of the work platform and calculate the resonance risk index;

[0065] Calculate the estimated lateral pressure under wind force based on the ambient wind parameters and generate a risk weighting factor based on the height of the working platform;

[0066] Calculate the critical slip risk threshold based on the surface friction coefficient and the tilt state of the work platform;

[0067] Generate current friction state based on surface friction coefficient and local pressure data;

[0068] Compare the critical slip risk threshold with the current friction state to derive the slip risk probability;

[0069] Analyze local pressure data to calculate the center deviation index and edge overload risk, and obtain load unevenness;

[0070] The resonance risk index, risk weighting coefficient, sliding risk probability and load unevenness are used to calculate the working platform stability index based on a weight distribution of 0.25:0.3:0.25:0.2.

[0071] In this embodiment of the present invention, a three-axis accelerometer and spectrum analysis module are mounted at the center and four corners of a construction lifting platform to continuously monitor its stability. These sensors continuously collect vibration signals within the 1 to 100 Hz range, extract the acceleration time-domain waveform, and perform a Fourier transform to obtain the spectral distribution. The primary frequency component of each channel within this frequency band is extracted, and the dominant frequency corresponding to the maximum amplitude is identified. The maximum peak acceleration value during the monitoring period is recorded, with the range controlled between 0.1g and 5.0g. This data is used to identify whether the platform has vibration concentration issues caused by specific mechanical structures or external forces. An anemometer is installed on the top of the platform, facing the wind, to collect real-time information on ambient wind speed and direction changes. The system sampling frequency is set to 1Hz, recording wind speed values ​​(0 to 25 meters per second) and wind direction angles (0 to 360 degrees) every second. A 10-second sliding average is used to mitigate the impact of sudden wind speed changes on data stability. Wind speed data is projected onto the platform's orientation to determine the wind impact on the platform's frontal side for subsequent risk weight analysis. A friction coefficient meter was used to measure the dynamic and static coefficients of friction on various work platform surfaces. Dynamic friction was measured by pulling a standard slider at a constant speed, recording the resistance to continuous motion. The values ​​were controlled within a range of 0.15 to 0.95. Static friction was measured by gradually increasing the pulling force to the maximum force at the slider's initial actuation, with values ​​ranging from 0.2 to 1.0. Experiments were conducted on both dry and wet surfaces, with repeated sampling under varying contamination conditions (such as oil and water stains) to determine the average surface friction coefficient under the current working conditions. Platform load monitoring utilizes an array of strain gauge pressure sensors positioned beneath the platform surface, forming a distributed pressure detection system with a sampling frequency of 10 Hz. The sensors have an accuracy of 1 kg / m² and support a maximum recording range of 2000 kg / m². During each sampling cycle, the system calculates the real-time pressure per unit area and combines it with the three-axis attitude angles (pitch and roll) measured by the gyroscope to determine the overall platform tilt. Tilt angle accuracy is controlled within ±0.5 degrees to ensure data accuracy. For the collected vibration frequency signals, an analysis window of 5 to 25 Hz is selected from the 1 to 100 Hz spectrum, representing the typical frequency range sensitive to human perception. The total energy contribution within this frequency band is calculated using the power spectral density integration method, and then combined with the overall vibration energy ratio to form a resonance risk index. If this energy contribution exceeds 40%, the platform is preliminarily deemed to have a high risk of human resonance interference, requiring further damping or structural reinforcement measures. Based on wind parameters, combined with the platform height (e.g., 15 meters) and windward area (e.g., 5 square meters), a simplified aerodynamic model is used to calculate the lateral force, which increases with the square of the wind speed. The angle between the wind direction and the platform's long axis is used as a projection factor, and an altitude correction factor is introduced (increasing the risk weight as the platform rises higher). This ultimately results in a risk-weighted coefficient between 0 and 1, reflecting the potential impact of wind on the platform's lateral stability at the current height and direction.The system calculates the initial conditions for platform slip based on the measured surface friction coefficient and the current platform tilt angle. The static friction coefficient multiplied by the projected gravity component is compared with the tangent of the tilt angle. The greater the tilt angle and the smaller the friction coefficient, the more likely it is to trigger slip. The system sets a preset slip risk threshold as the platform's critical tilt angle at a specific friction coefficient. The corresponding critical state is then identified as a slip boundary. Based on this threshold, the system combines local pressure data to calculate the product of the unit pressure and the friction coefficient within the actual contact area to determine the total available friction resistance. If the pressure concentration area coincides with a low-friction area, the friction state deteriorates, and the maximum resistance under this state is considered the platform's current friction state. This value is used as the actual anti-slip capability indicator for subsequent risk analysis. The critical slip risk threshold is compared with the current friction state. If the current friction resistance is below the threshold, the platform is at risk of slip. A probability distribution model is constructed through repeated sampling. The probability density of occurrences below the threshold within any time window is calculated to determine the platform's slip risk probability under the current operating conditions, which is used as part of the platform structural safety assessment. Statistical analysis of the distributed pressure data is performed to calculate the offset distance of the platform's pressure center of gravity from its geometric center, forming a center offset index. In addition, statistics are collected for overload events that occur in the edge area (within 1 meter of the platform boundary) that are more than 1.5 times the average value. The frequency and peak value are recorded to assess the edge overload risk. The two are combined to form a load unevenness index, which is used to identify whether the platform structure is balanced. The resonance risk index, wind risk weighting coefficient, sliding risk probability, and load unevenness are assigned weights of 0.25, 0.3, 0.25, and 0.2, respectively. The normalized weighted sum is performed to obtain the comprehensive working platform stability index. The stability index fluctuates between 0 and 1. The lower the value, the more stable the platform is under the current working conditions. The higher the value, the more necessary the structure reinforcement, load limit, or height limit treatment is. It is an important decision-making support data in the on-site safety management system.

[0072] Preferably, step S2 includes the following steps:

[0073] Step S21: Time-synchronize and pair the worker fatigue characteristics with the attention distraction index to generate a cognitive state matrix;

[0074] In this embodiment of the present invention, wearable sensors are used to collect workers' heart rate variability, galvanic skin response, and EEG frequency band characteristics, and fatigue characteristic values ​​are calculated every five minutes. For example, a heart rate variability below 20ms and a significant increase in alpha wave frequency are considered high fatigue. Simultaneously, a head posture tracker and an eye tracker are used to synchronously record the frequency of gaze drift, gaze dwell time, and head deflection amplitude of workers to calculate an attention distraction index. After the two types of data are collected, they are aligned using a unified timestamp and synchronized within a one-minute time window to ensure a one-to-one correspondence between fatigue and attention distraction data within each time period. A two-dimensional matrix is ​​then constructed, with the horizontal axis representing the time period sequence and the vertical axis representing the two indicators. This creates a cognitive state matrix that reflects the time series change trends and is used to assess changes in workers' cognitive load status during different work stages.

[0075] Step S22: mapping the action regularity into a behavior stability index in the range of 0-1;

[0076] The embodiment of the present invention performs numerical standardization mapping processing on the motion regularity index extracted in the early stage. The original numerical value of the motion regularity comes from the score of the three indicators of posture deviation, body balance index and joint trajectory repeatability synthesized by weight, which is usually between 0 and 100. In order to facilitate unified comparison with other indicators, the minimum-maximum normalization method is used to map the score to between 0 and 1, where 1 represents a highly stable motion pattern with small fluctuations, and 0 represents a highly unstable motion with obvious abnormalities. This behavioral stability index will be used as a key factor in reflecting individual operation consistency and state fluctuations in subsequent interaction modeling.

[0077] Step S23: analyzing the time series changes of the working platform stability index and extracting the platform stability fluctuation characteristics;

[0078] The embodiment of the present invention performs time series analysis on the stability index sequence of the working platform obtained above. The acquisition period is set to one stability value every 30 seconds to form a continuous change curve. Through the sliding window analysis method, the window length is set to 10 minutes, and the variance, range and fluctuation frequency of the stability index are calculated in each window to extract the stability fluctuation characteristics of the platform in this period. For example, when the platform has frequent increases in the vibration resonance index and violent fluctuations in the tilt angle within 10 minutes, the high fluctuation state label of this time period can be extracted. The final result uses the time label as the horizontal axis and the fluctuation intensity feature as the vertical axis to form a time series feature set that describes the change in the platform state.

[0079] Step S24: constructing a three-dimensional correlation map based on the cognitive state matrix, the behavioral stability index, and the platform stability fluctuation characteristics to form worker-environment interaction state data;

[0080] This embodiment of the present invention aligns three types of data—the cognitive state matrix, the behavioral stability index, and the platform stability fluctuation characteristics—by time and uniformly maps them to three-dimensional spatial coordinates to construct worker-environment interaction state data. Specifically, within each time unit, fatigue and distraction from the cognitive state matrix are combined to form cognitive load coordinates. The behavioral stability index is used as the Y-axis dimension, and the platform fluctuation characteristics are used as the Z-axis dimension. These three are combined to form a set of three-dimensional coordinate points representing the worker-environment interaction state at that moment. This interaction state data not only describes changes in the coupling between individuals and the environment but can also be used for cluster analysis to identify potential high-risk working conditions.

[0081] Step S25: extract risk precursor features from the worker-environment interaction status data to form a multi-dimensional risk precursor feature set.

[0082] The embodiment of the present invention performs time series analysis and cluster identification on continuously recorded worker-environment interaction state data to extract risk precursor features. The specific method is to use a density-based clustering algorithm to identify abnormal cluster areas in three-dimensional space, such as segments with frequent combinations of high cognitive load, low movement stability, and violent platform fluctuations. At the same time, a mutation detection algorithm is introduced to capture the transition process from a stable state to an unstable state, and to extract the difference in average values ​​before and after the change, the fluctuation rate, the state transition frequency, etc. as multidimensional features. Ultimately, a precursor feature set containing typical risk patterns such as high fatigue-unstable movement-violent environmental disturbance is formed, which is used to identify risk situations in which workers may become unbalanced, operate incorrectly, or experience sudden changes in working conditions in advance, and to achieve intelligent early warning and intervention control.

[0083] Preferably, step S25 includes the following steps:

[0084] Conduct time series analysis on cognitive components in worker-environment interaction data and extract cognitive load fluctuation curves;

[0085] Use cognitive load fluctuation curve to calculate the distribution of attention critical points and identify attention risk windows;

[0086] Perform pattern recognition on behavioral components in worker-environment interaction state data to extract postural instability intervals;

[0087] Analyze the temporal consistency between postural instability intervals and platform stability fluctuation characteristics to generate an environmentally induced behavioral risk index;

[0088] Identify high-risk state combinations based on attention risk windows and worker-environment interaction state data; associate high-risk state combinations with workspace coordinates to form a risk space distribution map;

[0089] Predict cognitive decline trends based on the cognitive state matrix and generate early warning indicators for attention decline;

[0090] Generate an environmental hazard zone map based on platform stability fluctuation characteristics and working environment parameters;

[0091] A dynamic risk association model is established through the spatiotemporal correlation between attention loss warning indicators and environmental danger zone maps;

[0092] The risk weight of the risk spatial distribution map is adjusted based on the risk level classification results of the dynamic risk association model to form a multidimensional risk precursor feature set.

[0093] When performing time-series analysis on cognitive components in worker-environment interaction data, the present invention first extracts time series data for fatigue and distraction indices from the cognitive state matrix, constructing a long-term cognitive state change curve with one-minute time units. The data is then smoothed using a moving average method to eliminate the effects of short-term fluctuations. The mean and standard deviation within each time window are calculated to analyze the rising and falling trends of cognitive load. Simultaneously, a change point detection algorithm is used to identify sudden changes in cognitive state. For example, using the CUSUM or Pelt algorithm, the time points when a focused state transitions to fatigue or distraction are identified. This allows the extraction of a complete cognitive load fluctuation curve, reflecting the cyclical changes in cognitive load across different work stages. This curve is suitable for analyzing the status of rail construction workers working continuously for 8 hours. When using the cognitive load fluctuation curve to calculate the critical attention point distribution, an attention change threshold is first set. For example, a combination of a fatigue index consistently exceeding a set value and a distraction index rapidly increasing beyond 0.6 is considered an attention instability boundary. Thresholds are applied to the fluctuation curve to identify the minimum and rapid turning points on the curve. The frequency and distribution of the minimum points over the entire time period are statistically analyzed to generate a temporal distribution map of attention critical points. Furthermore, segments in which three or more critical points occur consecutively and last for more than 10 minutes are labeled as attention risk windows, identifying periods of excessive cognitive load during work. For example, these risk windows are prone to occur during peak track laying periods or periods of intense environmental vibration. Pattern recognition of the behavioral components in the worker-environment interaction data primarily focuses on movement stability indicators and posture features. A long short-term memory (LSTM) network is trained on behavioral sequences to distinguish between normal stable postures and abnormal patterns such as swaying and frequent tilting. Using the standard deviation and trajectory offset of the shoulder, knee, and ankle joint trajectories as input features, a clustering algorithm is used to isolate unstable postural intervals. Furthermore, high-frequency resonance segments in the platform vibration data are combined to annotate abnormal states in the behavioral sequences. Segments lasting more than 15 seconds and with trajectory deviations exceeding ±5 cm are identified as typical unstable postural intervals. This is particularly applicable to identification scenarios involving cantilevered operations or high-altitude edge work. A time alignment analysis method was used to analyze the temporal consistency between postural instability intervals and platform stability fluctuation characteristics. First, the postural instability intervals were synchronized with the platform stability index fluctuation curve, and the platform vibration peak and tilt angle mutation point were extracted using a sliding window method. The overlapping time ratio and relative time offset of the two types of events were then calculated. If the two occurred synchronously multiple times within a minute, postural instability was identified as caused by platform fluctuations. Finally, an environmentally induced behavioral risk index was generated, which incorporates the degree to which platform factors induce worker movement instability and a temporal consistency score. A typical application of this index is when wind-induced vibrations on a bridge maintenance platform significantly increase worker movement fluctuations.To identify high-risk state combinations based on the attention risk window and worker-environment interaction data, time segments within the attention risk window are first cross-referenced with periods of unstable behavior and periods of high platform fluctuation. This approach identifies complex states that simultaneously exhibit high cognitive load, high behavioral fluctuations, and strong environmental disturbances. For example, a combination of three abnormal indicators occurring more than three times within an hour and lasting for more than five minutes is labeled a high-risk state. These time points are then mapped to the worker's spatial coordinates. Their relative position on the work platform is calibrated using real-time positioning data from wearable devices (such as UWB or RTK signals). This is then combined with a building BIM model or spatial grid model to create a spatial distribution map of the risk, highlighting areas prone to high-risk behaviors for dynamic risk management on the construction site. To predict cognitive decay trends based on the cognitive state matrix, a trend prediction model is constructed using time series of fatigue and attention distraction over the past several hours. Using time series regression methods, such as ARIMA or LSTM prediction algorithms, the upward slope of fatigue and the downward trend of attention index over the next 30 minutes are assessed. If both exceed a set rate of change threshold (e.g., a rate of attention decline exceeding 0.05 per minute), an attention decline warning indicator is generated, along with an estimated time of occurrence and duration. This indicator is suitable for proactive early warning intervention for workers engaged in long-term, high-intensity operations. To generate an environmental hazard zone map based on platform stability fluctuations and working environment parameters, multi-parameter data, such as the platform's vibration frequency resonance distribution, peak acceleration changes, wind speed and direction, and platform tilt, are spatially mapped. Through environmental physics model analysis, including platform resonance response simulation and wind load simulation, stability scores for different areas under varying environmental conditions are calculated. Low-stability areas, prone-to-slip zones, and areas subject to extreme loads under wind load direction are annotated on the platform's BIM model to generate an environmental hazard zone map for pre-construction risk planning and intelligent risk avoidance assistance during operations. Data fusion technology was used to construct a unified multidimensional indicator space, integrating attention loss warning indicators, environmental hazard zone maps, environmentally induced behavioral risk indicators, and risk spatial distribution maps to form a multidimensional risk precursor feature set. Each dimension represents an independent risk source, and tensor processing is used to uniformly represent time, space, risk intensity, and event type. Association rule mining and clustering methods were used to extract frequent co-occurrence patterns. For example, the risk combination of "front left of the platform + wind speed greater than 15m / s + attention index less than 0.4 + behavioral trajectory deviation greater than 5cm" was frequently identified as a precursor. The result is a structured, traceable, and predictable risk precursor feature set, supporting the deployment of an on-site multi-source data fusion warning system and real-time feedback mechanism.

[0094] Preferably, step S3 includes the following steps:

[0095] Step S31: extracting the average heart rate variability parameters, galvanic skin response baseline and normal range of myoelectric activity of workers from the pre-acquired historical safety operation data, and constructing a personal physiological parameter profile;

[0096] This embodiment of the present invention extracts archived records from workers' wearable physiological monitoring devices from historical safety work data, including parameters such as heart rate (HR), skin conductance (EDA), and surface electromyography (sEMG). By statistically analyzing this long-term, stable data, the mean and standard deviation of heart rate variability (HRV) are calculated. Indicators such as the standard deviation interval (SDNN) and the root mean square difference between adjacent intervals (RMSSD) are extracted as reference parameters for HRV. For electrodermal response, a baseline value of at least 15 minutes in a resting state is selected as a reference for the individual stress threshold. For electromyographic activity, the average EMG amplitude range of major upper and lower limb muscle groups (such as the biceps brachii and quadriceps femoris) during repetitions of standard work movements is analyzed to determine their normal EMG operating range. These normalized parameters constitute an individual's physiological parameter profile, facilitating subsequent personalized risk assessment and abnormal condition identification. For example, in the records of workers working on a tower at a power construction site, the mean heart rate variability of a worker was 68 milliseconds, the electrodermal baseline was 0.25 microsiemens, and the normal range of myoelectric activity was 20 to 40 microvolts.

[0097] Step S32: Marking posture features in the historical safe operation data that have a frequency exceeding a preset threshold as common posture features to obtain a human posture baseline value, where the common posture features include the habitual torso tilt angle, center of gravity distribution pattern, and joint movement preference;

[0098] The embodiment of the present invention collects statistics on the joint angles, center of gravity positions and limb distribution in each work task based on historical work motion capture data. By segmenting the time series and classifying and clustering each segment of posture data, the trunk inclination angle, joint deflection pattern and center of gravity movement trajectory with a recurrence frequency exceeding a preset threshold (such as 30 times / hour) are identified. For example, if a worker's forward lean angle in the ground paving task is always maintained between 20 and 25 degrees, and the frequency of this posture in the total work time reaches 35%, then this posture is marked as the "common posture feature" of the worker. In addition, the stable center of gravity distribution pattern, such as the preference for leaning to the left or moving backward, is identified based on the gait and center of gravity change map recorded by the plantar pressure sensor. The final posture reference value contains a specific angle range, joint rotation trend and distribution density information, which is suitable for use in scenarios with high action repeatability such as structural steel assembly and high-altitude bolt fixing.

[0099] Step S33: analyzing the adaptability of workers under different environmental conditions and extracting environmental tolerance parameters;

[0100] This embodiment of the present invention filters historical data to identify workers' performance records under different working conditions, including external parameters such as temperature, humidity, wind speed, noise level, and vibration intensity, as well as physiological responses (HRV, EDA) and behavioral performance (such as movement smoothness and task completion efficiency) within corresponding time periods. By comparing and analyzing the stability of work status under varying environmental parameters, the environmental tolerance range and upper limit are extracted. For example, if a worker's galvanic skin response fluctuates frequently and their movement duration increases by more than 20% when humidity exceeds 80%, this indicates poor adaptability to hot and humid environments. Statistical correlation analysis combined with principal component analysis (PCA) is used to extract key tolerance factors and define individualized environmental tolerance thresholds. For example, during tunnel construction, data collected from a worker shows a wind speed adaptation threshold of 4.5 m / s and a vibration tolerance of no more than 0.8 m / s², constituting their environmental tolerance parameter set.

[0101] Step S34: integrating the personal physiological parameter profile, common posture characteristics and environmental tolerance parameters into personal baseline parameters;

[0102] In this embodiment of the present invention, the physiological parameter profiles, common posture characteristics, and environmental tolerance parameters obtained in the aforementioned steps are integrated through a feature fusion algorithm to form individual baseline parameters. Specifically, the three types of data are first dimensionally unified and normalized. A feature vector group is then constructed, and a feature splicing method is used to form a unified feature description vector. An individual baseline model profile is then established for subsequent dynamic monitoring and risk deviation identification. For example, a worker's baseline parameters include: an HRV standard deviation of 70 milliseconds, a common forward tilt angle of 23 degrees, an upper temperature limit of 35°C, and a noise tolerance threshold of 85 decibels. This individual baseline parameter model can serve as the basis for adaptively adjusting the risk warning model for that worker, and is particularly suitable for the status management of workers who perform repetitive work for a long time, such as welding and testing positions.

[0103] Step S35: Constructing behavioral pattern features based on the historical safety operation data recording typical operation postures;

[0104] The embodiment of the present invention constructs a behavioral pattern dictionary by performing structured encoding on the sequences of work postures under different task types in historical safety work data, combined with the annotated data. Temporal pattern mining techniques (such as the PrefixSpan algorithm) are used to identify frequently recurring posture sequence segments to form a typical work posture model. The Hidden Markov Model (HMM) is further used to probabilistically model the behavioral pattern, describing the state transition path from the start action to the end action. For example, the three-stage "grasp-stretch-fix" action in the high-altitude installation task is analyzed. The timing logic and duration range of each action are determined by the specific elbow joint angle changes and upper limb position, which serve as the typical behavioral pattern characteristics of this type of task. This can be used to automatically identify "non-standard operations" in construction training simulation and behavioral deviation monitoring.

[0105] Step S36: Calculate the operation difficulty coefficient based on the operation type complexity and environmental condition change rate in the historical safety operation data.

[0106] This embodiment of the present invention statistically analyzes the structural complexity, spatial constraint level, cross-operation frequency, and environmental condition volatility of different operation types in historical safety operation data. For each operation type, a complexity factor (such as assembly difficulty and control accuracy requirements) is first defined. The volatility of environmental parameters (such as the number of wind speed changes per hour and the amplitude of temperature changes) is then introduced, and a weighted scoring method is used to form an operation difficulty coefficient evaluation model. For example, for an outdoor steel structure assembly operation, because the operation must be completed in strong winds and at high altitudes, the wind speed changes frequently up to 5 times per hour, and the spatial operation tolerance is less than 10 cm, the final operation difficulty coefficient is assessed to be 0.87 (with a full score of 1). This coefficient is used to adjust the model sensitivity and warning threshold in subsequent dynamic risk prediction to ensure that the system responds more sensitively and promptly in difficult operation scenarios.

[0107] It is particularly important that step S35 includes the following steps:

[0108] Step S351: extracting posture time series records from historical safety operation data and performing time window segmentation to obtain operation action segments;

[0109] The embodiment of the present invention retrieves time series data containing human posture information from a historical safe operation database. This data is usually recorded in real time by a motion capture system (such as an IMU inertial sensor, a depth camera, or a skeleton recognition algorithm). The recorded content includes the position coordinates of key joints, angle changes, center of gravity trajectory, etc. Taking 30 frames per second as an example, the continuous posture sequence is segmented into fixed time windows, such as a window of 10 seconds, and the overlapping step length is set to 2 seconds to ensure the recognition of movement continuity. The posture data within each time window constitutes an independent operation action segment, which is used for subsequent action pattern clustering. Taking the wiring operation of an electrical box as an example, multiple repeated segments of "lifting the arm - plugging the wire - crimping" are segmented from the continuous monitoring data. These action segments serve as the basic units for subsequent analysis.

[0110] Step S352: performing cluster analysis on the operation action segments to identify recurring action patterns in the same type of operations and obtain a typical action set;

[0111] In this embodiment of the present invention, the work action segments extracted in step S351 are classified using an unsupervised clustering algorithm based on a distance metric. For example, dynamic time warping (DTW) combined with the K-Medoids or DBSCAN clustering algorithm can be used to group similar actions based on the dynamic shape similarity metric between the action segments. The clustering process first extracts key frame features of the action segments, such as the shoulder, elbow, and wrist angle change sequence and the center of gravity movement trajectory, and converts them into feature vectors of uniform length. After clustering, representative central actions are extracted from each category to form a typical action set for that work type. For example, in the template installation task, cluster analysis identified three action categories: "lifting the board - calibrating - fixing." The "lifting the board" category contains 10 similar segments, making it a typical action.

[0112] Step S353: Calculate the occurrence frequency and duration of each action in the typical action set to obtain action usage statistics;

[0113] The embodiment of the present invention counts the frequency of occurrence of each action category in the typical action set in the historical operation data and the duration of each action. The frequency of occurrence statistics is based on the number of clusters of the category to which the action belongs. For example, a certain type of "turn around and pick up something" action appears more than 120 times in 1,000 segments, and its frequency is 12%. The duration statistics represent the duration characteristics of the action by calculating the mean and standard deviation of the time span in each type of action, which facilitates the identification of efficient and lengthy operations. For example, in high-altitude maintenance operations, a certain type of "stable probe" action has an average duration of 4.3 seconds and a standard deviation of 0.9 seconds, which is a time-stable action.

[0114] Step S354: extracting high-frequency action sequences based on action usage statistics to obtain a workflow model;

[0115] Based on the action frequency data obtained in step S353, this embodiment of the present invention filters actions whose frequency exceeds a set threshold (e.g., 10%) and constructs a temporal sequence model based on their order within the workflow. Sliding window statistics and sequential pattern mining methods (such as SPADE or GSP algorithms) are used to identify frequently occurring action sequences. For example, the sequence "tool removal—positioning—fixing" occurs 65% of the time in a certain type of assembly task and can be considered the standard workflow pattern for that task. The resulting workflow pattern can be used to provide process standardization recommendations and design early warning systems for process deviations.

[0116] Step S355: Calculating stability indexes for the actions in the typical action set to obtain action stability scores;

[0117] This embodiment of the present invention calculates a motion stability index for each action category in a typical action set. The specific calculation method involves evaluating the degree of fluctuation in each action in terms of spatial trajectory, action duration, and action amplitude. The smaller the fluctuation, the higher the stability. A comprehensive score can be calculated using the coefficient of variation, posture standard deviation, or trajectory consistency indicators (such as a similarity score). For example, in a welding task, the stability score for the "holding a gun steadily and moving" action reached 0.92 (out of a maximum score of 1), indicating that the action is highly repeatable and has minimal posture deviation, making it suitable as a reference for skill level assessment.

[0118] Step S356: analyzing the transition features between typical actions in the typical action set to obtain an action transition pattern;

[0119] The embodiment of the present invention analyzes the conversion relationship between each action in a typical action set and identifies the probability and transition rules of adjacent occurrence. An action state transition matrix can be constructed to count the probability of action A followed by action B, while recording the conversion time and occurrence conditions, such as whether the tool has changed or whether the working direction has changed. The hidden Markov model (HMM) can be further used to model the action conversion path and extract high-frequency paths. For example, a common conversion pattern in the bracket assembly process is "take the screw - lift the hand - position - tighten", and the frequency of this sequence is more than 80%, and the conversion time is kept within 1.5 seconds, forming an efficient operation path.

[0120] Step S357: Constructing behavioral pattern features based on typical action sets, workflow patterns, action stability scores, and action transition patterns.

[0121] The embodiment of the present invention integrates typical action sets, workflow patterns, action stability scores and action conversion patterns, and constructs a behavioral pattern feature set through a structured behavioral modeling method. The specific method is to use a graph structure or a vector group to describe the action nodes and their conversion edges. The nodes are accompanied by stability weights and frequency information, and the edges represent conversion probabilities and time consumption, forming a visual or embedded behavioral pattern map. This feature can be used for behavior prediction, process optimization and anomaly identification. For example, at a heavy equipment assembly site, the behavior map of a specific operator is compared with the standard map to quickly identify deviations in "insufficient action stability" or "abnormal conversion path", thereby improving the intelligence level of safety management.

[0122] Preferably, step S36 includes the following steps:

[0123] Step S361: extracting the operation type classification information from the historical safety operation data, and calibrating the difficulty level according to the preset difficulty standard data to obtain the basic difficulty coefficient of the operation type;

[0124] An embodiment of the present invention extracts the job type label corresponding to each job record from the historical safety operation data. The label can be classified and identified according to the job task classification standard, such as "high-altitude maintenance", "welding operation", "confined space operation", etc. Subsequently, a pre-established job difficulty standard library is called, which includes the score values ​​of difficulty dimensions such as operation complexity, physical requirements, skill requirements and psychological pressure corresponding to each job type. By comparing the correspondence between the job type in the historical data and the difficulty standard library, a basic difficulty score is assigned to each type of operation. For example, according to the enterprise safety standards and specifications, "high-altitude maintenance" is defined as high operation complexity and strong psychological pressure, with a score of 0.85; "ground handling" is simple to operate, with a score of 0.25. Finally, the basic difficulty coefficient corresponding to each type of operation is formed, which provides an initial reference for subsequent difficulty estimation.

[0125] Step S362: extracting environmental parameter records from historical safety operation data, calculating the rate of change per unit time, and obtaining environmental condition change rate data;

[0126] The embodiment of the present invention extracts dynamic parameter data related to the working environment from historical safety operation records, such as temperature, humidity, wind speed, illumination, noise intensity, and the concentration of harmful gases in the air. These data are recorded in real time by on-site environmental sensors or environmental monitoring systems. These parameters are sorted according to the time axis, and the rate of change per unit time is calculated, that is, the increase or decrease of each environmental parameter within 10 minutes or 30 minutes is divided by the time interval to obtain the rate of change of environmental conditions. For example, if the wind speed rises from 2.5 meters per second to 6.0 meters per second in 30 minutes, the rate of change is 3.5 divided by 30 minutes, which is about 0.117 meters per second per minute. Finally, a multidimensional rate of change data set is output to measure environmental volatility.

[0127] Step S363: Analyze the correlation between the environmental condition change rate data and the operation safety to obtain the environmental impact factor;

[0128] The embodiment of the present invention performs statistical correlation analysis by comparing historical environmental change rate data with operational safety event data (such as errors, alarms, or abnormal behavior markers) in the corresponding time period. The specific method is to calculate the correlation coefficient between the change rate of each environmental parameter and the frequency of operational anomalies, such as the Pearson correlation coefficient or the information gain index. If it is found that the operator's error rate increases significantly when the noise fluctuation is large, it means that the noise change has a greater impact on operational safety, and it is accordingly given a higher environmental impact weight. Based on the analysis results, an impact factor value is assigned to each environmental parameter to form a multi-dimensional environmental impact vector. Taking the "tunnel welding" operation as an example, the impact factors of noise and gas concentration changes are 0.4 and 0.35, respectively, while the impact of temperature and humidity is relatively low, only 0.1 and 0.05.

[0129] Step S364: Calculating a height risk coefficient based on the working height information in the historical safe operation data, thereby generating a height challenge index;

[0130] Embodiments of the present invention extract work altitude data from work records. This data can be provided by a positioning system (such as UWB positioning, 3D coordinate mapping) or a seatbelt sensor, reflecting the worker's actual height above the ground during operation. This data is combined with a preset altitude risk model, which is constructed based on the relationship between altitude levels and accident probability. Generally, higher altitudes are assumed to pose greater challenges to psychological and physical stability, resulting in a higher risk index. For example, altitudes below 1 meter are considered low risk (with a coefficient of 0.1), 1 to 3 meters are considered medium risk (with a coefficient of 0.4), and altitudes above 3 meters are considered high risk (with a coefficient of 0.8). For each type of work, an altitude risk coefficient is calculated based on the average operating altitude, and a psychological tolerance compensation factor is introduced (for example, a weight of 0.1 for first-time high-altitude work). This ultimately creates a composite altitude challenge index for the work. For example, in wind turbine blade maintenance work, where the average operating altitude is 60 meters, the challenge index can reach 0.95.

[0131] Step S365: constructing a difficulty benchmark value based on the basic difficulty coefficient of the task type, and generating an environmental impact factor in combination with the environmental impact factor;

[0132] The embodiment of the present invention uses the basic difficulty coefficient of the operation type obtained in step S361 as the core variable for constructing the difficulty benchmark; then, combined with the environmental impact factors extracted in step S363, the basic difficulty is weightedly adjusted. The specific operation can adopt a linear weight combination or a weighted scoring model, such as adding the weighted sum of the environmental impact factors to the basic difficulty coefficient, or setting the weight ratio according to empirical rules (such as the basic difficulty accounts for 60% and the environmental factor accounts for 40%). For example, the basic difficulty of a "pipeline corridor maintenance" operation is 0.6, and the total weight of the environmental factor is 0.25. After weighting, the adjusted difficulty value is 0.6 times 0.6 plus 0.25 times 0.4, which is approximately 0.53. This value reflects the result of the adjustment of the operation difficulty after being affected by the environment, and provides a reference for subsequent comprehensive calculations.

[0133] Step S366: Perform nonlinear combination mapping on the environment adjustment difficulty value and the height challenge index to obtain the operation difficulty coefficient.

[0134] In an embodiment of the present invention, a nonlinear combination mapping is performed on the environmental adjustment difficulty value obtained in step S365 and the altitude challenge index calculated in step S364 to reflect their impact on the overall difficulty of the operation under different weights and coupling states. A logistic regression function, a fuzzy comprehensive evaluation, or a neural network nonlinear mapping model can be used here to ensure that the challenge index has different marginal influence weights in different input intervals. For example, a fuzzy logic control setting is used: when the environmental difficulty is high but the altitude is low, the overall operation difficulty is moderate; when both are high, an exponential upward trend is shown. Taking "height rainstorm inspection" as an example, the environmental adjustment difficulty value is 0.72, the altitude challenge index is 0.85, and the overall operation difficulty coefficient output after nonlinear mapping is 0.91, indicating that the task belongs to a high-risk difficulty level and can be used as a basis for deploying highly qualified personnel or initiating auxiliary support measures.

[0135] Preferably, step S4 includes the following steps:

[0136] Step S41: Calculate the physiological safety tolerance range based on the individual baseline parameters to obtain the physiological risk threshold parameters;

[0137] The embodiment of the present invention calculates the safe allowable range of various physiological indicators of workers, such as heart rate variability, galvanic skin response, and electromyographic activity, based on personal baseline parameters. These physiological indicators are set based on the worker's daily physiological state and work intensity, with reference to certain physiological tolerance standards. For example, for heart rate variability, the normal range is 60-100ms. Exceeding this range may indicate physical fatigue or stress response. On this basis, physiological models are used to set physiological risk thresholds taking into account factors such as personal health status, job type, and work intensity. For example, if heart rate variability exceeds its normal fluctuation range by more than 30%, the system can mark it as a potential physiological risk. Ultimately, personalized physiological risk threshold parameters are formed to monitor the worker's current physiological safety status.

[0138] Step S42: Calculate the behavior deviation from the safety limit based on the behavior pattern characteristics to obtain the behavior risk threshold parameter;

[0139] The embodiment of the present invention analyzes the behavioral data such as posture changes, movement regularity, and stability displayed by workers during the operation process based on the behavioral pattern features extracted in step S357. By comparing the workers' common postures and typical movements, the deviation of each behavioral pattern is calculated, that is, the difference between the worker's current behavior pattern and the baseline behavior pattern. For example, if a worker has an unstable posture or the duration of the movement exceeds the set safety limit during high-altitude operation, these abnormal behaviors will be regarded as risky behaviors. Based on historical data and combined with safe operation standards, a risk threshold for behavioral deviation is set, and when the deviation exceeds the set safety range, a behavioral risk threshold parameter is generated. For example, a certain operation standard stipulates that the time for a single rotation action should not exceed 10 seconds, and behaviors that exceed this time are marked as potential risk behaviors. The threshold parameters can be personalized according to specific operation tasks.

[0140] Step S43: establishing an environmental risk tolerance upper limit according to the environmental tolerance parameter to obtain an environmental risk threshold parameter;

[0141] This embodiment of the present invention analyzes the impact of the environment on workers based on environmental tolerance parameters (such as temperature, humidity, air pressure, and air velocity). By analyzing the correlation between environmental parameters and safety incidents in historical operational data, the maximum tolerance range of these environmental parameters for worker health and safety is determined. For example, if a worker works in an environment with a temperature of 40°C and a humidity of 80% for a long period of time, they may suffer from physiological problems such as overheating or dehydration. Therefore, in practical applications, by accumulating environmental data and comparing it with operational safety, an upper limit on environmental risk tolerance is generated, such as a safety threshold range for a given temperature and humidity combination. If this range is exceeded, the environmental risk threshold parameter triggers an alarm, indicating potential danger.

[0142] Step S44: adjusting the physiological risk threshold parameter, the behavioral risk threshold parameter, and the environmental risk threshold parameter according to the task difficulty coefficient to obtain a personalized risk threshold matrix;

[0143] In an embodiment of the present invention, the physiological risk threshold, behavioral risk threshold, and environmental risk threshold obtained in steps S41, S42, and S43 are combined with the operation difficulty coefficient to comprehensively adjust the above thresholds. The operation difficulty coefficient is obtained through the analysis in the aforementioned steps S361 to S366, reflecting the influence of multidimensional factors such as the working environment, task complexity, and worker adaptability. For example, if a worker is engaged in high-altitude work or needs to operate heavy equipment, the operation difficulty coefficient is relatively high. In this case, the physiological, behavioral, and environmental risk thresholds need to be further lowered to ensure safety. By weighted adjustment of each risk threshold parameter (such as based on weighted average or fuzzy logic methods), a personalized risk threshold matrix is ​​ultimately obtained. This matrix reflects the multidimensional risk assessment results of workers under different working conditions and can effectively guide workers to take appropriate preventive measures.

[0144] Step S45: performing real-time risk deviation calculation on the personalized risk threshold matrix and the multi-dimensional risk precursor feature set;

[0145] The embodiment of the present invention uses the personalized risk threshold matrix generated in step S44 and the multi-dimensional risk precursor feature set obtained in step S25 to calculate the risk deviation of real-time worker operation data. This step uses dynamic monitoring technology, such as real-time physiological monitoring equipment, motion tracking systems, and environmental monitoring sensors, to obtain workers' physiological status, behavioral characteristics, and environmental parameters in real time. By comparing these real-time data with the personalized risk threshold matrix, the deviation of each data item is calculated. For example, if a worker's real-time heart rate variability exceeds the upper and lower limits of their personalized physiological threshold, the system will calculate its deviation and give a corresponding risk level. This process can dynamically reflect the immediate risk of workers' operations and promptly warn of potential safety hazards.

[0146] Step S46: Perform time series prediction analysis on the real-time risk deviation and calculate the potential fall risk probability in the next 30 seconds in combination with the dangerous area map of the working environment;

[0147] This embodiment of the present invention performs a time series analysis on the real-time risk deviation calculated in step S45 and, in combination with a map of hazardous areas in the work environment, predicts the potential fall risk within the next 30 seconds. Specifically, this method uses a time series prediction model (such as an ARIMA model or a long short-term memory (LSTM) network) to model the real-time risk deviation data, predicting risk trends in the short term. The method also calculates the potential fall risk probability based on the distribution of hazardous areas in the work environment (such as high-altitude work and slippery areas). For example, in the event of increased wind speed or platform sway, the system, combined with the high-risk areas in the environmental hazardous area map, predicts the fall risk a worker may face and provides a corresponding risk value.

[0148] Step S47: Convert the potential falling risk probability into a risk level score to form a risk prediction report.

[0149] In the embodiment of the present invention, the potential fall risk probability calculated in step S46 is converted into a specific risk level score using a risk scoring model. The risk level score is usually based on a preset scoring standard, and the potential risk probability is divided into different levels such as low risk, medium risk and high risk. For example, if the potential fall risk probability is 0.05, it is rated as low risk, if it is 0.25, it is medium risk, and if it is 0.7, it is high risk. Finally, the system generates a risk prediction report, which includes the worker's current safety status, potential risk assessment, possible future safety incidents and corresponding early warning measures, to help safety management personnel take preventive measures in a timely manner and reduce the probability of accidents.

[0150] It is particularly important that step S5 includes the following steps:

[0151] Divide warning levels according to risk prediction reports and obtain graded warning information;

[0152] Obtain the worker's current location data and the three-dimensional model data of the work scene;

[0153] Select the warning method based on the graded warning information and trigger the corresponding warning feedback to obtain dynamic risk situation data;

[0154] Based on the dynamic risk situation data and the worker's current location data, risk visualization is marked on the three-dimensional model of the work scene to obtain a high-altitude work risk situation map;

[0155] Generate safety monitoring instructions based on the high-altitude operation risk situation map and pre-acquired personnel location information, and issue alarms to relevant operators through the call system.

[0156] Based on the risk prediction report generated above, the system in this embodiment of the present invention categorizes risks into different warning levels based on different risk probabilities and assessment criteria. Warning levels are typically categorized as low, medium, and high. Each level corresponds to different response measures and processes. For example, if the probability of a potential fall risk is less than 0.1, it is assessed as low risk, and the system will only provide a warning. If the probability is between 0.1 and 0.4, it is considered medium risk, and the system will initiate appropriate monitoring and alert management personnel. If the probability exceeds 0.4, it is considered high risk, and the system will immediately trigger warning measures and require immediate emergency response. The basis for categorizing warning levels is typically derived from a comprehensive analysis of multiple factors, including historical accident data, the risk factor of the work environment, and deviations in worker work behavior. The system requires real-time access to the worker's current location data, typically using positioning devices such as GPS modules, Bluetooth positioning, and RFID sensors to accurately locate the specific location of the work area. Secondly, the system also requires 3D model data of the current work scene. This data is typically derived from a previously constructed work scene model or 3D image data generated through technologies such as laser scanning and photogrammetry. Using this data, the system can clearly understand the worker's location and the specific details of the work environment, providing a foundation for subsequent risk analysis and early warning. For example, in a complex high-altitude work environment, workers may be located on platforms at varying heights. Accurate worker location and scene model data allows for more precise risk prediction and management. Based on the warning level determined in step S51, the system selects the appropriate warning method. For example, for low-risk situations, the system may only provide a visual alert, such as displaying a warning icon; for medium-risk situations, the system may alert the worker and safety supervisor through voice warnings or vibrations; and for high-risk situations, the system may trigger more urgent early warning feedback, such as automatically broadcasting an alarm, illuminating a red warning light, or even providing specific safety instructions and escape routes to the worker through voice commands. The specific warning feedback method can be customized based on the work environment, individual worker differences, and equipment conditions to ensure timely and effective transmission of risk information. Risks are visually marked by combining the worker's current location data obtained in step S52 with the three-dimensional model data of the work scene. During this process, the worker's real-time location is dynamically annotated in the 3D model. As the risk of the work environment changes, risk areas in the model are presented with different colors, brightness, icons, and other methods. For example, if a worker is on an aerial work platform in an area with a high risk of falling, this area will be marked as red in the 3D scene, and the worker's location may be marked as a flashing warning point. This dynamic risk visualization clearly shows the current risks faced by workers and possible danger zones, providing managers and workers with real-time risk assessment and decision-making basis.Based on the high-altitude work risk situation map generated in step S54 and combined with the workers' real-time location information, safety monitoring instructions are generated for relevant workers. This instruction alerts the workers through the call system or voice broadcast system. For example, when the system detects a high-risk area, it automatically broadcasts to the workers through the loudspeaker: "Attention, you have entered a high-risk area for falling. Please take protective measures immediately." At the same time, the system can also pass these instructions to the safety monitor or on-site person in charge to ensure that the workers can receive timely supervision and guidance. In addition, the system can also issue instructions as needed to require workers to stop the current high-risk work or perform specific safety actions, such as using fall protection equipment, adjusting the work position, etc., to ensure work safety.

[0157] This aspect also provides a visual monitoring and early warning system for preventing falls during high-altitude operations, which is used to implement the above-mentioned visual monitoring and early warning method for preventing falls during high-altitude operations. The visual monitoring and early warning system for preventing falls during high-altitude operations includes:

[0158] The multi-source data acquisition module is used to collect workers' physiological status data to generate worker fatigue characteristics and attention distraction index; collect behavioral posture data to analyze movement regularity; collect work environment parameters and evaluate the work platform stability index;

[0159] The interactive feature extraction module is used to interactively correlate worker fatigue characteristics, attention distraction index, movement regularity, and work platform stability index to obtain worker-environment interaction status data; and to extract risk precursor features from the worker-environment interaction status data to form a multi-dimensional risk precursor feature set;

[0160] The baseline parameter and behavior pattern module is used to establish individual baseline parameters based on pre-acquired historical safety operation data, and record typical operation postures to construct behavior pattern characteristics; the operation difficulty coefficient is calculated based on the complexity of the operation type and the rate of change of environmental conditions in the historical safety operation data;

[0161] The risk assessment and prediction module is used to generate a personalized risk threshold matrix based on individual baseline parameters, behavioral pattern characteristics, and task difficulty coefficients. It calculates the risk deviation degree in real time based on the personalized risk threshold matrix and multi-dimensional risk precursor feature set, predicts the risk development trajectory, and assesses the potential fall risk probability to form a risk prediction report.

[0162] The early warning and visual feedback module is used to generate graded early warnings and visual feedback based on the risk prediction report, form a high-altitude operation risk situation map, and issue alarms to operators through the call system.

[0163] The present invention is therefore intended to be illustrative and non-restrictive in all respects, with the scope of the invention being defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the application documents are intended to be embraced therein.

[0164] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A visual monitoring and early warning method for preventing falling during high-altitude operations, characterized in that: The following steps are involved: Step S1: Collecting the worker's physiological state data to generate the worker's fatigue characteristics and attention distraction index; Collect behavioral posture data to analyze movement regularity; collect work environment parameters and evaluate the work platform stability index; collect worker physiological state data to generate fatigue characteristics and attention distraction index, specifically: Collect workers' heart rate variability parameters, including RR interval standard deviation and low-frequency / high-frequency power ratio; Collect skin electrical response data, including 0.01-1.0 Skin conductance levels in the S range and spontaneous fluctuation frequency of 0–12 beats per minute; Measure the power density changes in the 20-200 Hz frequency band and the muscle fatigue index of 10-60% to obtain the electromyographic activity signal; Extract heart rate irregularity index based on heart rate variability parameters and obtain cardiogenic fatigue factor; The sympathetic nerve activity level was calculated based on the galvanic skin response data; Fatigue analysis is performed on cardiogenic fatigue factors, sympathetic nerve activity levels, and myoelectric activity signals to generate a comprehensive fatigue score; The fatigue accumulation rate is calculated based on the time trend of the comprehensive fatigue score to form the worker fatigue characteristics; Analyze the mutation characteristics of electromyographic activity signals, identify small jitters and uncoordinated patterns, and calculate the rate of decrease in movement accuracy; The stability of the autonomic nervous system was assessed based on the consistency of fluctuations between heart rate variability parameters and galvanic skin response data; Cognitive load index was calculated based on the rate of decline in motor accuracy and the stability of the autonomic nervous system; Based on the change pattern of cognitive load indicators, an attention distraction index is generated; Step S2: Interactively correlate worker fatigue characteristics, distraction index, movement regularity, and work platform stability index to obtain worker-environment interaction status data; extract risk precursor features from the worker-environment interaction status data to form a multidimensional risk precursor feature set; Step S3: Establishing individual baseline parameters based on pre-acquired historical safety operation data, and recording typical operation postures to construct behavioral pattern characteristics; Calculate the operation difficulty coefficient based on the operation type complexity and environmental condition change rate in historical safety operation data; Step S4: Generate a personalized risk threshold matrix based on individual baseline parameters, behavioral pattern characteristics, and task difficulty coefficient; Calculate the risk deviation in real time based on the personalized risk threshold matrix and multi-dimensional risk precursor feature set, predict the risk development trajectory, and assess the potential fall risk probability to generate a risk prediction report; Step S5: Generate graded warnings and visual feedback based on the risk prediction report, form a high-altitude operation risk situation map, and send an alarm to the operator through the call system.

2. The visual monitoring and early warning method for preventing falling during high-altitude operations according to claim 1 is characterized in that: The collected behavior posture data and analysis of movement regularity in step S1 include: Collect the worker's trunk tilt parameters, including forward tilt angle 0-90°, side tilt angle ±45°, and torsion angle ±60°; Measure the displacement of ±25cm on the X axis and ±20cm on the Y axis relative to the standard standing position to obtain the center of gravity offset data Collect key joint motion information, including the range of motion and six-axis motion trajectory of the shoulder, wrist, knee, and ankle joints, to form raw posture data; Perform time series processing on the sampling frequency of the original attitude data and extract the attitude stability index within every 30 seconds; Compare the trunk tilt parameters with the preset standard posture template of the work task and calculate the posture deviation; A steady-state analysis of the center of gravity offset data was performed in a 3-second time window to extract acceleration mutation points of ±2 cm / s², identify body sway patterns with a frequency of 0.5-5 Hz, and obtain the body balance index. Based on the posture stability index, the trajectory deviation of ±3 cm within the 95% confidence interval within the action cycle was extracted to obtain the joint trajectory repeatability; The regularity of the movement is calculated based on the posture deviation, body balance index, and joint trajectory repeatability with a weight ratio of 0.3:0.4:0.

3.

3. The visual monitoring and early warning method for preventing falling during high-altitude operations according to claim 1 is characterized in that: Collecting working environment parameters and evaluating the working platform stability index in step S1: Collect vibration frequency data of the work platform, measuring the amplitude distribution and main frequency components in the 1-100 Hz frequency range, and recording the peak acceleration of 0.1-5.0g; Collect environmental wind parameters, including wind speed of 0-25m / s and wind direction changes of 0-360°; Measure the surface friction coefficient of the work platform, ranging from 0.15-0.95 dynamic friction value and 0.2-1.0 static friction value; Monitor the load distribution on the platform, record local pressure data from 0-2000kg / m², and measure the tilt of the working platform; Extract the energy ratio of human-sensitive frequencies within the range of 5-25Hz from the vibration frequency data of the work platform and calculate the resonance risk index; Calculate the estimated lateral pressure under wind force based on the ambient wind parameters and generate a risk weighting factor based on the height of the working platform; Calculate the critical slip risk threshold based on the surface friction coefficient and the tilt state of the work platform; Generate current friction state based on surface friction coefficient and local pressure data; Compare the critical slip risk threshold with the current friction state to derive the slip risk probability; Analyze local pressure data to calculate the center deviation index and edge overload risk, and obtain load unevenness; The resonance risk index, risk weighting coefficient, sliding risk probability and load unevenness are used to calculate the working platform stability index based on a weight distribution of 0.25:0.3:0.25:0.

2.

4. The visual monitoring and early warning method for preventing falling during high-altitude operations according to claim 3 is characterized in that: Step S2 includes the following steps: Step S21: Time-synchronize and pair the worker fatigue characteristics with the attention distraction index to generate a cognitive state matrix; Step S22: mapping the action regularity into a behavior stability index in the range of 0-1; Step S23: analyzing the time series changes of the working platform stability index and extracting the platform stability fluctuation characteristics; Step S24: constructing a three-dimensional correlation map based on the cognitive state matrix, the behavioral stability index, and the platform stability fluctuation characteristics to form worker-environment interaction state data; Step S25: extract risk precursor features from the worker-environment interaction status data to form a multi-dimensional risk precursor feature set.

5. The visual monitoring and early warning method for preventing falling during high-altitude operations according to claim 4 is characterized in that: Step S25 includes the following steps: Conduct time series analysis on cognitive components in worker-environment interaction data and extract cognitive load fluctuation curves; Use cognitive load fluctuation curve to calculate the distribution of attention critical points and identify attention risk windows; Perform pattern recognition on behavioral components in worker-environment interaction state data to extract postural instability intervals; Analyze the temporal consistency between postural instability intervals and platform stability fluctuation characteristics to generate an environmentally induced behavioral risk index; Identify high-risk state combinations based on attention risk windows and worker-environment interaction state data; associate high-risk state combinations with workspace coordinates to form a risk space distribution map; Predict cognitive decline trends based on the cognitive state matrix and generate early warning indicators for attention decline; Generate an environmental hazard zone map based on platform stability fluctuation characteristics and working environment parameters; A dynamic risk association model is established through the spatiotemporal correlation between attention loss warning indicators and environmental danger zone maps; The risk weight of the risk spatial distribution map is adjusted based on the risk level classification results of the dynamic risk association model to form a multidimensional risk precursor feature set.

6. The visual monitoring and early warning method for preventing falling during high-altitude work according to claim 5 is characterized in that: Step S3 includes the following steps: Step S31: extracting the average heart rate variability parameters, galvanic skin response baseline and normal range of myoelectric activity of workers from the pre-acquired historical safety operation data, and constructing a personal physiological parameter profile; Step S32: Marking posture features in the historical safe operation data that have a frequency exceeding a preset threshold as common posture features to obtain a human posture baseline value, where the common posture features include the habitual torso tilt angle, center of gravity distribution pattern, and joint movement preference; Step S33: analyzing the adaptability of workers under different environmental conditions and extracting environmental tolerance parameters; Step S34: integrating the personal physiological parameter profile, common posture characteristics and environmental tolerance parameters into personal baseline parameters; Step S35: Constructing behavioral pattern features based on the historical safety operation data recording typical operation postures; Step S36: Calculate the operation difficulty coefficient based on the operation type complexity and environmental condition change rate in the historical safety operation data.

7. The visual monitoring and early warning method for preventing falling during high-altitude work according to claim 6 is characterized in that: Step S36 includes the following steps: Step S361: extracting the operation type classification information from the historical safety operation data, and calibrating the difficulty level according to the preset difficulty standard data to obtain the basic difficulty coefficient of the operation type; Step S362: extracting environmental parameter records from historical safety operation data, calculating the rate of change per unit time, and obtaining environmental condition change rate data; Step S363: Analyze the correlation between the environmental condition change rate data and the operation safety to obtain the environmental impact factor; Step S364: Calculating a height risk coefficient based on the working height information in the historical safe operation data, thereby generating a height challenge index; Step S365: constructing a difficulty benchmark value based on the basic difficulty coefficient of the task type, and generating an environmental impact factor in combination with the environmental impact factor; Step S366: Perform nonlinear combination mapping on the environment adjustment difficulty value and the height challenge index to obtain the operation difficulty coefficient.

8. The visual monitoring and early warning method for preventing falling during high-altitude work according to claim 7 is characterized in that: Step S4 includes the following steps: Step S41: Calculate the physiological safety tolerance range based on the individual baseline parameters to obtain the physiological risk threshold parameters; Step S42: Calculate the behavior deviation from the safety limit based on the behavior pattern characteristics to obtain the behavior risk threshold parameter; Step S43: establishing an environmental risk tolerance upper limit according to the environmental tolerance parameter to obtain an environmental risk threshold parameter; Step S44: adjusting the physiological risk threshold parameter, the behavioral risk threshold parameter, and the environmental risk threshold parameter according to the task difficulty coefficient to obtain a personalized risk threshold matrix; Step S45: performing real-time risk deviation calculation on the personalized risk threshold matrix and the multi-dimensional risk precursor feature set; Step S46: Perform time series prediction analysis on the real-time risk deviation and calculate the potential fall risk probability in the next 30 seconds in combination with the dangerous area map of the working environment; Step S47: Convert the potential falling risk probability into a risk level score to form a risk prediction report.

9. A visual monitoring and early warning system for preventing falling during high-altitude operations, characterized in that: Used to execute the visual monitoring and early warning method for preventing falling from high altitude work according to claim 1, the visual monitoring and early warning system for preventing falling from high altitude work comprises: The multi-source data acquisition module is used to collect workers' physiological status data to generate worker fatigue characteristics and attention distraction index; collect behavioral posture data to analyze movement regularity; collect work environment parameters and evaluate the work platform stability index; The interactive feature extraction module is used to interactively correlate worker fatigue characteristics, attention distraction index, movement regularity, and work platform stability index to obtain worker-environment interaction status data; and to extract risk precursor features from the worker-environment interaction status data to form a multi-dimensional risk precursor feature set; The baseline parameter and behavior pattern module is used to establish individual baseline parameters based on pre-acquired historical safety operation data, and record typical operation postures to construct behavior pattern characteristics; the operation difficulty coefficient is calculated based on the complexity of the operation type and the rate of change of environmental conditions in the historical safety operation data; The risk assessment and prediction module is used to generate a personalized risk threshold matrix based on individual baseline parameters, behavioral pattern characteristics, and task difficulty coefficients. It calculates the risk deviation degree in real time based on the personalized risk threshold matrix and multi-dimensional risk precursor feature set, predicts the risk development trajectory, and assesses the potential fall risk probability to form a risk prediction report. The early warning and visual feedback module is used to generate graded early warnings and visual feedback based on the risk prediction report, form a high-altitude operation risk situation map, and issue alarms to operators through the call system.

Images