Smart safety hat system

Abstract

The embodiment of the application provides a kind of intelligent safety helmet system.The system includes: data acquisition module, for collecting the work data of field personnel, work data includes geographical environment data, physiological state data, motion state data, voice data and image data;Data processing module is used for pre-processing, feature extraction and data analysis to work data, and generates work guidance information and safety risk level based on analysis result;Path planning module is used for generating work path and safety boundary range based on geographical environment data;Remote monitoring and rescue module is used for generating alarm information when detecting abnormal state, and alarm information, geographical environment data and field image information are sent to remote management platform, to realize remote monitoring and rescue.Through the above system, the real-time, accuracy and collaboration of power field risk identification, work guidance and emergency rescue are improved.

Description

Technical Field

[0001] This application relates to the field of smart wearable devices, and more particularly to a smart safety helmet system. Background Technology

[0002] In power industry field operations, personnel often need to perform equipment inspection, maintenance, and emergency repair tasks in complex and high-risk environments. For example, scenarios such as high-altitude tower work, substation equipment maintenance, or emergency repairs after natural disasters present the following challenges: The working environment is complex, requiring personnel to carry various auxiliary tools (such as lighting equipment, walkie-talkies, photographic equipment, and positioning devices), leading to cumbersome operations and potential safety hazards; if technical difficulties arise during operations, traditional telephone communication is insufficient to clearly convey on-site information and quickly obtain expert guidance; there are safety risks such as impacts, falls, and unconsciousness at the work site, and a lack of real-time monitoring and early warning mechanisms; furthermore, equipment information, maintenance knowledge, and operating procedures are scattered, and new employees rely on experienced staff for guidance, resulting in low efficiency and a high risk of errors. Currently, the demand for intelligent, integrated, and real-time power operations is increasingly urgent, necessitating intelligent equipment that can integrate multimodal data acquisition, intelligent analysis, remote collaboration, and safety early warning functions to improve operational efficiency, reduce safety risks, and optimize management processes.

[0003] In current technology, power field operations mainly rely on a combination of traditional safety helmets and distributed tools. Traditional safety helmets only provide basic protection, so workers need to carry additional tools. Regarding technical support, workers must contact experts by phone or walkie-talkie to obtain assistance when encountering problems.

[0004] However, existing safety helmets cannot provide workers with real-time and accurate technical support, and the operation of multiple independent devices carried in addition is cumbersome and prone to safety hazards. Summary of the Invention

[0005] This application provides an intelligent safety helmet system to solve the problems in the prior art that it is difficult to provide workers with real-time and accurate technical support, and that the operation of multiple independent devices carried in addition is cumbersome and prone to safety hazards.

[0006] In a first aspect, embodiments of this application provide an intelligent safety helmet system, comprising:

[0007] The data acquisition module is used to collect work data from on-site personnel. The work data includes geographical environment data, physiological state data, motion state data, voice data, and image data.

[0008] The data processing module is used to preprocess, extract features, and analyze the operation data, and generate operation guidance information and safety risk levels based on the analysis results;

[0009] The path planning module is used to generate the operation path and safety boundary range based on the geographic environment data.

[0010] The remote monitoring and rescue module is used to generate alarm information when an abnormal state is detected, and send the alarm information, geographical environment data and on-site image information to the remote management platform to realize remote monitoring and rescue.

[0011] Based on the aforementioned technical means, the data acquisition module collects geographical environment data, physiological state data, motion state data, voice data, and image data of on-site personnel. Combined with the data processing module, path planning module, and remote monitoring and rescue module for collaborative processing, it realizes real-time perception, intelligent analysis, and remote linkage management of on-site operation status. This can improve the intelligence level of on-site operations, reduce on-site safety risks, and enhance the efficiency of operation guidance and emergency rescue capabilities.

[0012] In one possible implementation, the data acquisition module further includes: a triaxial accelerometer, a temperature sensor, a heart rate and blood oxygen sensor, a camera, a positioning unit, and an audio unit;

[0013] The triaxial accelerometer is used to collect acceleration data in different directions in real time in order to identify abnormal movement behavior of the personnel on site.

[0014] The temperature sensor is used to collect the ambient temperature of the work environment;

[0015] The heart rate and blood oxygen sensor is used to collect heart rate and blood oxygen saturation data of on-site personnel.

[0016] The camera is used to collect images and / or video data of the work site;

[0017] The positioning unit is used to obtain the real-time geographical location of the operator;

[0018] The audio unit is used to collect voice signals and realize two-way voice communication.

[0019] Based on the aforementioned technical means, by setting up a three-axis accelerometer, temperature sensor, heart rate and blood oxygen sensor, camera, positioning unit and audio unit, multi-dimensional data collection of on-site personnel's movement status, physiological status, environmental status and work scenario information is realized, thereby improving the completeness and real-time nature of on-site data acquisition, and providing a reliable data foundation for subsequent safety risk identification and intelligent analysis.

[0020] In one possible implementation, the data processing module includes a motion state analysis unit;

[0021] The motion state analysis unit is used to perform the following steps: calculate the three-axis vector magnitude based on the acceleration data collected by the three-axis accelerometer, and perform sliding window smoothing on the three-axis vector magnitude to obtain acceleration feature data; extract the peak acceleration, average magnitude, and fluctuation amplitude features from the acceleration feature data; and identify abnormal motion behavior based on the peak acceleration, average magnitude, and fluctuation amplitude features.

[0022] Based on the above technical means, by performing triaxial vector magnitude calculation, sliding window smoothing, and peak acceleration, average magnitude, and fluctuation amplitude feature extraction on triaxial acceleration data, accurate identification of abnormal motion behaviors such as falls, impacts, and violent shaking can be achieved, thereby improving the accuracy and stability of motion anomaly detection and reducing the occurrence of false alarms.

[0023] In one possible implementation, the data processing module further includes a physiological state analysis unit.

[0024] The physiological state analysis unit is used to calculate heart rate variability data based on the heart rate data, and construct a comprehensive feature vector based on the heart rate data, blood oxygen saturation data, and heart rate variability data; and to identify the risk level of the comprehensive feature vector based on a time series model to output the corresponding risk level result.

[0025] Based on the aforementioned technical means, by fusing and analyzing heart rate data, blood oxygen saturation data, and heart rate variability data, and combining them with time series models for risk level identification, dynamic assessment of the physiological state of on-site personnel is achieved. This enables timely detection of risks such as fatigue, hypoxia, or physical abnormalities, thereby improving the life safety protection capabilities of on-site workers.

[0026] In one possible implementation, the data processing module further includes a voice processing unit;

[0027] The speech processing unit is used to perform endpoint detection on the acquired speech signal to obtain speech data; extract speech feature vectors from the speech data based on the Mel-Cepstral Coefficient feature extraction algorithm; and recognize the speech feature vectors based on the speech recognition model to obtain text information.

[0028] Based on the aforementioned technical means, by performing endpoint detection, Mel-frequency cepstral coefficient feature extraction, and speech recognition processing on the speech signal, the speech content of on-site personnel can be converted into text information, thereby improving the accuracy of speech recognition in complex working environments and providing a data foundation for subsequent intelligent question answering and semantic understanding.

[0029] In one possible implementation, when performing endpoint detection, the voice processing unit detects valid voice frames in the voice signal based on short-time energy and short-time zero-crossing rate. When the short-time energy is greater than a preset silence threshold and the short-time zero-crossing rate is within a preset range, the corresponding voice frame is determined to be a valid voice frame, and the voice data corresponding to the valid voice frame is output.

[0030] Based on the above technical means, by detecting effective speech frames based on short-time energy and short-time zero-crossing rate, the effective speech part and the silence part in the speech signal can be effectively distinguished, thereby reducing the impact of environmental noise and invalid speech data on the speech recognition results and improving the efficiency and accuracy of speech recognition processing.

[0031] In one possible implementation, the data processing module further includes a knowledge question answering unit, which is used to perform word segmentation and dependency parsing on the text information to identify user intent, and to perform matching retrieval in the knowledge base based on the identified user intent, and output corresponding technical guidance information.

[0032] Based on the aforementioned technical means, by performing word segmentation and dependency parsing on text information, and combining it with knowledge base matching and retrieval, the recognition of user voice intent and the intelligent output of technical guidance information are realized. This can improve the efficiency of on-site operators in obtaining technical support and reduce reliance on human experience.

[0033] In one possible implementation, the data processing module further includes an image analysis unit, which is used to analyze the collected image and / or video data according to the target detection algorithm to identify the location of personnel and / or safety hazards in the work environment, and generate on-site risk warning information based on the identification results.

[0034] Based on the aforementioned technical means, by using target detection algorithms to analyze image and / or video data, intelligent identification of personnel locations, safety hazards, and on-site conditions in the work environment is achieved, thereby enabling timely detection of potential risks and improving on-site safety monitoring and risk warning capabilities.

[0035] In one possible implementation, the path planning module is used to construct a path model based on the starting position, target position, and environmental terrain data corresponding to the geographical environment data of the on-site personnel; generate the optimal operation path based on the path optimization algorithm; set a safety boundary based on the operation area range; and trigger a position offset alarm when it is detected that the operator has exceeded the safety boundary.

[0036] Based on the aforementioned technical means, by constructing a path model based on the starting position, target position, and environmental terrain data, and combining it with a path optimization algorithm to generate the optimal operation path and set safety boundaries, intelligent planning and range monitoring of on-site personnel's operation paths are realized, thereby improving operation efficiency and reducing the safety risks caused by accidentally entering dangerous areas.

[0037] In one possible implementation, the remote monitoring and rescue module is used to generate an abnormal alarm event when abnormal movement behavior is detected, and upload the geographical environment data, physiological state data and image data corresponding to the abnormal alarm event to the remote management platform.

[0038] Based on the aforementioned technical means, by generating abnormal alarm events when abnormal movement behavior is detected, and uploading the corresponding geographical environment data, physiological state data, and image data to the remote management platform, remote real-time monitoring and coordinated rescue of abnormal on-site conditions are realized, thereby shortening emergency response time and improving accident rescue efficiency and on-site safety assurance capabilities.

[0039] This application provides an intelligent safety helmet system. The system first acquires real-time work data from on-site personnel through a data acquisition module, thereby achieving multi-dimensional data perception of on-site personnel, the work environment, and work status, improving the completeness and real-time nature of on-site information acquisition. Subsequently, a data processing module preprocesses, extracts features, and analyzes the acquired work data. Specifically, it performs vector magnitude calculation and feature analysis on motion state data to identify abnormal movement behaviors such as falls and impacts; it performs fusion analysis on physiological state data to determine the health risk level of workers; it performs speech recognition and semantic analysis on voice data to obtain user intent and generate corresponding work guidance information; and it performs target recognition and risk detection on image data to identify on-site safety hazards. This achieves intelligent analysis and risk assessment of on-site work status, improving the level of intelligence in on-site operations. The system enhances safety and security capabilities. Furthermore, the path planning module constructs a path model based on geographic environmental data and generates work paths and safety boundaries by combining target location, terrain, and safe zone range. It uses path optimization algorithms to plan safe work routes for on-site personnel and monitors personnel activity in real time. When personnel deviate from the safe zone, a position deviation alarm is triggered, thereby improving on-site work efficiency and reducing safety risks caused by accidental entry into dangerous areas. Finally, the remote monitoring and rescue module generates alarm information when abnormal conditions are detected and uploads the alarm information, geographic environmental data, and on-site image information to the remote management platform. This allows managers to remotely obtain on-site status, personnel location, and risk information, and promptly initiate remote voice interaction, video monitoring, and emergency rescue processes, thereby shortening accident response time and improving on-site emergency rescue efficiency and overall safety assurance capabilities. Attached Figure Description

[0040] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0041] Figure 1 Schematic diagram of the structure of the smart safety helmet system provided in the embodiments of this application Figure 1 ;

[0042] Figure 2 Schematic diagram of the structure of the smart safety helmet system provided in the embodiments of this application Figure 2 ;

[0043] Figure 3 Schematic diagram of the structure of the smart safety helmet system provided in the embodiments of this application Figure 3 .

[0044] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0045] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application.

[0046] In power industry field operations, personnel often need to perform equipment inspection, maintenance, and emergency repair tasks in complex and high-risk environments. For example, scenarios such as high-altitude tower work, substation equipment maintenance, or emergency repairs after natural disasters present the following challenges: The working environment is complex, requiring personnel to carry various auxiliary tools (such as lighting equipment, walkie-talkies, photographic equipment, and positioning devices), leading to cumbersome operations and potential safety hazards; if technical difficulties arise during operations, traditional telephone communication is insufficient to clearly convey on-site information and quickly obtain expert guidance; there are safety risks such as impacts, falls, and unconsciousness at the work site, and a lack of real-time monitoring and early warning mechanisms; furthermore, equipment information, maintenance knowledge, and operating procedures are scattered, and new employees rely on experienced staff for guidance, resulting in low efficiency and a high risk of errors. Currently, the demand for intelligent, integrated, and real-time power operations is increasingly urgent, necessitating intelligent equipment that can integrate multimodal data acquisition, intelligent analysis, remote collaboration, and safety early warning functions to improve operational efficiency, reduce safety risks, and optimize management processes.

[0047] In current technologies, power field operations primarily rely on a combination of traditional safety helmets and distributed tools. Traditional safety helmets only provide basic protection and cannot integrate multimodal sensors (such as accelerometers, temperature sensors, and heart rate monitoring modules), requiring workers to carry additional lighting equipment, walkie-talkies, and positioning devices, which is inconvenient and prone to missing critical information. Regarding technical support, workers need to contact experts by phone or walkie-talkie when encountering problems, but voice communication struggles to convey crucial information such as on-site images and environmental data, and is affected by signal stability, resulting in low efficiency for remote guidance. In terms of safety monitoring, existing solutions largely rely on manual observation or post-event analysis, unable to detect dangerous conditions such as falls, impacts, or unconsciousness in real time, nor can they dynamically track work paths to avoid restricted areas. Furthermore, power field knowledge bases and operating procedures are mostly in the form of paper documents or scattered electronic materials, lacking a unified intelligent question-and-answer system, leading to high learning costs for new employees and limited operational efficiency.

[0048] Most existing on-site safety protection equipment only possesses basic protective functions and lacks the ability to perceive and intelligently analyze the status of workers, the on-site environment, and operational risks in real time. This leads to problems such as fragmented information acquisition, untimely technical guidance, insufficient safety risk early warning capabilities, and low efficiency in emergency rescue during on-site operations. Especially in complex work scenarios such as power inspection, equipment maintenance, and emergency repairs, workers often need to carry multiple tools such as walkie-talkies, lighting equipment, positioning equipment, and recording equipment, which not only increases the operational burden but also makes it difficult to obtain timely and effective support when encountering technical problems or dangerous situations. Based on this, the inventors recognized that with the development of Internet of Things (IoT), artificial intelligence (AI), and multi-sensor fusion technologies, intelligent management of on-site operations can be achieved by integrating functions such as on-site personnel status perception, intelligent voice interaction, image recognition and analysis, path planning, and remote monitoring and rescue. Therefore, an intelligent safety helmet system has been proposed. In practical implementation, the data acquisition module first uses multi-source sensors to collect real-time data on the geographical environment, physiological state, motion state, voice, and image data of on-site personnel, achieving a comprehensive perception of the on-site environment and personnel status. Subsequently, the data processing module preprocesses, extracts features, and performs intelligent analysis on the collected data. Specifically, by calculating the three-axis vector magnitude, smoothing the data using a sliding window, and extracting features from the acceleration data, it identifies abnormal motion behaviors such as falls, impacts, and violent shaking. By fusing and analyzing heart rate, blood oxygen, and heart rate variability data, it dynamically assesses the physical condition and risk level of on-site personnel. Finally, by performing endpoint detection, Mel-frequency cepstral coefficient feature extraction, and semantic analysis on the voice signal, it achieves voice content recognition, user intent understanding, and knowledge recognition. The system performs database matching and retrieval to output corresponding technical guidance information. Simultaneously, it combines target detection algorithms to analyze image and / or video data, enabling the identification of on-site personnel locations, equipment status, and safety hazards. Furthermore, the path planning module constructs a path model based on on-site personnel location data, target locations, and environmental terrain data. It then uses path optimization algorithms to generate safe operating paths and safety boundary ranges, thereby intelligently planning and monitoring the work routes of on-site personnel. When the system detects abnormal states such as falls, impacts, physiological abnormalities, or deviations from safe areas, the remote monitoring and rescue module automatically generates alarm information and uploads the alarm information, on-site location data, physiological status data, and image data to the remote management platform. This allows back-end administrators to monitor the on-site situation in real time and quickly initiate remote monitoring and emergency rescue procedures.The above technical solutions enable intelligent management of on-site personnel, work environment, and work risks throughout the entire process. This not only improves the safety and efficiency of on-site operations but also enhances risk warning capabilities, remote technical guidance capabilities, and emergency rescue capabilities. As a result, it effectively reduces safety hazards in complex work scenarios and reduces reliance on human experience and traditional auxiliary equipment.

[0049] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will be described below with reference to the accompanying drawings.

[0050] Figure 1 Schematic diagram of the structure of the smart safety helmet system provided in the embodiments of this application Figure 1 ; Figure 2 Schematic diagram of the structure of the smart safety helmet system provided in the embodiments of this application Figure 2 ; Figure 3 Schematic diagram of the structure of the smart safety helmet system provided in the embodiments of this application Figure 3 ;like Figures 1 to 3 As shown, the system includes: a data acquisition module 10, a data processing module 20, a path planning module 30, and a remote monitoring and rescue module 40. The data acquisition module 10 also includes a triaxial accelerometer 101, a temperature sensor 102, a heart rate and blood oxygen sensor 103, a camera 104, a positioning unit 105, and an audio unit 106. The data processing module 20 includes a motion state analysis unit 201, a physiological state analysis unit 202, a voice processing unit 203, a knowledge question answering unit 204, and an image analysis unit 205.

[0051] The data acquisition module is used to collect operational data from on-site personnel. This data includes geographical environment data, physiological state data, motion state data, voice data, and image data. Specifically, a triaxial accelerometer 101 is installed on the top or side of the smart safety helmet to collect real-time acceleration data in the X, Y, and Z directions. Continuous sampling acquires information on posture changes during personnel movement to identify abnormal motion behaviors. The sampling frequency can be set to 50Hz–100Hz to ensure real-time response to sudden events such as falls, impacts, and violent shaking. The design utilizes the vector synthesis principle of triaxial acceleration to uniformly represent motion states in different directions. The formula for calculating the triaxial vector magnitude is as follows:

[0052]

[0053] in, These represent the acceleration values ​​in the X, Y, and Z directions, respectively. This method eliminates detection bias caused by movement in a single direction, thereby improving the accuracy of abnormal movement behavior identification. For example, when workers are walking normally, the magnitude of the three-axis vector usually remains near the gravitational acceleration. However, in the event of a fall or impact, there will be a sudden change in the instantaneous peak value, thus allowing for rapid identification of dangerous situations through changes in the magnitude.

[0054] Furthermore, to avoid the impact of on-site vibration, equipment shaking, and random noise on the detection results, the motion state analysis unit 201 performs sliding window smoothing on the acquired three-axis vector magnitude. This is achieved by selecting the first N sampling points at the current moment and performing mean filtering to reduce misjudgments caused by accidental fluctuations. The calculation process is as follows:

[0055]

[0056] in, This represents the smoothed result at the current moment, where N represents the sliding window length. For example, when the system collects data every 10ms, it can select the five most recent sampled values ​​for smoothing calculation to obtain more stable motion trend data. Subsequently, the system further extracts peak acceleration, average modulus, and fluctuation amplitude features, where fluctuation amplitude is the difference between the maximum and minimum values. By fusing these multiple features, the system determines whether there are abnormal motion behaviors such as falls, collisions, or violent shaking among personnel on site, thereby improving the stability and reliability of abnormal behavior identification.

[0057] Temperature sensor 102 (such as an NTC thermistor) is used to collect the ambient temperature at the work site. Based on the temperature change characteristics of the thermistor, the actual ambient temperature is calculated by detecting the voltage change corresponding to the change in the thermistor's resistance. The specific calculation formula is as follows:

[0058]

[0059] in, This refers to the voltage signal value output by the temperature sensor (such as an NTC thermistor divided by a fixed resistor), where T is the actual ambient temperature. The reference resistor is the fixed resistance value in the voltage divider circuit connected in series with the thermistor (unit: ohms). ); The power supply voltage for the sensor circuit (unit: volts V).

[0060] When the ambient temperature exceeds a preset threshold, such as above 60℃ or below -20℃, the system automatically generates a temperature anomaly alarm, alerting personnel to stop work or evacuate the danger zone to prevent accidents such as burns from high temperatures or frostbite from low temperatures. This design enables real-time monitoring of harsh environments, improving operational safety in complex environments.

[0061] The heart rate and oxygenation sensor 103 is mainly used for real-time monitoring of the physiological state of on-site personnel. Its design principle is based on PPG (Photoplethysmography) detection technology. It emits infrared and visible light to illuminate the surface of human skin and uses a photodetector to receive the light signals reflected or transmitted through human tissue. Since the absorption rate of hemoglobin in human blood for different wavelengths of light changes periodically with the heartbeat cycle, the system can acquire corresponding pulse waveform signals based on changes in light intensity. In practice, the heart rate and oxygenation sensor 103 can be installed at the point where the helmet strap contacts the human skin, such as the forehead or behind the ears, to ensure stable light signal acquisition. The sensor continuously acquires PPG waveform data at a frequency of 1 Hz to 10 Hz and transmits it to the physiological state analysis unit 202 for processing.

[0062] During data processing, the physiological state analysis unit 202 first filters the original PPG waveform to remove power frequency interference, motion artifacts, and random noise. A bandpass filter can be used to preserve the effective frequency range corresponding to the human pulse, for example, 0.5Hz to 5Hz, thereby improving the accuracy of pulse peak detection. Subsequently, the system performs peak detection on the filtered PPG waveform to identify the pulse peak point corresponding to each heartbeat, and calculates the real-time heart rate (HR) based on the time interval between adjacent pulse peaks. The calculation formula is as follows:

[0063] HR=60 / RR

[0064] Here, RR represents the time interval between two adjacent pulse peaks, measured in seconds. Using this method, the system can acquire real-time information on the heart rate changes of personnel on-site.

[0065] Furthermore, the system calculates blood oxygen saturation using the ratio of the absorption of infrared light to that of visible light. The principle behind this is that oxyhemoglobin and deoxyhemoglobin have different absorption rates for different wavelengths of light. Therefore, the blood oxygen level can be estimated by the ratio of the intensity of the two wavelengths of light. The calculation process can be expressed as follows:

[0066]

[0067] Where R represents the ratio of the AC to DC components of infrared light to visible light, and A and B are empirical calibration coefficients. Using this method, the system can detect in real time whether personnel on-site are at risk of oxygen deficiency.

[0068] Subsequently, the system further calculates heart rate variability (HRV) to reflect the physical workload, mental stress, and fatigue levels of on-site personnel. Specifically, the system calculates HRV parameters based on the changes in the RR interval between multiple consecutive pulse peaks. For example, the standard deviation of the RR interval (SDNN) can be used as an evaluation index, and its calculation formula is as follows:

[0069]

[0070] Among them, RR i This represents the i-th RR interval. This represents the average RR interval. Typically, when field personnel are engaged in prolonged high-intensity work or are under high stress, HRV will decrease significantly. Therefore, changes in HRV can be used to identify fatigue or abnormal physiological states.

[0071] After obtaining heart rate, blood oxygen saturation, and HRV data, the physiological state analysis unit 202 further constructs a comprehensive feature vector to describe the overall physiological state of the personnel on site within the current time window. Specifically, the system fuses physiological parameters collected at multiple times within a fixed time window (e.g., 30s or 60s) to construct the following feature vector:

[0072]

[0073] in, This indicates the current heart rate data. This indicates the current blood oxygen saturation data. This represents the heart rate variability data at the current moment. This represents ambient temperature data. This represents motion state data. By integrating physiological data with environmental and motion state data, a more comprehensive reflection of the actual working status of personnel on site can be obtained.

[0074] Subsequently, the system inputs the comprehensive feature vector sequence over a continuous time period into the time series model for dynamic risk identification. In this embodiment, an LSTM (Long Short-Term Memory) model can be used to model the time series data. Its design principle lies in the fact that LSTM can utilize memory units to store historical states over a long period, thereby identifying the dynamic trends of physiological states changing over time. For example, when personnel are working continuously in high-temperature environments, the system can not only detect the current increase in heart rate, but also combine this with the past trends of continuously increasing heart rate, continuously decreasing HRV, and gradually decreasing blood oxygen to determine whether personnel are at risk of fatigue, heatstroke, or hypoxia.

[0075] In practice, the system inputs feature vectors from multiple consecutive time points into the LSTM network in chronological order, for example:

[0076]

[0077] The LSTM network updates historical states through input gates, forget gates, and output gates, extracts dynamic change features from the time series, and finally outputs the corresponding hidden state vector. Subsequently, the system inputs the hidden state vector into a multilayer perceptron (MLP) classifier to classify the current physiological state into risk levels. The output can be divided into normal state, low-risk state, moderate-risk state, and high-risk state.

[0078] For example, when the system detects that a worker's heart rate remains above 120 beats per minute for five consecutive minutes, their HRV (heart rate variability) significantly decreases, and their blood oxygen saturation is below 90%, the system can determine that the worker is at risk of severe fatigue or hypoxia and automatically generate a physiological abnormality alarm. Simultaneously, the corresponding physiological status data is uploaded to the remote management platform to alert management personnel for timely intervention. Through this design, the system can not only detect the instantaneous physiological state of workers on site but also analyze trends in physical condition changes over time, thereby improving the accuracy of abnormal physiological state identification and early warning capabilities, and ensuring continuous protection of the life safety of on-site workers.

[0079] Camera 104 is used to collect images and / or video data of the work site. Its design principle is to perform real-time visual perception of the site environment through high-definition image acquisition equipment, and transmit the acquired image frames or video streams to the image analysis unit 205 for intelligent analysis and processing. In specific implementation, camera 104 can be a wide-angle high-definition camera, installed at the front of the smart safety helmet to ensure that the site environment within the worker's field of vision can be completely captured. Camera 104 continuously acquires video data at a preset frame rate, for example, acquiring real-time video streams at a frame rate of 25fps to 30fps, and transmits them to the image analysis unit 205 via a wireless communication module or local bus interface. Through this method, the system can obtain real-time information on the status of on-site equipment, personnel behavior, and environmental changes, thereby providing basic data support for subsequent risk identification and remote monitoring.

[0080] During image analysis, the image analysis unit 205 first preprocesses the acquired image and / or video data to improve the accuracy of subsequent target recognition. Specifically, the image analysis unit 205 performs size normalization, image enhancement, and noise suppression on the input image. For example, histogram equalization can enhance image contrast in low-light environments, and Gaussian filtering can reduce image noise caused by dust, rain, fog, or electromagnetic interference, thereby improving image quality in complex working environments. Subsequently, the system further inputs the processed image into a deep learning target detection model for analysis.

[0081] In this embodiment, the image analysis unit 205 can use the YOLO (You Only Look Once) target detection algorithm to perform real-time detection of on-site target objects. Its design principle is to use a convolutional neural network to extract features from the input image and directly complete target location prediction and category recognition during a single network inference process. Specifically, the system first divides the input image into multiple grid regions. Each grid is responsible for predicting the target bounding box and target category probability within the corresponding region. The bounding box parameters can be expressed as:

[0082] B=(x,y,w,h,c)

[0083] Where x and y represent the center coordinates of the target bounding box, w and h represent the width and height of the target bounding box, and c represents the target confidence score. The system extracts edge features, texture features, and semantic features from images using a convolutional neural network, and identifies target objects in the field based on a trained model, such as workers, safety helmets, insulating gloves, electrical equipment, smoke, flames, and foreign objects. Because the YOLO algorithm uses a single-stage detection structure, it features fast detection speed and high real-time performance, meeting the needs of real-time risk identification in the field.

[0084] In another implementation, the image analysis unit 205 can also employ the Faster-RCNN object detection algorithm for high-precision recognition. Specifically, it first generates candidate target regions using a Region Proposal Network (RPN), then classifies and regresses bounding boxes on these regions, thereby improving detection accuracy for small targets and complex scenes. For example, in power maintenance scenarios, when workers wear small or partially obscured insulating gloves, Faster-RCNN can improve the accuracy of identifying insulating protective equipment. Through this multi-model detection mechanism, the system can balance real-time performance and detection accuracy according to the needs of different work scenarios.

[0085] Furthermore, after identifying the target object, the image analysis unit 205 performs behavioral analysis and risk assessment on the target detection results. In specific implementation, the system can construct on-site risk rules based on target category, target location relationship, and target status information. For example, when the system detects that on-site workers are not wearing insulated gloves, not wearing safety helmets, or entering high-voltage hazardous areas, it can be determined as a risk of personnel violating regulations; when the system detects smoke, flames, electric arc flashes, or foreign object accumulation in the equipment area, it can be determined as a risk of abnormal equipment operation; when the system detects that on-site personnel are stationary for a long time or have fallen to the ground, it can further determine whether a personnel fall accident has occurred by combining the motion state analysis results. Through the above methods, the system can achieve automatic identification and real-time monitoring of on-site safety hazards.

[0086] To improve the stability of target recognition results, the image analysis unit 205 further incorporates continuous time frames for target tracking analysis. Specifically, the system establishes the target's motion trajectory by analyzing changes in the target's position within consecutive video frames and uses a target association algorithm to determine the target's motion state. For example, when personnel continuously approach the boundary of a danger zone across multiple consecutive video frames, the system can generate early warning information instead of triggering an alarm only after entering the danger zone, thereby increasing the lead time for risk warnings.

[0087] After risk identification is completed, the image analysis unit 205 generates corresponding on-site risk warning information and uploads the risk type, risk level, on-site image, and target location data to the remote management platform. For example, when the system detects smoke in the equipment area, it can automatically capture the corresponding risk image and generate an alarm message of "suspected fire in equipment area"; when the system detects that the operator is not wearing insulated gloves, it can generate an alarm message of "personnel protective equipment missing" for violation of operating procedures. After receiving the alarm information, the remote management platform can display the on-site image and the location of the risk target in real time and notify the back-end management personnel for processing.

[0088] For example, in a power emergency repair scenario, when on-site workers enter a high-voltage power distribution area to inspect equipment, camera 104 continuously collects on-site video streams, and image analysis unit 205 monitors the status of on-site personnel's protective equipment and the operating status of the equipment in real time. When the system detects that the workers are not wearing insulated gloves and there are electric arcs in the nearby equipment area, the system immediately generates a high-level risk alarm and simultaneously uploads the on-site video, personnel location, and risk information to the remote management platform. Back-end management personnel can use remote voice communication to remind on-site personnel to stop work and evacuate the danger zone, thereby effectively reducing the risk of electric shock accidents.

[0089] Through the above design, the system can realize intelligent analysis of on-site images and video data, which not only improves the ability to identify on-site personnel behavior and monitor equipment status, but also enhances the real-time early warning capability and remote collaborative management capability of on-site safety hazards, thereby effectively improving the overall safety assurance level in complex operating environments.

[0090] The positioning unit 105 is primarily used to acquire the real-time geographical location of personnel on site. By fusing positioning data from both the GPS and BeiDou positioning systems, it improves the accuracy and stability of location acquisition in complex working environments. Specifically, the positioning unit 105 simultaneously receives GPS and BeiDou satellite signals, jointly calculates the current location of personnel, and optimizes the positioning results by incorporating time synchronization mechanisms, satellite signal quality considerations, and multipath error correction algorithms. Because power line inspections, substation maintenance, and underground or mountainous work environments are prone to issues such as building obstruction, electromagnetic interference, and unstable signals from a single satellite system, the dual-satellite fusion positioning method enhances continuous positioning capabilities and anti-interference capabilities in weak signal environments. For example, when GPS satellite signals are blocked, the system can still maintain positioning continuity using BeiDou satellite signals, thereby ensuring the real-time nature and reliability of personnel location data.

[0091] Furthermore, the path planning module 30 constructs a path model based on the real-time location of on-site personnel, the target location, and environmental terrain data acquired by the positioning unit 105. The environmental terrain data may include information such as on-site road structure, equipment distribution, hazardous area locations, obstacle areas, and restricted areas. The system first abstracts the entire work area into a topological map structure composed of multiple path nodes and connecting edges. Each path node represents a passable location, and the connecting edges between nodes represent the corresponding path access relationships.

[0092] To improve the safety of path planning, the system further assigns corresponding path weights to different path edges. Path weights include not only path distance but also factors such as hazard level, obstacle complexity, and environmental risk coefficient. For example, path edges near high-voltage equipment areas correspond to higher risk weights, while ordinary passage areas correspond to lower risk weights.

[0093] During path generation, the path planning module 30 employs a path optimization algorithm to search for path nodes in the work area to generate a safe work path with the minimum total path weight. In this embodiment, Dijkstra's algorithm can be used for optimal path search. Its design principle is based on a greedy strategy to progressively search for the path node with the minimum current cost and continuously update the shortest path distance between the starting node and each target node. Specifically, the system takes the current position of the personnel on site as the starting node S, the target work area as the target node G, and calculates the cumulative path cost between the starting node and each candidate node. Its objective function can be expressed as:

[0094]

[0095] Where C(P) represents the total cost of the current path, This represents the distance of the i-th path segment. This represents the risk weight of the corresponding path segment. The system continuously compares the total cost of each candidate path and ultimately obtains the optimal path with the minimum total cost. By introducing a risk coefficient into the path weight, the system can not only select the shorter path but also prioritize avoiding high-risk areas, thereby improving the safety of path planning.

[0096] For example, in a substation inspection scenario, there are high-voltage equipment areas, construction areas, and temporary obstacle areas on site. The system first constructs a path node network based on the site map and assigns higher risk weights to paths near high-voltage equipment areas. When on-site personnel travel from the entrance area to the target equipment for maintenance, the path planning module 30 comprehensively considers path distance and risk level, automatically generating a safe inspection path that avoids dangerous areas. Simultaneously, when the system detects changes in the site environment, such as the addition of obstacles, abnormal temperature in equipment areas, or an expansion of dangerous areas, the path planning module 30 can dynamically update the path model based on real-time environmental data and replan the work path, thereby improving the path adaptability in complex and dynamic environments.

[0097] Furthermore, the path planning module 30 also constructs electronic safety boundaries based on the work area. Specifically, the system generates corresponding electronic fence boundaries according to the work permit range, hazardous equipment areas, and prohibited areas, and determines in real time whether the current position of on-site personnel exceeds the safety boundary range. When the system detects that the on-site personnel's position exceeds the preset safety area, it automatically triggers a position offset alarm and uploads the alarm information, current position, and on-site image data to the remote management platform. For example, when on-site personnel accidentally enter a high-voltage hazardous area, the system can immediately issue a voice alarm prompt and simultaneously notify the back-end management personnel for remote intervention, thereby reducing the risk of safety accidents caused by accidentally entering a hazardous area.

[0098] Furthermore, the path planning module 30 can optimize the path planning results by combining historical path data and on-site operational behavior data. For example, the system can record the actual travel paths, dwell times, and occurrences of risk events by on-site personnel in different work scenarios, and adaptively adjust the path weights based on historical data, thereby improving the rationality and accuracy of subsequent path planning results. Through the above design, the system can not only achieve real-time acquisition of on-site personnel locations, but also realize safe path planning and dynamic boundary monitoring based on environmental risk information, thereby improving on-site operational efficiency, enhancing the ability to control hazardous areas, and effectively reducing safety risks in complex work environments.

[0099] The audio unit 106 is used to acquire on-site voice signals and realize two-way voice communication. It utilizes a microphone array to directionally acquire on-site voice data and combines voice enhancement and noise reduction technologies to improve the quality of voice acquisition in complex working environments. Specifically, the audio unit 106 can employ an array structure composed of multiple microphones, installed on both sides or the front of the smart safety helmet. Beamforming technology is used to enhance the direction of on-site personnel's voice while suppressing noise signals generated by surrounding equipment operation, wind noise, and environmental electromagnetic interference. For example, in substations, high-voltage equipment areas, or large machinery construction sites, where environmental noise is strong, traditional single-microphone acquisition methods are prone to voice distortion. However, a microphone array can improve the signal-to-noise ratio of the target voice signal, thereby enhancing the on-site voice acquisition effect. The acquired voice data is then transmitted to the voice processing unit 203 for analysis and processing.

[0100] During speech processing, the speech processing unit 203 first preprocesses the original speech signal. The purpose of this preprocessing is to extract the valid speech portion from the continuous audio stream and remove silence segments and background noise to reduce the impact of invalid data on subsequent speech recognition results. Specifically, the system first divides the continuous speech signal into multiple speech frames with a fixed time length, for example, each frame is set to a length of 20ms to 30ms. A certain overlap is maintained between adjacent frames to ensure the continuity of the speech signal. Subsequently, the system performs endpoint detection on each speech frame based on short-time energy and short-time zero-crossing rate.

[0101] Among them, short-time energy is mainly used to reflect the energy changes of the speech signal within a short period of time. Its design principle is to calculate the sum of squares of the amplitudes of all sampling points in the current speech frame to reflect whether the current speech frame contains valid speech information. The calculation formula is as follows:

[0102]

[0103] Here, x(m) represents the original speech signal, and w(m) represents the window function. Typically, the short-time energy of the speech frame corresponding to a person speaking is significantly higher than that of a silent state. Therefore, the system can use short-time energy to distinguish between valid speech and silent regions. For example, when a person issues a voice command, the speech frame energy increases rapidly, while it remains at a lower energy level when not speaking.

[0104] Furthermore, the system also uses the short-time zero-crossing rate to analyze the frequency changes of the speech signal. The short-time zero-crossing rate represents the number of symbol changes in a speech signal per unit time, and its calculation formula is as follows:

[0105]

[0106] Where N represents the number of sampling points in the current speech frame, sgn(x) represents the sign function, and ZCR reflects the number of frequent signal switching. When the speech signal frequency is high or there is a lot of noise, the short-time zero-crossing rate is usually high; while the zero-crossing rate of vowel parts in speech is relatively low. Therefore, the system can combine short-time energy and short-time zero-crossing rate to more accurately detect the effective speech region. For example, when there is continuous mechanical noise in the scene, although some noise frames may have high energy, their zero-crossing rate characteristics are different from those of real speech. Therefore, the system can use dual features for joint judgment to avoid misidentification.

[0107] In practice, the speech processing unit 203 first calculates the short-time energy and short-time zero-crossing rate for each frame of speech data, and compares the calculation results with preset thresholds. When the short-time energy is greater than the preset silence threshold and the short-time zero-crossing rate is within the preset range, the system determines the current speech frame as a valid speech frame and retains the corresponding speech data; otherwise, the corresponding speech frame is treated as a silence frame or background noise frame and discarded. Through this method, the system can automatically extract valid speech segments, thereby reducing the computational load of subsequent speech recognition models and improving the accuracy of speech recognition in complex environments.

[0108] After endpoint detection, the system further performs feature extraction processing on the valid speech frames. In this embodiment, the speech processing unit 203 uses the Mel Frequency Cepstral Coefficients (MFCC) algorithm to extract speech feature vectors. By simulating the human ear's perception characteristics of different frequencies of sound, the speech signal is converted from the time domain to the frequency domain, and spectral feature information in the speech is extracted. Specifically, the system first performs a Fast Fourier Transform (FFT) on the valid speech frames to convert the speech signal into a spectral signal; then, the spectral signal is filtered through a Mel filter bank to obtain a frequency distribution that conforms to the characteristics of human hearing; subsequently, logarithmic operations and Discrete Cosine Transform (DCT) are performed on the filtering results to generate the corresponding MFCC feature vectors. Through the above processing, the system can effectively extract the pronunciation features in the speech and reduce the impact of environmental noise and speaker differences on the speech recognition results.

[0109] For example, when a field worker issues a voice command to "open the equipment inspection record," the system first extracts the valid voice segment through endpoint detection, and then generates the corresponding voice feature vector using the MFCC algorithm.

[0110]

[0111] in, This represents the cepstral feature parameters in the corresponding frequency band. The system then inputs this speech feature vector into the speech recognition model for recognition.

[0112] During speech recognition, the system can employ speech recognition models based on HMM (Hidden Markov Model), DNN (Deep Neural Network), or Transformer structures to decode and analyze speech feature vectors. Its design principle is to learn the mapping relationship between speech features and text through a large amount of speech training data, thereby converting the input speech content into corresponding text information. For example, when a person on-site issues the voice command "Query yesterday's equipment temperature record," the system can recognize the corresponding speech content as the text information "Query yesterday's equipment temperature record" and send this text information to the knowledge question-answering unit 204 for subsequent semantic analysis and knowledge retrieval.

[0113] Furthermore, to further improve the stability of speech recognition in complex environments, the system can also correct the recognition results by combining contextual semantic information and industry terminology dictionaries. For example, in power inspection scenarios, the system can prioritize the recognition of professional terms such as "insulator," "circuit breaker," and "busbar," thereby reducing the misrecognition rate of ordinary general speech recognition models in professional scenarios.

[0114] Through the above design, the audio unit 106 and the voice processing unit 203 can achieve high-quality acquisition of on-site voice signals, effective voice detection, feature extraction and voice recognition processing. This not only improves the accuracy of voice recognition in complex noise environments, but also enhances the voice interaction capability between on-site personnel and the system, thereby improving the ease of operation, efficiency of technical guidance and remote collaborative operation capability during on-site operations.

[0115] The knowledge-based question-answering unit 204 performs semantic analysis, user intent recognition, and knowledge matching retrieval on the text information generated after speech recognition. By combining natural language processing technology and knowledge base retrieval technology, it enables intelligent voice interaction between on-site personnel and the smart safety helmet system. In specific implementation, the knowledge-based question-answering unit 204 receives the text information output by the speech processing unit 203 and first preprocesses the text information, including text normalization, invalid character filtering, and stop word removal, to improve the accuracy of subsequent semantic analysis. For example, when an on-site personnel issues the voice command "Help me check yesterday's inspection record for transformer No. 1," the system first filters out redundant interjections in the text and standardizes the time expression format, thereby forming standardized semantic text.

[0116] Subsequently, the knowledge-based question-answering unit 204 performs word segmentation on the text information. Its design principle is to break down continuous text into word units with actual semantic meaning, facilitating subsequent intent recognition and semantic analysis. For example, for the text information "Show yesterday's security inspection records," the system uses a Chinese word segmentation algorithm to break it down into keywords such as "show," "yesterday," and "security inspection records." Here, "show" corresponds to the action, "yesterday" corresponds to the time condition, and "security inspection records" corresponds to the query target. Through this method, the system can extract key semantic information from continuous sentences, providing a foundation for subsequent semantic understanding.

[0117] After word segmentation, the knowledge-based question-answering unit 204 further utilizes dependency parsing to analyze the semantic relationships within the text. Its design principle is to identify subject-predicate, verb-object, and modification relationships by analyzing the dependencies between words in a sentence, thereby recognizing the user's true intent. For example, for the statement "Query yesterday's temperature alarm record for device number three," the system can not only identify keywords such as "query," "device number three," and "temperature alarm record," but also determine that "yesterday" corresponds to the time-limited condition of "alarm record," rather than a description of the device's status. Through dependency parsing, the system can improve the accuracy of semantic understanding in complex sentence scenarios.

[0118] Furthermore, the knowledge-based question-answering unit 204 constructs a user intent vector based on word segmentation and dependency parsing results to represent the semantic features corresponding to the current statement. For example, the user statement can be mapped to the following structured semantic expression:

[0119] I=[Action,Target,Time,Location,Condition]

[0120] Here, Action represents the action, such as "query," "display," or "open"; Target represents the target object, such as "inspection records" or "equipment status"; Time represents the time condition; Location represents the location object; and Condition represents additional constraints. For the statement "Display yesterday's security inspection records," its corresponding intent vector can be represented as:

[0121] I=[Display, Security Inspection Records, Yesterday, NULL, NULL]

[0122] Through the above methods, the system can convert natural language statements into structured semantic information, thereby improving the efficiency of subsequent knowledge matching.

[0123] After obtaining the user's intent, the knowledge-based question-answering unit 204 further performs knowledge matching and retrieval based on the knowledge base. In specific implementation, the system pre-constructs a knowledge base related to on-site operations. This knowledge base may include equipment operating procedures, fault handling plans, safe operating procedures, historical inspection records, emergency response plans, and expert experience documents. Various documents in the knowledge base can be pre-indexed with corresponding keywords and semantic feature vectors to improve subsequent retrieval efficiency.

[0124] During the knowledge matching process, the system uses text similarity algorithms to match the user's intent vector with candidate documents in the knowledge base. For example, the cosine similarity algorithm can be used to calculate the similarity between the user's semantic vector and the document vectors in the knowledge base; the calculation formula is as follows:

[0125]

[0126] Here, Q represents the user intent vector, and K represents the document feature vector in the knowledge base. The system calculates the similarity between different documents and the current user intent, and selects the knowledge content with the highest similarity as the matching result. Through this method, the system can achieve rapid and intelligent retrieval of on-site technical issues.

[0127] For example, when on-site personnel issue a voice command such as "How to handle abnormal transformer temperature," the system first identifies the user's intent as "fault handling query" through semantic analysis. It then retrieves relevant fault handling solutions, safe operating procedures, and historical cases from the knowledge base and returns the matching results to the on-site personnel. The returned results can be output via voice broadcast or displayed on a terminal. For example, the system can output guidance such as "Please first check if the transformer cooling system is operating normally and check if the load current exceeds the limit," thereby assisting on-site personnel in quickly handling fault problems.

[0128] Furthermore, to improve the accuracy of knowledge-based question answering in complex scenarios, the knowledge-based question answering unit 204 can also perform continuous semantic analysis by combining contextual historical dialogue information. For example, when a person on site first asks "Query the status of transformer No. 1" and then continues to ask "Retrieve yesterday's alarm records as well," the system can automatically identify "yesterday's alarm records" as "transformer No. 1" based on contextual information, thereby achieving intelligent semantic association in continuous dialogue scenarios.

[0129] Furthermore, the knowledge-based question-and-answer unit 204 can also dynamically recommend knowledge based on on-site environmental data. For example, when the system detects abnormal equipment temperature or personnel entering high-risk areas, it can proactively push relevant safety operating procedures or risk warnings to on-site personnel without requiring them to actively search for information, thereby improving on-site risk prevention and control capabilities.

[0130] Through the above design, the knowledge question answering unit 204 can realize intelligent semantic analysis of on-site voice text, user intent recognition, and knowledge matching retrieval. This not only improves the efficiency of obtaining on-site technical guidance information, but also enhances the intelligent interaction capability between on-site personnel and the system, thereby reducing the dependence of on-site operations on human experience and improving the efficiency of fault handling and safety management capabilities in complex operating environments.

[0131] The remote monitoring and rescue module 40 generates anomaly alarm events when the system detects abnormal states such as falls, impacts, physiological abnormalities, prolonged stillness, or deviation from safe areas. It then uploads corresponding geographical environmental data, physiological state data, and image data to the remote management platform to achieve remote monitoring, remote coordination, and emergency rescue of on-site risk events. Its design principle involves integrating motion state analysis results, physiological state analysis results, positioning data, and on-site image analysis results to comprehensively assess the current safety status of on-site personnel. When preset risk conditions are met, an alarm mechanism is automatically triggered, thereby transforming traditional passive rescue into proactive early warning and real-time coordinated rescue.

[0132] In practical implementation, the remote monitoring and rescue module 40 first receives the abnormal movement behavior recognition results output by the motion state analysis unit 201, the risk level results output by the physiological state analysis unit 202, the position boundary status output by the path planning module 30, and the risk recognition results output by the image analysis unit 205, and then performs fusion analysis on the multi-source state data. For example, when the system detects that the instantaneous peak value of the triaxial acceleration of the on-site personnel exceeds the preset impact threshold, and the personnel remain stationary for a period of time afterward, the system can determine that the on-site personnel are at risk of falling; when the system detects that the on-site personnel's heart rate is continuously abnormally high, their blood oxygen saturation is continuously below the preset threshold, and their HRV is significantly reduced, the system can determine that the on-site personnel are at risk of physiological abnormalities such as fatigue, heatstroke, or hypoxia; when the system detects that the on-site personnel's position exceeds the preset safety boundary range, the system can determine that the on-site personnel are at risk of accidentally entering a dangerous area. Through the above-mentioned multi-source information fusion method, the system can reduce the false alarm problem caused by false alarms from a single sensor and improve the accuracy and reliability of abnormal state recognition.

[0133] Furthermore, the remote monitoring and rescue module 40 automatically generates an abnormal alarm event after identifying an abnormal state. Specifically, the system constructs an alarm event data packet based on the abnormality type, risk level, duration, and on-site environmental conditions, the structure of which can be represented as follows:

[0134] Alarm=[Type,Level,Time,Location,Status,Image,Video]

[0135] Here, Type indicates the type of abnormal event, such as a fall, impact, physiological abnormality, or area displacement; Level indicates the risk level; Time indicates the time the alarm occurred; Location indicates the current location of the personnel on site; Status indicates the current physiological status data; and Image and Video represent on-site image and video information, respectively. Through this structured alarm method, the system can uniformly manage different types of risk events, improving the data processing efficiency of the remote platform.

[0136] After an alarm is generated, the system further uploads the alarm information to the remote management platform via a wireless communication module. In practice, the system can achieve data communication between field devices and the remote platform via 4G, 5G, or wireless LAN, and employs data compression and priority transmission mechanisms to improve data transmission efficiency in emergency alarm situations. For example, in the event of a fall, the system prioritizes uploading the alarm type, location, and key image frames to ensure that backend management personnel can obtain critical accident information immediately; subsequently, it gradually uploads the complete video stream, physiological status data, and historical behavioral data, thereby improving the efficiency of remote rescue response.

[0137] Upon receiving an alarm, the remote management platform first displays the real-time location of personnel on an electronic map based on location data, and then visualizes the accident scene using on-site images and video information. For example, the backend management interface can display the coordinates of on-site personnel, the current risk level, real-time video footage, and the distribution of surrounding equipment, helping managers quickly understand the situation. Furthermore, the remote management platform can automatically match the nearest rescue or maintenance personnel based on the location and generate the optimal rescue route to improve on-site rescue efficiency.

[0138] Meanwhile, the remote management platform can also establish remote contact with on-site personnel via voice or video communication. In practice, when the system detects that on-site personnel are at risk of falling or experiencing physiological abnormalities but are still conscious, back-end administrators can proactively inquire about the situation via two-way voice communication and guide them to take emergency measures. For example, if on-site personnel experience discomfort due to high-temperature work, back-end administrators can remotely instruct them to stop working and move to a safe area to rest, thus preventing the accident from escalating. If on-site personnel cannot respond to voice calls, the system automatically escalates the current alarm level and notifies emergency rescue personnel to immediately proceed to the scene.

[0139] For example, in a power line inspection scenario, when on-site personnel enter mountainous transmission line areas to maintain equipment, the complex terrain and harsh environment pose a risk. If a personnel fall during the inspection, the triaxial accelerometer will detect the instantaneous impact signal, the positioning unit will detect that the personnel's position has not changed for an extended period, the image analysis unit will detect the personnel's fallen state, and the physiological state analysis unit will detect abnormal heart rate fluctuations. The system automatically classifies this as a high-risk fall accident based on this combined information. Subsequently, the remote monitoring and rescue module 40 automatically generates an abnormal alarm event and uploads the accident location, on-site video, physiological state data, and risk level to the remote management platform. Back-end administrators can quickly locate the accident site using an electronic map and confirm the accident situation using the on-site video. Simultaneously, they will immediately notify the nearest maintenance personnel to proceed to the scene for rescue, significantly shortening the accident response time.

[0140] Furthermore, the remote monitoring and rescue module 40 can record and store the entire process of alarm events. Specifically, the system can save the alarm time, on-site status data, image and video data, and rescue handling process to a historical database for subsequent accident analysis, accountability, and safety management optimization. For example, managers can analyze historical alarm records to determine if high-risk events frequently occur in a particular area and further optimize on-site work processes or hazardous area delineation strategies, thereby improving the overall level of safety management.

[0141] Through the above design, the remote monitoring and rescue module 40 can not only realize real-time detection and automatic alarm of abnormal conditions on site, but also realize on-site status visualization, remote voice and video linkage, and rapid rescue dispatch, thereby improving the risk warning capability, accident response efficiency and overall on-site safety assurance capability in complex working environments.

[0142] Furthermore, the remote management platform is also used to display the real-time location of workers on an electronic map, establish remote video call connections, and generate rescue deployment instructions based on on-site video and location information, thereby achieving visualized monitoring of the accident site, remote collaborative command, and intelligent dispatch of rescue resources. By integrating positioning data, on-site video data, and real-time communication technology, the platform synchronizes the status of on-site personnel, accident location, and surrounding environment information to the remote management center in real time, enabling back-end management personnel to quickly grasp the on-site situation and implement remote rescue command.

[0143] In practice, the remote management platform first receives real-time location data uploaded by the positioning unit 105 and dynamically displays the current location of on-site personnel on an electronic map based on a GIS geographic information system. This electronic map may include a work area structure map, a hazardous area distribution map, an equipment location distribution map, and inspection route information. The system updates the on-site personnel's location in real time based on the uploaded latitude and longitude data and uses a trajectory recording function to save the historical movement paths of on-site personnel. For example, in a power transmission line inspection scenario, back-end management personnel can use the electronic map to view the current tower location of on-site inspection personnel, the inspection route, and the distance to the target equipment in real time, thereby improving the visualization and management capabilities of the on-site operation process.

[0144] Furthermore, when the system detects an abnormal alarm event, the remote management platform automatically highlights the corresponding location node on the electronic map and generates a corresponding risk label based on the risk level information. For example, when a person falls on-site, the system can display the accident location on the electronic map with a flashing red icon, while also displaying the current risk level, physiological status data, and a thumbnail of the scene image, allowing back-end administrators to quickly identify high-risk events. This approach significantly improves the efficiency of accident location and on-site dispatch.

[0145] Simultaneously, the remote management platform is also used to establish remote video call connections to enable remote interaction between rescue personnel and on-site workers. Specifically, when the system detects an abnormal situation on-site, the remote management platform automatically sends a video communication request to the smart safety helmet terminal and establishes a two-way audio and video communication link through camera 104 and audio unit 106. This utilizes a real-time audio and video transmission protocol to encode and transmit on-site video streams and voice data, thereby achieving real-time remote communication between back-end management personnel and on-site workers.

[0146] For example, when personnel on-site experience discomfort in high-voltage equipment areas, back-end administrators can monitor the environment and personnel's condition in real time via video call and remotely guide them to take emergency measures, such as cutting off equipment power, moving away from dangerous areas, or wearing protective gear. If on-site personnel are unable to operate the terminal due to an accident, the system can automatically activate video communication and continuously upload on-site video footage, allowing back-end personnel to monitor the accident scene in real time. These methods improve remote collaborative handling capabilities and emergency command efficiency after an accident.

[0147] Furthermore, the remote management platform generates rescue deployment instructions based on on-site video and location information. In practice, the system first obtains geographical information about the accident site, surrounding roads, and the distribution of nearby rescue resources based on the on-site location data, and then combines this with on-site video analysis results to determine the severity of the accident. For example, the system comprehensively assesses the accident risk level by analyzing the flame range, smoke concentration, personnel status, and equipment damage in the on-site video. Subsequently, the system automatically generates a corresponding rescue deployment plan based on the accident level.

[0148] In this embodiment, the rescue deployment instruction may include rescue personnel dispatch information, rescue route information, on-site risk warning information, and required rescue equipment information, and its structure can be represented as follows:

[0149] R=[P,L,E,W,T]

[0150] Where P represents rescue personnel information, L represents rescue route information, E represents required rescue equipment information, W represents on-site risk warning information, and T represents estimated arrival time. The system generates the optimal rescue dispatch plan based on the distance relationship between the current accident location and the location of rescue resources, and automatically selects and dispatches the rescue personnel closest to the accident site who possess the corresponding handling capabilities.

[0151] For example, in a power transmission line maintenance scenario in mountainous areas, if a worker falls and a fire breaks out in the high-voltage equipment area, the system first uses an electronic map to determine the location of the accident near the inspected line in the mountainous area. Simultaneously, it identifies the presence of flames and smoke spreading through on-site video footage. Subsequently, the remote management platform automatically generates rescue deployment instructions, including: dispatching the nearest maintenance and repair personnel to the scene, planning the optimal rescue route avoiding dangerous areas, notifying firefighters to bring insulated fire extinguishers to the accident area, and reminding rescue personnel of the risk of high-voltage electric shock. After confirmation by backend administrators, the corresponding rescue deployment instructions can be issued with a single click, thereby improving on-site emergency response efficiency.

[0152] Furthermore, the remote management platform can dynamically adjust the rescue process based on on-site video information. For example, when the system detects that the fire at the accident site is expanding or the location of personnel at the site has changed, the system can update the rescue route and risk warning information in real time, and regenerate the corresponding rescue deployment plan, thereby improving the rescue adaptability in complex and dynamic environments.

[0153] Through the above design, the remote management platform can not only realize the real-time visualization of the location of on-site personnel, but also realize remote video interaction, real-time perception of on-site status, and intelligent dispatch of rescue resources, thereby improving the ability to locate accidents, remote collaborative command, and on-site emergency rescue efficiency, and further enhancing the overall safety assurance capability in complex operating environments.

[0154] This application provides an intelligent safety helmet system. The system first acquires real-time work data from on-site personnel through a data acquisition module, thereby achieving multi-dimensional data perception of on-site personnel, the work environment, and work status, improving the completeness and real-time nature of on-site information acquisition. Subsequently, a data processing module preprocesses, extracts features, and analyzes the acquired work data. Specifically, it performs vector magnitude calculation and feature analysis on motion state data to identify abnormal movement behaviors such as falls and impacts; it performs fusion analysis on physiological state data to determine the health risk level of workers; it performs speech recognition and semantic analysis on voice data to obtain user intent and generate corresponding work guidance information; and it performs target recognition and risk detection on image data to identify on-site safety hazards. This achieves intelligent analysis and risk assessment of on-site work status, improving the level of intelligence in on-site operations. The system enhances safety and security capabilities. Furthermore, the path planning module constructs a path model based on geographic environmental data and generates work paths and safety boundaries by combining target location, terrain, and safe zone range. It uses path optimization algorithms to plan safe work routes for on-site personnel and monitors personnel activity in real time. When personnel deviate from the safe zone, a position deviation alarm is triggered, thereby improving on-site work efficiency and reducing safety risks caused by accidental entry into dangerous areas. Finally, the remote monitoring and rescue module generates alarm information when abnormal conditions are detected and uploads the alarm information, geographic environmental data, and on-site image information to the remote management platform. This allows managers to remotely obtain on-site status, personnel location, and risk information, and promptly initiate remote voice interaction, video monitoring, and emergency rescue processes, thereby shortening accident response time and improving on-site emergency rescue efficiency and overall safety assurance capabilities.

[0155] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0156] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0157] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0158] Based on the above disclosure, those skilled in the art will understand that the computer-readable storage medium and its related implementation methods provided in the embodiments of the present invention are not limited to the specific hardware or logical partitioning forms described above. In actual product or system deployments, computer execution instructions can be encapsulated in different software distribution packages or firmware modules and distributed via networks, optical discs, or mobile storage devices. When computer execution instructions are loaded and executed by one or more processors, in addition to implementing the aforementioned method steps, hardware resource configurations can be automatically identified according to specific application scenarios, and thread priorities or memory allocation strategies can be dynamically adjusted to optimize execution efficiency without manual intervention. Furthermore, the computer-readable storage medium can also be integrated into IoT terminals, edge computing nodes, or cloud servers to form a distributed instruction storage and collaborative execution system to meet differentiated needs such as high concurrency, low latency, or data privacy protection.

[0159] Furthermore, it should be emphasized that the logical functions corresponding to the methods, units, modules, and storage media in this application, without departing from the core idea of ​​this invention, allow developers to make equivalent refactoring, renaming, or re-division based on specific programming languages ​​(such as C++, Java, Python, etc.) or framework characteristics. For example, the functions of multiple "units" can be merged into a single service process, or message queues or pipe communication can be used instead of direct function calls. Even if these implementation details are not completely the same as the foregoing embodiments, as long as they ultimately achieve the same technical effect and no creative effort is required, they should all be considered to fall within the protection scope of this invention. Similarly, any intermediate data processing forms, exception handling mechanisms, or logging methods embodied in the instructions in the computer-readable storage medium when executed are also derivative technologies of this invention and should not be construed as substantial limitations on the original solution.

[0160] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0161] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A smart safety helmet system, characterized in that, Its features include: The data acquisition module is used to collect work data from on-site personnel. The work data includes geographical environment data, physiological state data, motion state data, voice data, and image data. The data processing module is used to preprocess, extract features, and analyze the operation data, and generate operation guidance information and safety risk levels based on the analysis results; The path planning module is used to generate the operation path and safety boundary range based on the geographic environment data. The remote monitoring and rescue module is used to generate alarm information when an abnormal state is detected, and send the alarm information, geographical environment data and on-site image information to the remote management platform to realize remote monitoring and rescue.

2. The system according to claim 1, characterized in that, The data acquisition module also includes: a triaxial accelerometer, a temperature sensor, a heart rate and blood oxygen sensor, a camera, a positioning unit, and an audio unit; The triaxial accelerometer is used to collect acceleration data in different directions in real time in order to identify abnormal movement behavior of the personnel on site. The temperature sensor is used to collect the ambient temperature of the work environment; The heart rate and blood oxygen sensor is used to collect heart rate and blood oxygen saturation data of on-site personnel. The camera is used to collect images and / or video data of the work site; The positioning unit is used to obtain the real-time geographical location of the operator; The audio unit is used to collect voice signals and realize two-way voice communication.

3. The system according to claim 2, characterized in that, The data processing module includes a motion state analysis unit; The motion state analysis unit is used to perform the following steps: calculate the three-axis vector magnitude based on the acceleration data collected by the three-axis accelerometer, and perform sliding window smoothing on the three-axis vector magnitude to obtain acceleration feature data; extract the peak acceleration, average magnitude, and fluctuation amplitude features from the acceleration feature data; Abnormal motion behavior is identified based on the peak acceleration, average modulus, and fluctuation amplitude characteristics.

4. The system according to claim 3, characterized in that, The data processing module also includes a physiological state analysis unit. The physiological state analysis unit is used to calculate heart rate variability data based on the heart rate data, and construct a comprehensive feature vector based on the heart rate data, blood oxygen saturation data, and heart rate variability data; and to identify the risk level of the comprehensive feature vector based on a time series model to output the corresponding risk level result.

5. The system according to claim 1, characterized in that, The data processing module also includes a voice processing unit; The speech processing unit is used to perform endpoint detection on the acquired speech signal to obtain speech data; extract speech feature vectors from the speech data based on the Mel-Cepstral Coefficient feature extraction algorithm; and recognize the speech feature vectors based on the speech recognition model to obtain text information.

6. The system according to claim 5, characterized in that, When performing endpoint detection, the speech processing unit detects valid speech frames in the speech signal based on short-time energy and short-time zero-crossing rate. When the short-time energy is greater than a preset silence threshold and the short-time zero-crossing rate is within a preset range, the corresponding speech frame is determined to be a valid speech frame, and the speech data corresponding to the valid speech frame is output.

7. The system according to claim 5, characterized in that, The data processing module also includes a knowledge question answering unit, which is used to perform word segmentation and dependency parsing on the text information to identify user intent, and to perform matching and retrieval in the knowledge base based on the identified user intent, and output corresponding technical guidance information.

8. The system according to claim 1, characterized in that, The data processing module also includes an image analysis unit, which is used to analyze the collected image and / or video data according to the target detection algorithm to identify the location of personnel and / or safety hazards in the work environment, and generate on-site risk warning information based on the identification results.

9. The system according to claim 1, characterized in that, The path planning module is used to construct a path model based on the starting position, target position, and environmental terrain data corresponding to the geographical environment data of the on-site personnel; generate the optimal operation path based on the path optimization algorithm; set safety boundaries based on the operation area; and trigger a position offset alarm when it is detected that the operator has exceeded the safety boundary.

10. The system according to claim 1, characterized in that, The remote monitoring and rescue module is used to generate an abnormal alarm event when abnormal movement behavior is detected, and upload the corresponding geographical environment data, physiological state data and image data to the remote management platform.

Images