A dangerous identification system and method based on three-dimensional data fusion comparison method

Abstract

This invention relates to a hazard identification system and method based on three-dimensional data fusion and comparison, belonging to the field of hazard identification technology. The invention preprocesses multi-source data; uses an improved Kalman filter combined with a zero-rate update algorithm to fuse and correct three-dimensional position and attitude data; based on the object-centered fusion concept, it spatially aligns and correlates features of multi-source data in a unified three-dimensional world coordinate system to generate a three-dimensional fusion dataset; and performs multi-dimensional comparative analysis based on the three-dimensional fusion dataset, a preset safety benchmark dataset, and a historical normal dataset to identify hazard sources and determine their hazard levels. This invention, through three-dimensional data fusion and multi-dimensional comparison, achieves accurate identification of spatial and composite hazard sources in complex high-risk scenarios, improving safety protection capabilities.

Description

Technical Field

[0001] This invention belongs to the field of hazard identification technology, and in particular to a hazard identification system and method based on three-dimensional data fusion and comparison. Background Technology

[0002] Current smart safety helmets still have many unresolved technical shortcomings. Existing products mostly focus on upgrading single functions or simply adding modules, failing to form a multi-dimensional and in-depth hazard identification system. Specifically, this manifests in the following ways: 1. Limited Data Comparison Methods and Poor Hazard Identification Accuracy: Existing smart safety helmets rely heavily on single-type sensors (integrated impact sensors, gas sensors, or positioning sensors) for data acquisition. Data comparison only involves a simple comparison of single-dimensional data with preset thresholds, without in-depth fusion and multi-dimensional comparative analysis of multi-source data. For example, judging environmental hazards solely based on gas concentration thresholds without combining data such as the worker's three-dimensional position, the three-dimensional shape of surrounding objects, and the human physiological state for linked analysis makes it difficult to identify complex, multi-faceted hazards in various scenarios (such as the combined hazard of a falling object's trajectory and the worker's position, or the linked hazard of toxic gas leaks and abnormal personnel posture). This leads to false alarms and missed alarms, failing to meet the safety protection needs of complex and high-risk scenarios.

[0003] 2. Lack of 3D spatial data fusion capability and lack of spatial hazard identification: Existing technologies mostly focus on the collection and comparison of 2D planar data, without introducing 3D data fusion technology. It is impossible to obtain 3D spatial information of the work scene (such as the 3D positioning of surrounding equipment, the 3D motion trajectory of falling objects from heights, the 3D boundary of the work area, etc.), making it difficult to identify potential spatial hazards, such as 3D distance warnings between falling objects from heights and workers, and spatial positioning hazards caused by equipment offset, resulting in serious safety blind spots.

[0004] 3. Outdated data fusion and comparison algorithms with poor adaptability: Although some existing smart safety helmets attempt to integrate multiple types of sensors, the data fusion is merely a simple feature overlay without employing advanced 3D data fusion algorithms. This fails to address redundancy and conflict issues between multi-source data (such as location data, environmental data, and image data), resulting in low accuracy of the fused data. Furthermore, the comparison analysis logic is simplistic, failing to dynamically compare historical normal data with real-time scene data, making it difficult to adapt to the hazard identification needs of different work scenarios and resulting in poor algorithm robustness.

[0005] Furthermore, based on existing patent search results (such as multi-functional safety helmets based on the Internet of Things, safety helmet monitoring methods based on multimodal image recognition, and smart safety helmets based on digital twins), their core technologies have not yet broken through the limitations of "single data comparison" and "simple module stacking," have not deeply integrated three-dimensional data fusion with multi-dimensional comparative analysis, and have not optimized the module integration structure for the wearable characteristics of safety helmets, resulting in significant technological gaps.

[0006] Based on the shortcomings of the existing technologies and industry needs, a hazard identification method integrating three-dimensional data fusion and comparison is developed to solve the problems of single data comparison, module stacking, and lack of spatial hazard identification in existing smart safety helmets. This method improves the accuracy, comprehensiveness, and real-time performance of hazard identification and optimizes the wearing experience, which has become an urgent technical problem to be solved. It also provides the core starting point and direction of thinking for the development of this invention. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of existing technologies and propose a hazard identification system and method based on three-dimensional data fusion and comparison. This invention addresses the technical deficiencies of existing smart safety helmets, such as single data acquisition methods, simple comparison methods, crude module integration, lack of three-dimensional spatial data fusion capabilities, and outdated data fusion and comparison algorithms with poor adaptability. By integrating and optimizing multi-source, multi-dimensional data in three dimensions, optimizing functional modules, and designing advanced three-dimensional data fusion and comparison algorithms, this invention improves the accuracy, comprehensiveness, and spatial protection capabilities of hazard identification, optimizes wearing comfort and battery life, and enhances algorithm robustness to adapt to different high-risk work scenarios.

[0008] The technical problem solved by this invention is achieved through the following technical solution: A hazard identification system based on a three-dimensional data fusion and comparison method, installed on a safety helmet, includes a data acquisition module, a main control module, a power supply module, a display module, and an alarm module. The power supply module is connected to the data acquisition module, the main control module, the display module, and the alarm module. The main control module is connected to the data acquisition module, the display module, and the alarm module. The data acquisition module is used to collect multi-source, multi-dimensional data. The main control module is used to preprocess the collected data, perform three-dimensional data fusion, multi-dimensional comparative analysis, and hazard identification. The power supply module supplies power to each module. The display module uses a miniature OLED screen. The alarm module includes a voice broadcast unit and a vibration unit. The speakers of the voice broadcast unit are embedded on both sides of the safety helmet shell, and the vibration unit is installed inside the safety helmet liner.

[0009] Moreover, the main control module adopts an improved Kalman filter algorithm combined with the object-centered fusion idea to achieve three-dimensional deep fusion of multi-source data, and constructs a multi-dimensional comparison system of three-dimensional fusion dataset + preset safety benchmark dataset + historical normal dataset + real-time three-dimensional image to achieve accurate identification of multi-dimensional hazard sources.

[0010] A hazard identification method based on a three-dimensional data fusion and comparison system includes the following steps: Step 1: The data acquisition module collects data from multiple sources; Step 2: The main control module performs noise reduction, calibration, normalization, and anomaly removal on the collected multi-source data to generate a preprocessed dataset. Step 3: The main control module calls the improved Kalman filter combined with the zero-rate update algorithm to fuse and correct the 3D position and attitude data; based on the object center fusion idea, the corrected position and attitude data are spatially aligned and feature-correlated with other data in a unified 3D world coordinate system to generate a 3D fusion dataset. Step 4: Based on the three-dimensional fusion dataset, the preset safety benchmark dataset, and the historical normal dataset, the main control module performs multi-dimensional comparative analysis to identify hazard sources and determine the hazard level. Step 5: The alarm module triggers the corresponding alarm mode based on the identified hazard source and hazard level.

[0011] Moreover, the multi-source data in step 1 includes three-dimensional position and attitude data, environmental data, three-dimensional image data, human body state data, and impact data.

[0012] Moreover, the specific implementation method of step 2 is as follows: the data preprocessing unit receives the data collected by each data acquisition module, performs noise reduction processing using a moving average filtering algorithm to eliminate data noise; calibrates the data of each sensor using preset calibration parameters to correct sensor errors; normalizes data of different magnitudes to convert them into a dataset of the same magnitude; uses the 3σ criterion to remove abnormal data, retains valid data, and generates a preprocessed data set.

[0013] Furthermore, step 3, which involves calling an improved Kalman filter combined with a zero-rate update algorithm to fuse and correct the 3D position and attitude data, includes the following steps: Step 3.1: Obtain the real-time detection value of the foot pressure sensor; Step 3.2: When the real-time detected value is greater than the preset pressure threshold, it is determined to be the gait support period, and the zero-speed update algorithm is triggered. The integral error of the inertial measurement unit is corrected through the observation update equation, wherein the zero-speed observation value is set to zero. Step 3.3: When the real-time detection value is less than or equal to the preset pressure threshold, it is determined to be an oscillation period, and state prediction of the improved Kalman filter is performed.

[0014] Furthermore, step 3, based on the object-centric fusion concept, involves the following steps to form a three-dimensional fusion dataset: Step 3.4: Establish the unified three-dimensional world coordinate system with the geometric center of the identified core object as the origin; Step 3.5: Transform the raw data collected by each sensor from its respective local coordinate system to the unified three-dimensional world coordinate system. The coordinate transformation formula is as follows: in, These are the original coordinates in the sensor's local coordinate system; for The rotation matrix describes the attitude mapping relationship between the local coordinate system and the world coordinate system, and is calculated from the attitude angles acquired by the IMU; for Translation vector, describing the positional offset between the origin of the local coordinate system and the origin of the world coordinate system; To achieve spatial alignment of data from different sensors in the transformed world coordinate system; Step 3.6: Extract the three-dimensional feature parameters of each object, and associate the environmental data, human body state data, and impact data with the three-dimensional spatial coordinates and feature parameters of the corresponding objects to generate the three-dimensional fusion dataset containing spatial information, environmental information, human body information, and impact information.

[0015] Furthermore, the specific implementation method of step 4 is as follows: Step 4.1: Calculate the relative deviation between real-time data and preset safety benchmark values, as well as the deviation between real-time data and historical normal data, to identify single indicator anomalies; Step 4.2: Calculate the trend slope of the real-time data sequence using the sliding window linear regression method to identify rapid upward or downward trends in the data; Step 4.3: Perform point cloud segmentation and target detection on the 3D image data, identify dangerous objects, and calculate the relative distance and collision probability between dangerous objects and workers; Step 4.4: Establish a linkage judgment matrix, integrate the analysis results of environment, human body status and spatial dimensions, and identify complex hazard sources; Step 4.5: Calculate the comprehensive risk value based on the preset weighting coefficients, and determine the hazard level based on the comprehensive risk value.

[0016] Furthermore, the specific calculation method for the comprehensive risk value is as follows: Sr=W1×Pc+W2×Rc+W3×min(Di,1)+W4×min(Oi,1) Where Pc is the spatial collision probability, Rc is the composite hazard correlation degree, Di is the relative deviation, Oi is the deviation degree, and W1, W2, W3, and W4 are the corresponding weighting coefficients.

[0017] The advantages and positive effects of this invention are: This invention precisely addresses the problems of existing data acquisition methods being singular and comparison methods being simplistic. It achieves three-dimensional deep fusion and comparative analysis of multi-source, multi-dimensional data, effectively eliminating blind spots in hazard identification, reducing false alarms and false negatives, and significantly improving the accuracy and comprehensiveness of hazard identification. Simultaneously, this core method overcomes the shortcoming of existing technologies lacking three-dimensional spatial data fusion capabilities. Relying on advanced three-dimensional data fusion algorithms to acquire three-dimensional spatial information of the work scenario, it achieves accurate identification of spatial hazard sources, filling blind spots in spatial hazard protection. Addressing the issues of outdated and poorly adaptable existing data fusion and comparison algorithms, this invention designs an adaptive three-dimensional data fusion and comparison algorithm. This effectively solves the problems of multi-source data redundancy and conflict, improves the accuracy of fused data, and achieves dynamic comparison by combining historical and real-time data, enhancing the algorithm's robustness and adaptability to different high-risk work scenarios. Detailed Implementation

[0018] A hazard identification system based on a three-dimensional data fusion and comparison method, installed on a safety helmet, includes a data acquisition module, a main control module, a power supply module, a display module, and an alarm module. The power supply module connects to all three modules, and the main control module also connects to them. The data acquisition module collects multi-source, multi-dimensional data. The main control module performs preprocessing, three-dimensional data fusion, multi-dimensional comparative analysis, and hazard identification on the collected data. The power supply module powers all modules. The display module uses a miniature OLED screen. The alarm module includes a voice broadcast unit and a vibration unit. The speakers of the voice broadcast unit are embedded on both sides of the helmet shell, and the vibration unit is installed inside the helmet liner. The main control module adopts an improved Kalman filter algorithm combined with the object-centric fusion concept to achieve three-dimensional deep fusion of multi-source data. It constructs a multi-dimensional comparison system of three-dimensional fusion dataset + preset safety benchmark dataset + historical normal dataset + real-time three-dimensional image to achieve accurate identification of multi-dimensional hazard sources.

[0019] A hazard identification method based on a three-dimensional data fusion and comparison system includes the following steps: Step 1: The data acquisition module collects multi-source data. The operator puts on a safety helmet and starts the system. The power supply module supplies power to each module, and the data acquisition module starts synchronously to collect multi-source data (3D position and attitude data, environmental data, 3D image data, human body status data, and impact data). The acquisition frequency is 1 time / second (the acquisition frequency of 3D image data can be adjusted according to the working mode).

[0020] Step 2: The main control module performs noise reduction, calibration, normalization, and anomaly removal on the collected multi-source data to generate a preprocessed dataset.

[0021] The data preprocessing unit receives data collected by each submodule, performs noise reduction processing using a moving average filtering algorithm to eliminate data noise; calibrates the data of each sensor using preset calibration parameters to correct sensor errors; normalizes data of different magnitudes to convert them into a dataset of uniform magnitude; and uses the 3σ criterion to remove outlier data, retains valid data, and generates a preprocessed dataset.

[0022] Step 3: The main control module calls the improved Kalman filter combined with the zero-rate update algorithm to fuse and correct the 3D position and attitude data; based on the object center fusion idea, the corrected position and attitude data are spatially aligned and feature-correlated with other data in a unified 3D world coordinate system to generate a 3D fusion dataset.

[0023] Step 3.1: Fusion of 3D position and attitude data based on EKF-ZUPT.

[0024] To address the core issue of integrated cumulative errors in position and attitude data acquired by IMUs (Inertial Measurement Units), which negatively impact positioning accuracy, this module precisely triggers the ZUPT (Zero-Range Update) algorithm during the gait support phase for error correction. Simultaneously, it combines this with the EKF (Improved Kalman Filter) algorithm to achieve data fusion, effectively offsetting integral drift and improving the stability and accuracy of position and attitude data. The core implementation process and formulas are as follows: Gait support phase determination (ZUPT trigger condition): Introducing foot pressure sensors to work in conjunction with preset pressure thresholds. When the sensor detects values ​​in real time satisfy When the current gait support phase is determined, the ZUPT algorithm is immediately triggered to correct the IMU integration error in real time; when When the period is in a swing phase, only EKF state prediction is performed, and zero-rate update is not triggered.

[0025] Its state prediction equation is: in, It is a state vector, specifically containing three-dimensional position. 3D velocity 3D attitude angle (Roll angle, pitch angle, yaw angle), a total of 9 dimensions; for The state transition matrix describes the evolution of the state vector over time. for The control matrix is ​​used to map the effect of control inputs on the state; Input quantities for IMU measurement, including triaxial acceleration. and triaxial angular velocity .

[0026] Observation update equation (when ZUPT is triggered): Among them, when ZUPT is triggered, the zero-velocity observation (That is, the theoretical value of foot speed is 0); for The observation matrix is ​​used to extract the velocity components in the state vector and match them with the observed values. The Kalman gain is obtained by minimizing the covariance of the estimation error, and its calculation formula is as follows: ,in For the prediction error covariance matrix, The noise covariance matrix is ​​observed (preset based on the IMU sensor accuracy).

[0027] Error correction logic: After ZUPT is triggered, the velocity error generated by IMU integration is forcibly corrected through the observation update equation, and then the position and attitude angle errors are corrected in reverse to break the "error accumulation" closed loop; after the support period ends, the normal prediction-update process of EKF is returned to ensure the continuity of data in the non-support period.

[0028] Step 3.2: Multi-source data fusion based on object center.

[0029] Using the core objects identified in 3D image data (workers, equipment, falling objects, etc.) as the fusion core, a unified 3D world coordinate system is established to break down the coordinate barriers between different sensor data. Environmental perception data, human physiological data, impact perception data, and object features are precisely aligned to achieve integrated fusion of multi-dimensional data. The specific implementation steps and formulas are as follows: Establishing a unified coordinate system: A three-dimensional world coordinate system is established with the geometric center of the core object as the origin. This coordinate system is fixed to the object itself and updates synchronously with the object's movement, ensuring spatial consistency of all related data.

[0030] Coordinate transformation (unifying multi-sensor data): This transforms the raw data collected by each sensor from its local coordinate system to a unified world coordinate system. The transformation formula is as follows: .

[0031] in, These are the original coordinates in the local coordinate system of the sensor (such as the installation coordinates of a gas sensor or the detection coordinates of an impact sensor). for The rotation matrix describes the attitude mapping relationship between the local coordinate system and the world coordinate system, and is calculated from the attitude angles acquired by the IMU; for Translation vector, describing the positional offset between the origin of the local coordinate system and the origin of the world coordinate system; To achieve spatial alignment of data from different sensors in the transformed world coordinate system.

[0032] Feature association and dataset generation: Feature extraction: Extracting the three-dimensional feature parameters of each object, including geometric features (volume V, surface area S, center coordinates). Motion characteristics (speed) Trajectory of motion ).

[0033] Data correlation: Integrating environmental data (toxic gas concentration C, temperature and humidity T / H), human body status data (heart rate HR, blood oxygen SpO2), and impact data (impact force F, impact angle). The data is associated with the three-dimensional spatial coordinates and feature parameters of the corresponding object, and the object and attribute to which the data belongs are labeled (e.g., “Worker A - Heart rate HR=110 beats / min”, “Falling object B - Impact force F=50N”).

[0034] Fusion Dataset Output: Integrates all related data to generate a four-dimensional / three-dimensional fusion dataset containing "spatial information, environmental information, human information, and impact information". This provides standardized data input for subsequent multi-dimensional comparative analysis.

[0035] Step 4: Based on the aforementioned 3D fusion dataset, preset safety benchmark dataset, and historical normal dataset, the main control module performs multi-dimensional comparative analysis to identify hazards and determine their levels. The multi-dimensional comparative analysis unit calls the two basic datasets (preset safety benchmark dataset and historical normal dataset), using the 3D fusion dataset generated in Step 3 as the analysis object. Through a four-level process of "basic threshold comparison - trend judgment - spatial identification - linkage comparison," it completes single-indicator anomaly identification, data trend prediction, spatial hazard identification, and composite hazard identification. The specific implementation process is as follows: Step 4.1: Basic threshold comparison (single indicator anomaly identification).

[0036] By employing dual comparisons (safety benchmark comparison and historical normal comparison), abnormal situations in single data indicators are identified, and interference from accidental fluctuations is eliminated. The core formula and judgment logic include: comparison with a preset safety benchmark dataset and comparison with a historical normal dataset.

[0037] (1) Comparison with the preset security benchmark dataset: Calculate the relative deviation between real-time data and the safety baseline: in, This represents the real-time value of the i-th indicator in the 3D fusion dataset (such as toxic gas concentration, heart rate, impact force, etc.). The safety thresholds for the corresponding indicators in the preset safety benchmark dataset are set according to industry standards and work scenarios, such as safety thresholds for toxic gas concentrations. Heart rate safety threshold (times / minute) The relative deviation reflects the degree to which real-time data deviates from the safety baseline.

[0038] Judgment rule: Preset deviation threshold (Generally taken as 0.2, but can be adjusted according to the scenario), when When this occurs, the indicator is determined to be "abnormal to the safety baseline," meaning it exceeds the safety range.

[0039] (2) Comparison with historical normal datasets: Calculate the deviation between real-time data and historical normal data: in, This represents the average of the corresponding indicators in the historical normal dataset (average data from the normal operating period over the past 30 days). The variance of the indicator corresponding to the historical normal dataset reflects the fluctuation range of the historical data; The deviation is the difference between real-time data and historical normal levels.

[0040] Judgment rule: Preset deviation threshold (Generally, 2 is taken, corresponding to the 95% confidence interval), when When this occurs, the indicator is determined to be "historical deviation anomaly," meaning it deviates from the fluctuation range under normal operating conditions.

[0041] Final determination of single indicator anomaly: When an indicator meets both the "safety benchmark anomaly" and "historical deviation anomaly" criteria, it is determined to be a single indicator anomaly; when only one indicator is anomaly, it is determined to be "suspicious fluctuation", and no warning will be triggered for the time being, while subsequent data will be continuously tracked.

[0042] Step 4.2: Determine data change trends (predict anomalies in advance) By employing the sliding window linear regression method, trend fitting is performed on real-time data sequences to enable early prediction of data change trends, thus avoiding the problem of delayed early warning caused by relying solely on threshold comparison.

[0043] Sliding window settings: Set the sliding window size to N (preset based on the data acquisition frequency, generally 10 acquisition cycles, with an acquisition cycle of 0.1s, meaning the window covers data within 1 second). The window slides in real time as data acquisition progresses, always extracting the real-time data sequence of the latest N acquisition cycles. ,in This represents the real-time data value during the i-th acquisition cycle.

[0044] Trend slope calculation (linear regression fitting): By fitting the changing trend of the data sequence using linear regression, the slope k of the trend is calculated as follows: Where i is the index of the data in the window (1~N), the sign of the slope k reflects the trend direction (k>0 is an upward trend, k<0 is a downward trend), and the absolute value of k reflects the rate of trend change (the larger the absolute value, the faster the change).

[0045] Trend Anomaly Detection: Preset trend threshold (Upward trend anomaly threshold) and (Abnormal threshold for downward trend), adjusted according to different indicator characteristics (e.g., gas concentration) blood oxygen ).

[0046] Judgment rule: If If the data shows a rapid upward trend (e.g., a sudden increase in toxic gas concentration, or a rapid increase in impact force); If the data shows a rapid downward trend (such as a sudden drop in blood oxygen or an abnormal drop in heart rate), an early warning will be triggered immediately when the trend is abnormal, allowing workers time to respond to emergencies.

[0047] Step 4.3: Identification of Spatial Dimension Object Characteristics and Potential Hazard Sources Based on the 3D image data and spatial coordinate information in the 3D fusion dataset, a two-step method of "point cloud segmentation + target detection" is adopted to identify spatial object features and potential hazards (such as falling objects from heights and illegally placed equipment). Combined with distance calculation and trajectory prediction, the spatial hazard level is determined.

[0048] Point cloud preprocessing and segmentation: Point cloud preprocessing (denoising and downsampling) is performed on the 3D image data to remove invalid point clouds caused by environmental noise (such as dust and light interference); Euclidean clustering algorithm is used to determine the point cloud size based on the Euclidean distance between point clouds. The point cloud is divided into different point cloud clusters, and each point cloud cluster corresponds to a spatial object (workers, equipment, falling objects, etc.).

[0049] Extract the 3D feature parameters of each point cloud cluster: volume Surface area Center coordinates Forming an object feature set .

[0050] Target feature matching (hazardous object identification): Call the preset dangerous object feature library (Including standard characteristics of dangerous objects such as falling objects and unauthorized equipment, such as the volume range and surface area threshold of falling objects), calculate the similarity between the characteristics of the object to be identified and the characteristics of the dangerous object: in, Let j be the j-th dimension feature of the object to be identified. M represents the j-th dimension feature of the corresponding object in the hazardous object feature library, where M is the feature dimension (here M=3, i.e., volume, surface area, and center coordinates). This represents the feature similarity (ranging from 0 to 1). The higher the value, the higher the matching degree.

[0051] Judgment rule: Preset similarity threshold (Generally taken as 0.7), when When this happens, the object is determined to be a "dangerous object".

[0052] Space Hazard Assessment (Collision Risk Prediction): Taking the workers as the core protected object, calculate the three-dimensional relative distance between the hazardous object and the workers: in The coordinates of the center of the hazardous object, The coordinates are the center coordinates of the operator.

[0053] Fitting the motion trajectory of a dangerous object using continuous frame point cloud data. Combined with the relative distance L, the collision probability is predicted. The determination formula is: in, Preset a safe distance (adjusted according to the type of hazardous object, such as falling objects from heights). Illegal equipment ).

[0054] Judgment rule: When When the location is identified as a "potential source of danger in the spatial dimension", a spatial early warning is triggered.

[0055] Step 4.4: Multi-dimensional linkage comparison (identification of composite hazard sources).

[0056] A multi-dimensional linkage judgment matrix is ​​established, integrating the analysis results from environmental, human condition, and spatial dimensions. This breaks through the limitations of single-dimensional identification and accurately identifies complex hazard sources in complex scenarios (such as "toxic gas leak + personnel physiological abnormalities + space restriction"). The specific implementation steps are as follows: Determine the linkage dimensions and core indicators: Select three core linkage dimensions, and for each dimension, select key indicators to ensure coverage of the entire "environment-human-space" scenario: Environmental dimension: Toxic gas concentration C (core indicator); Human body condition dimension: Heart rate (HR) (core indicator); Spatial dimensions: relative distance L, collision probability (Core Indicators).

[0057] Establish linkage judgment rules (matrix-based judgment): Based on the abnormal combination of indicators of different dimensions, set graded judgment rules to clarify the risk level and hazard type of composite hazard sources. The specific rules are shown in the table below: Output comparative analysis results: Based on the linkage judgment rules, the final output is a standardized comparative analysis result of "risk level - hazard type - impact range", where the impact range is determined by the center coordinates of the object. The determination of the relative distance L provides a clear triggering basis for the subsequent hazard identification and alarm modules.

[0058] Step 5: The alarm module triggers the corresponding alarm mode based on the identified hazard source and hazard level.

[0059] Based on the comparative analysis results, the hazard identification unit determines the hazard source and hazard level according to the preset hazard level judgment logic; it triggers the corresponding alarm mode of the alarm module according to the hazard level, and displays the hazard information through the display module; the communication unit transmits the hazard identification results and 3D fusion dataset to the background management system and handheld terminal in real time to achieve synchronous early warning.

[0060] The specific logic for determining the hazard level is as follows: Four core determination indicators are selected, and quantitative values ​​are assigned based on multi-dimensional comparative analysis results. Indicator 1: Degree of deviation of a single parameter, with the value being the relative deviation Di, where Di ∈ [0, +∞).

[0061] Indicator 2: Historical data deviation degree, with the value of deviation degree Oi, where Oi ∈ [0, +∞).

[0062] Indicator 3: Space collision probability, with the value of collision probability Pc, where Pc ∈ [0, 1].

[0063] Indicator 4: Composite hazard correlation degree, with the value of Rc, where Rc ∈ [0, 1], which is output by a multi-dimensional linkage judgment matrix. When there is no composite hazard, Rc = 0; when there is a triple linkage hazard of environment - human - space, Rc = 1.

[0064] The index weight distribution is set according to the influence degree of different indicators on operation safety, and the weight coefficients are as follows: The weight of space collision probability W1 = 0.4; the weight of composite hazard correlation degree W2 = 0.3; the weight of single parameter deviation degree W3 = 0.15; the weight of historical data deviation degree W4 = 0.15.

[0065] Calculate the comprehensive risk value Sr: Sr = W1 × Pc + W2 × Rc + W3 × min(Di, 1) + W4 × min(Oi, 1) Perform upper limit truncation processing on Di and Oi. When the value is greater than 1, calculate it as 1.

[0066] Hazard level classification and judgment rules: Based on the comprehensive risk value Sr, the hazard sources are divided into 4 levels, and the corresponding judgment rules are as follows: Hazard levelComprehensive risk value rangeJudgment conditionsExplanationLevel I (General risk)0 < Sr ≤ 0.2Single indicator slightly exceeds the standard, no space collision risk, no composite hazard, and the data trend is stableLevel II (Relatively large risk)0.2 < Sr ≤ 0.5Multiple indicators exceed the standard or the data trend is abnormal, the space collision probability Pc < 0.3, and there is a single-dimensional hazardLevel III (Major risk)0.5 < Sr ≤ 0.8The composite hazard correlation degree Rc ≥ 0.5, the space collision probability 0.3 ≤ Pc < 0.7, and the physiological indicators of personnel or environmental indicators significantly exceed the standardLevel IV (Extraordinarily major risk)Sr > 0.8The composite hazard correlation degree Rc = 1, the space collision probability Pc ≥ 0.7, and the physiological indicators of personnel are severely abnormal or there is about to be a collision of falling objects from a height Output of level judgment result: According to the calculation result of the comprehensive risk value, the hazard identification unit outputs the judgment result of hazard source level + hazard type + influence range, for example: "Level III major risk - toxic gas leakage + abnormal heart rate of personnel - within a 5-meter operation radius".

[0067] Step 6: The main control unit stores the preprocessed data, 3D fusion dataset, comparative analysis results, and hazard identification records in real time; at the same time, it regularly updates the historical normal dataset (automatically updated every morning at midnight, retaining data for the past 30 days). According to the needs of different work scenarios, the preset safety benchmark dataset can be updated through the background management system to improve the adaptability of hazard identification.

[0068] It should be emphasized that the embodiments described in this invention are illustrative rather than limiting. Therefore, this invention includes, but is not limited to, the embodiments described in the specific implementation. Any other implementations derived by those skilled in the art based on the technical solutions of this invention are also within the scope of protection of this invention.

Claims

1. A hazard identification system based on a three-dimensional data fusion and comparison method, installed on a safety helmet, characterized in that: It includes a data acquisition module, a main control module, a power supply module, a display module, and an alarm module. The power supply module is connected to the data acquisition module, the main control module, the display module, and the alarm module. The main control module is connected to the data acquisition module, the display module, and the alarm module. The data acquisition module is used to collect multi-source, multi-dimensional data. The main control module is used to preprocess the collected data, perform three-dimensional data fusion, multi-dimensional comparative analysis, and hazard identification. The power supply module is used to supply power to each module. The display module uses a miniature OLED screen. The alarm module includes a voice broadcast unit and a vibration unit. The speakers of the voice broadcast unit are embedded on both sides of the helmet shell, and the vibration unit is installed inside the helmet liner.

2. The hazard identification system based on a three-dimensional data fusion and comparison method according to claim 1, characterized in that: The main control module adopts an improved Kalman filter algorithm combined with the object-centric fusion concept to achieve three-dimensional deep fusion of multi-source data, and constructs a multi-dimensional comparison system of three-dimensional fusion dataset + preset safety benchmark dataset + historical normal dataset + real-time three-dimensional image to achieve accurate identification of multi-dimensional hazard sources.

3. A hazard identification method for a hazard identification system based on three-dimensional data fusion and comparison as described in claim 1 or 2, characterized in that: Includes the following steps: Step 1: The data acquisition module collects data from multiple sources; Step 2: The main control module performs noise reduction, calibration, normalization, and anomaly removal on the collected multi-source data to generate a preprocessed dataset. Step 3: The main control module calls the improved Kalman filter combined with the zero-rate update algorithm to fuse and correct the 3D position and attitude data; Based on the idea of ​​object-centered fusion, the corrected position and pose data are spatially aligned and feature-associated with other data in a unified three-dimensional world coordinate system to generate a three-dimensional fusion dataset. Step 4: Based on the three-dimensional fusion dataset, the preset safety benchmark dataset, and the historical normal dataset, the main control module performs multi-dimensional comparative analysis to identify hazard sources and determine the hazard level. Step 5: The alarm module triggers the corresponding alarm mode based on the identified hazard source and hazard level.

4. The identification method of a hazard identification system based on three-dimensional data fusion and comparison method according to claim 3, characterized in that: The multi-source data in step 1 includes three-dimensional position and attitude data, environmental data, three-dimensional image data, human body state data, and impact data.

5. The identification method of a hazard identification system based on three-dimensional data fusion and comparison method according to claim 3, characterized in that: The specific implementation method of step 2 is as follows: the data preprocessing unit receives the data collected by each data acquisition module, performs noise reduction processing using a moving average filtering algorithm to eliminate data noise; calibrates the data of each sensor using preset calibration parameters to correct sensor errors; normalizes data of different magnitudes to convert them into a dataset of the same magnitude; uses the 3σ criterion to remove abnormal data, retains valid data, and generates a preprocessed data set.

6. The identification method of a hazard identification system based on three-dimensional data fusion and comparison method according to claim 3, characterized in that: Step 3, which involves calling an improved Kalman filter combined with a zero-rate update algorithm to fuse and correct the 3D position and attitude data, includes the following steps: Step 3.1: Obtain the real-time detection value of the foot pressure sensor; Step 3.2: When the real-time detected value is greater than the preset pressure threshold, it is determined to be the gait support period, and the zero-speed update algorithm is triggered. The integral error of the inertial measurement unit is corrected through the observation update equation, wherein the zero-speed observation value is set to zero. Step 3.3: When the real-time detection value is less than or equal to the preset pressure threshold, it is determined to be an oscillation period, and state prediction of the improved Kalman filter is performed.

7. The identification method of a hazard identification system based on three-dimensional data fusion and comparison method according to claim 3, characterized in that: Step 3, based on the object-centric fusion concept, involves the following steps to form a 3D fused dataset: Step 3.4: Establish the unified three-dimensional world coordinate system with the geometric center of the identified core object as the origin; Step 3.5: Transform the raw data collected by each sensor from its respective local coordinate system to the unified three-dimensional world coordinate system. The coordinate transformation formula is as follows: ； in, These are the original coordinates in the sensor's local coordinate system; for The rotation matrix describes the attitude mapping relationship between the local coordinate system and the world coordinate system, and is calculated from the attitude angles acquired by the IMU; for Translation vector, describing the positional offset between the origin of the local coordinate system and the origin of the world coordinate system; To achieve spatial alignment of data from different sensors in the transformed world coordinate system; Step 3.6: Extract the three-dimensional feature parameters of each object, and associate the environmental data, human body state data, and impact data with the three-dimensional spatial coordinates and feature parameters of the corresponding objects to generate the three-dimensional fusion dataset containing spatial information, environmental information, human body information, and impact information.

8. The identification method of a hazard identification system based on three-dimensional data fusion and comparison method according to claim 3, characterized in that: The specific implementation method of step 4 is as follows: Step 4.1: Calculate the relative deviation between real-time data and preset safety benchmark values, as well as the deviation between real-time data and historical normal data, to identify single indicator anomalies; Step 4.2: Calculate the trend slope of the real-time data sequence using the sliding window linear regression method to identify rapid upward or downward trends in the data; Step 4.3: Perform point cloud segmentation and target detection on the 3D image data, identify dangerous objects, and calculate the relative distance and collision probability between dangerous objects and workers; Step 4.4: Establish a linkage judgment matrix, integrate the analysis results of environment, human body status and spatial dimensions, and identify complex hazard sources; Step 4.5: Calculate the comprehensive risk value based on the preset weighting coefficients, and determine the hazard level based on the comprehensive risk value.

9. The identification method of a hazard identification system based on a three-dimensional data fusion and comparison method according to claim 8, characterized in that: The specific calculation method for the comprehensive risk value is as follows: Sr=W1×Pc+W2×Rc+W3×min(Di,1)+W4×min(Oi,1); Where Pc is the spatial collision probability, Rc is the composite hazard correlation degree, Di is the relative deviation, Oi is the deviation degree, and W1, W2, W3, and W4 are the corresponding weighting coefficients.

Images