A safety hidden danger point intelligent identification positioning and closed-loop disposal method and system for complex scenes

Abstract

The application discloses a kind of safety hidden danger point intelligent identification positioning and closed loop treatment method and system for complex scene, collect sensor node event data, distinguish fixed node and mobile terminal according to event source, determine observation source coordinates and coarse positioning range;According to observation characteristics in coarse positioning range, form hidden danger existence credibility field, generate candidate hidden danger instance by connected domain clustering;Correct multi-source observation time stamp and associate observation data, solve terminal pose and hidden danger coordinates in combination with ranging, attitude and relative observation constraint, generate continuous space envelope after judging observability;According to hidden danger credibility and risk attribute, generate control instruction;Verify treatment effect through five marks of identity, location, action, effect and evidence integrity, only when check passes, cancel and update model.The application can effectively solve the problem that safety hidden danger point identification positioning is not accurate and treatment closed loop is imperfect in complex scene, improve the intelligent level of industrial safety management.

Description

Technical Field

[0001] This invention belongs to the fields of industrial safety sensing, Internet of Things communication, intelligent inspection, multi-source observation correlation, spatial positioning, uncertainty propagation and equipment linkage control technology, specifically involving a method and system for intelligent identification, location and closed-loop handling of safety hazard points. Background Technology

[0002] In industrial parks and their storage areas, energy facility areas, construction areas, and underground spaces, safety hazards may manifest as abnormal environmental parameters, abnormal equipment operation, structural changes, abnormal stacking of goods, abnormal heat, abnormal gas or dust, and obscured suspected hazardous targets. These hazards typically do not carry wireless positioning tags and are characterized by their diverse types, non-fixed locations, susceptibility to signal interference, and uncertain discovery time. The safety hazard points referred to in this article include not only point targets but also representative coordinates and spatial envelopes of area or volumetric hazard regions; representative coordinates are used for location and scheduling, while the spatial envelope is used to determine the re-measurement range and control range.

[0003] In the prior art, Chinese patent application CN118149802A discloses a multi-source fusion Kalman filtering method and system for unmanned vehicles with observation time delay. This technology uses a strapdown inertial navigation system as a time and space reference, and performs spatiotemporal synchronization of visual inertial odometry and global navigation satellite system to construct SINS / GNSS federated sub-filters and SINS / VIO federated sub-filters with time delay. This technology is mainly applied in the field of unmanned vehicle navigation, using a combination of SINS, GNSS, and VIO sensors, and employing a federated Kalman filtering algorithm to solve the observation time delay problem. However, this technology only involves unmanned vehicle navigation and positioning, and does not address the identification and handling of industrial safety hazards, nor does it consider issues such as the diversion of safety hazard sources, observability determination, and closed-loop verification.

[0004] Furthermore, existing technologies typically rely on periodic manual inspections, single sensor threshold alarms, or direct wireless positioning of alarm terminals. Directly using the coordinates of alarm terminals obtained through wireless positioning as hazard coordinates ignores the observation distance, direction, attitude, and sensor installation offset between the terminal and the hazard. Multiple data sources arrive at different times, and multiple similar hazards may exist simultaneously in the same scene. Without time error propagation and instance association, observations belonging to different hazards can easily be incorrectly placed into the same optimization problem. Simply relying on manual confirmation or status field judgment for item cancellation cannot prove the actual action of the equipment, the effectiveness of the handling, or the completeness of the evidence. Existing fingerprint positioning technologies mainly use short-range communication technologies such as Wi-Fi and Bluetooth; LoRa is typically used for long-range low-power communication and is not used for fingerprint positioning. Existing multi-source fusion positioning technologies typically assume perfect time synchronization and do not consider time error propagation and uncertainty propagation. Existing positioning systems typically assume the system is always observable and do not consider observability determination. Existing closed-loop verification typically only verifies the execution result and does not consider multi-dimensional verification of time, location, action, and status.

[0005] Therefore, this method addresses four questions sequentially: where does the data come from, how many potential hazards are there on-site, where is each hazard located, and whether the remediation has been truly completed. The system first separates the location of the observation equipment from the location of the hazard, then treats unconnected suspected areas as separate candidate hazard areas. Next, it aligns the observation times of different equipment, incorporates the position and attitude errors caused by time deviations into the covariance, and only associates observations that truly belong to the same candidate hazard. After location, the system checks whether the existing data can stably determine the location and generates a continuous area covering the possible location and safety margin. After remediation, the system determines whether the item can be closed based on equipment feedback, retest results, and signature verification. Summary of the Invention

[0006] This invention provides a method and system for intelligent identification, location, and closed-loop handling of safety hazard points in complex scenarios. The method first distinguishes between the observation source and the target hazard, then processes multiple suspected hazards separately; subsequently, it aligns observation times, filters data belonging to the same hazard, and combines this with sensor installation locations for joint location; only when existing observations are sufficient to stably determine the location is the result output. Finally, the system expands the possible locations to a continuous control range and uses five computable pieces of evidence to determine whether the handling is complete, thereby avoiding coordinate confusion, observation mismatch, unstable location results, and relying solely on manual status verification.

[0007] To achieve the above objectives, the first aspect of the present invention provides a method for intelligent identification, location, and closed-loop handling of security hazard points in complex scenarios, comprising the following steps:

[0008] S1, data acquisition, event identification, and low-power transmission;

[0009] S2. Event source diversion and coarse regional localization;

[0010] S3, Generation of Hidden Danger Credibility Field and Candidate Instances;

[0011] S4, time alignment, data attribution determination, and precise positioning;

[0012] S5. Risk Classification and Technical Response;

[0013] S6. Feedback verification and model update.

[0014] Preferably, step S1 specifically involves: collecting environmental, equipment, image, posture, or inspection event data from fixed sensor nodes or mobile inspection terminals, combining scene, area, time window, terminal, hazard category, or status fields into an event identifier, and uploading it to the edge processing node via a low-power wireless communication network.

[0015] Preferably, the low-power wireless communication network is a LoRa network; during the deployment phase, the calibration terminal sequentially sends multiple frames of data packets in each spatial grid, records the received signal strength, signal-to-noise ratio, data packet reception rate, and arrival time difference under clock synchronization conditions according to the gateway number, and saves the mean and standard deviation of each component; during online detection, the mobile inspection terminal sends data packets in the same way, and each gateway forms a real-time observation vector in a fixed order; for each effective component, the difference between the real-time value and the grid mean is calculated, and the standard deviation of the component is used for normalization. Then, the squares of the normalized differences are averaged and the square root is taken to obtain the distance of each grid; the first L grids and their adjacent grids are selected as the coarse positioning range according to the distance from smallest to largest. If the number of effective components is insufficient, a retransmission is requested.

[0016] Preferably, step S2 specifically involves: determining the observation source coordinates and coarse positioning range based on the event source; when the event originates from a fixed sensing node, using the valid installation coordinates in the device registry as the observation source coordinates, and forming a coarse positioning area based on the effective detection radius and field of view direction; when the event originates from a mobile inspection terminal, comparing the real-time signal characteristics formed by multiple wireless receiving nodes within the same event time window with the area fingerprint to determine the terminal's coarse positioning range.

[0017] Preferably, step S3 specifically involves: within the coarse positioning area or range, forming a hazard existence confidence field based on the uncertainty of the observation position, observation direction, effective field of view, occlusion visibility, and category feature matching results; obtaining a candidate location set through multi-frame thresholding and connected component clustering; registering different connected components as candidate hazard instances and calculating the initial representative coordinates of each instance.

[0018] Preferably, step S3 calculates the spatial response, category feature matching score, and hazard existence confidence level sequentially according to the following three formulas:

[0019]

[0020] in: Indicates spatial response, The location to be determined. Let i be the coordinates of the i-th observation point. For spatial attenuation scale, For visibility, For an effective field of view, Only in the effective field Non-zero inside; This represents the matching score of the i-th observation point for location x belonging to the c-th type of hazard. To standardize the feature vector, This is a reference template for class c; This indicates the confidence level of location x belonging to the c-th type of hazard; n represents the number of valid observation points. To measure the confidence level weight; Measure how consistent the representation of source i is with that of category c at x; Output only when the sum of the weights involved in the calculation is greater than 0. .

[0021] Preferably, step S3 performs threshold comparison on each grid at the most recent K detection times, retains grids that reach the threshold at least q times; performs two-dimensional eight-neighbor or three-dimensional twenty-six-neighbor connected component division on the retained grids and deletes connected components with too small area or volume, registers each remaining connected component as an independent candidate instance; calculates the weighted centroid of each instance using the average confidence of the most recent K frames as the weight, and uses the geometric center of the connected component when the weight sum is 0.

[0022] Preferably, step S4 specifically involves: correcting the timestamps of multi-source observations, using terminal linear velocity and angular velocity to compensate asynchronous observations within the allowed time difference to a unified reference time, propagating clock correction error, linear velocity error and angular velocity error to the effective observation covariance, and assigning observations that have reached the association threshold to corresponding candidate instances based on the corrected time difference, spatial distance, category feature similarity and event identifier consistency.

[0023] By combining ranging constraints, attitude constraints, projection of the hazard relative to the terminal, orientation, distance or amplitude constraints, and sensor installation extrinsic parameters, the terminal position, terminal attitude, hazard representative coordinates and position covariance are jointly solved for each candidate instance.

[0024] After removing the numerical regularization term, the rank and condition number of the joint information matrix are judged. The positioning result is output only when the observability threshold is met. The positioning information ellipsoid and the candidate instance extended region are closed, and the connected region containing the coordinates of the hidden danger is extracted as the continuous spatial envelope.

[0025] Preferably, in step S4, temporal proximity, spatial proximity, category feature similarity, and event identifier consistency are all converted into scores ranging from 0 to 1, and the association score between the i-th observation and the r-th candidate instance is calculated according to the following formula:

[0026]

[0027] in: For correlation scoring, , , and The scores are, in order, time, space, category, and event identifier. , , and The weights are non-negative and sum to 1; only the maximum score not lower than the association threshold is considered. The observations are assigned to the corresponding instances, and the remaining observations enter the supplementary collection queue.

[0028] Preferably, in step S4, the time is first corrected according to the following formula:

[0029]

[0030] in: This is the original timestamp. This is the clock correction amount. To correct the timestamp, For reference time, The propagation time difference is positive or negative; when | If the observation exceeds the calibration limit, discard the observation; otherwise, use the terminal linear velocity and angular velocity to convert the observation to the reference time, and add the additional covariance formed by the clock, linear velocity and angular velocity errors to the original observation covariance.

[0031] Preferably, the ranging and positioning unit is a UWB positioning unit, the attitude observation unit is an inertial measurement unit, and the relative observation sensor includes a visual sensor or a millimeter-wave sensor; the UWB ranging model is...

[0032] in: For the first Distance values ​​at each anchor point For the location of the inspection terminal, For anchor point coordinates, For ranging error; visual or millimeter-wave sensors transform relative observations to the global coordinate system through calibrated rotation and translation extrinsic parameters, and then weight and sum the ranging, attitude, visual and millimeter-wave residuals according to their respective covariance and confidence level, and jointly solve for the terminal position, terminal attitude and hazard representative coordinates.

[0033] Preferably, in step S4, after removing the numerical regularization term, the information matrix is ​​checked to see if it is full rank and if the condition number does not exceed the calibration upper limit, and the position uncertainty is calculated according to the following formula:

[0034]

[0035] in: For the covariance of the location of the hidden danger, The positional uncertainty; only when the observability condition is satisfied and If the position does not exceed the allowed upper limit, output the positioning result. Then, perform expansion and closure operations on the positioning information region, candidate instance region, and safety distance to obtain a connected region containing optimized coordinates that does not merge with adjacent instances as a continuous spatial envelope.

[0036] Preferably, step S5 specifically involves generating a collection control command or disposal control command with an event number, instruction sequence number, one-time random number, target space envelope, control parameters, and validity period based on the credibility of the hidden danger, the uncertainty of the location, and the risk attributes.

[0037] Preferably, step S5 calculates the risk score according to the following formula:

[0038]

[0039] in: To normalize the position uncertainty, S represents the maximum allowable limit; S represents the overall risk score. To maximize the credibility of the potential risks, As for the severity of the consequences, To the degree of obstruction, To improve the ability to handle [the situation], , , and For non-negative weights that sum to 1, It is a non-negative processing capacity deduction coefficient; when the location uncertainty or occlusion degree is too high, a supplementary sampling instruction is issued; when S reaches the linkage threshold, a processing instruction with event number, sequence number, one-time random number, target space envelope, control parameters and validity period is issued.

[0040] Preferably, step S6 specifically involves: receiving device feedback and retest data, calculating five binary flags for identity, location, action, effect, and evidence integrity based on the consistency of identity digests before and after processing, whether the evidence coordinates after processing are within the spatial envelope, the deviation between the device feedback telemetry and control vector, the changes in technical indicators before and after processing, and the anti-replay signature verification results. The flag is deactivated only when all five flags are 1, and the observation weight, category reference template, and credibility judgment threshold are updated using the verification samples that have passed the verification and have no conflicts.

[0041] Preferably, step S6 calculates the closed-loop state of the j-th task according to the following formula:

[0042] in: , , , and The indicators are, in order: identity, location, action, effect, and evidence integrity, all of which are binary variables. The identity indicator only compares fields that do not change with the disposal, such as event number, instance number, device number, and instruction binding random number. The location indicator checks whether the evidence coordinates are within the spatial envelope. The action indicator compares the control quantity and the feedback quantity. The effect indicator compares the technical indicators before and after the disposal. The evidence indicator verifies that the serial number, random number, validity period, and digital signature were not used. An item is cancelled only when all five indicators are 1, and the observation weight, category template, and judgment threshold are updated only using samples with valid signatures, no conflicts after review, and the required number.

[0043] The second aspect of this invention provides an intelligent identification, location, and closed-loop handling system for safety hazard points, including a data acquisition and source discrimination module, a mobile terminal coarse positioning module, a fixed node observation source coordinate module, a hazard credibility field and instance decomposition module, a clock correction and error propagation module, a multi-source observation correlation module, a ranging and positioning and attitude observation module, a relative observation sensor module, an external parameter calibration and coordinate transformation module, a joint positioning and observability determination module, a continuous spatial envelope module, a risk and control command module, an equipment feedback and anti-replay verification module, and an evidence review and update module; the data acquisition and source discrimination module will... The node registration coordinates are only used as the observation source coordinates; the instance splitting module establishes different hidden danger state quantities for different connected domains; the error propagation module adds clock and motion uncertainties to the effective observation covariance; the multi-source observation association module only outputs observations that reach the association threshold to the corresponding instance; the joint positioning and observability determination module outputs the positioning result when the information matrix meets the threshold; the continuous spatial envelope module generates a connected control range containing optimized coordinates; the device feedback and anti-replay verification module outputs five closed-loop flags based on identity digest, envelope inclusion, action deviation, technical indicator improvement, and cryptographic verification results.

[0044] Compared with the prior art, the present invention has the following beneficial technical effects:

[0045] (1) In this invention, the registered coordinates of the fixed sensor nodes are only used as the location of the observation source, and the mobile inspection terminal obtains the coarse location of the area through communication fingerprints; both types of sources must be combined with direction, distance or other relative observations to obtain the coordinates of the hidden danger, avoiding the direct use of the coordinates of the sensor nodes or inspection terminals as the coordinates of the hidden danger. This design effectively solves the problem of confusion between the location of the observation equipment and the location of the hidden danger in the traditional method, and improves the positioning accuracy.

[0046] (2) In the LoRa implementation of this invention, the fingerprint distance uses component normalization and effective masking, which can still output a continuous coarse terminal positioning range when some gateways are missing or the arrival time difference is unavailable. This robust design ensures that the system can still work stably under complex electromagnetic environment and network conditions, and will not cause the entire system to fail due to the failure of individual sensors.

[0047] (3) In this invention, candidate connected components are split by instance, and multi-source observations are associated according to correction time, spatial distance, category similarity, and event identifier. Observations that do not reach the association threshold are placed in an isolation queue, which can reduce erroneous fusion when similar hidden dangers coexist. This mechanism effectively solves the data association problem in multi-target scenarios and avoids the positioning deviation caused by erroneous fusion of observations of different hidden dangers.

[0048] (4) In this invention, time compensation errors are incorporated into the effective observation covariance, and the joint positioning results must also pass the full rank of the information matrix and the condition number threshold. When a continuous spatial envelope cannot be formed or observability is insufficient, sensor turning, supplementary ranging, or remeasurement is automatically triggered. This adaptive quality control mechanism ensures the reliability and accuracy of the positioning results and avoids outputting unreliable positioning results when data is insufficient. Compared with the prior art that assumes perfect time synchronization and does not consider the propagation of time errors and uncertainty, this invention is the first to consider the propagation of time errors and uncertainty in multi-source fusion positioning.

[0049] (5) The identity digest, envelope inclusion relationship, instruction and telemetry deviation, change of handling indicators and anti-replay signature verification of the present invention determine five closed-loop flags respectively, avoiding the need to cancel items based solely on manual status fields; the model is only allowed to be updated after the verified sample meets the quantity and verification conditions. This strict technical evidence verification mechanism ensures the authenticity and traceability of the handling closed loop and improves the technical credibility of security management. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the overall implementation process of the present invention;

[0051] Figure 2 This is a schematic diagram illustrating the generation of the potential credibility field, candidate instances, and initial representative coordinates in this invention.

[0052] Figure 3This is a schematic diagram of the observation correlation, time error propagation, and joint positioning system of the present invention;

[0053] Figure 4 This is a schematic diagram illustrating the coordinate relationships between multiple instances of potential hazards, terminals, and multi-source observations in this invention.

[0054] Figure 5 This is a schematic diagram illustrating the verification of the five technical evidences and the closed-loop state transition for replay prevention in this invention. Detailed Implementation

[0055] The following description, with reference to the accompanying drawings, illustrates several preferred embodiments of the present invention to make its technical content clearer and easier to understand. The present invention can be embodied in many different forms, and the scope of protection of the present invention is not limited to the embodiments mentioned herein.

[0056] In the accompanying drawings, components with the same structure are indicated by the same numerical designation, and components with similar structures or functions are indicated by similar numerical designations. The dimensions and thicknesses of each component shown in the drawings are arbitrary, and the present invention does not limit the dimensions and thicknesses of each component. To make the illustrations clearer, the thickness of some components has been appropriately exaggerated in the drawings.

[0057] Example 1

[0058] like Figure 1 As shown, this embodiment provides a method for intelligent identification, location, and closed-loop handling of security hazard points in complex scenarios, including the following steps:

[0059] Data acquisition, event identification, and low-power transmission; event source diversion and coarse area localization.

[0060] Fixed sensor nodes and mobile inspection terminals collect environmental, equipment, image, attitude, and event data, and combine scene, region, time window, terminal, hazard category, and status fields into event identifiers. The registered coordinates of fixed nodes only indicate the location of the observation source; the mobile inspection terminal uses multiple LoRa gateways for area-level coarse positioning.

[0061] During deployment, the calibration terminal sequentially sends multiple frames of data packets to each spatial grid, and saves the received signal strength, signal-to-noise ratio, data packet reception rate, and the mean and standard deviation of available time-of-arrival differences according to the gateway number. During online detection, real-time observation vectors are formed in the same order, and the distance between the real-time vector and the fingerprint of the l-th grid is calculated.

[0062]

[0063] In the formula, The distance between the real-time observation and the l-th grid fingerprint is represented by d; the total number of feature components is represented by d; and the component index is represented by g. This represents the g-th component observed in real time. This represents the calibration mean of the component corresponding to the l-th grid. This represents the calibration standard deviation after minimum scale constraint; This is a valid mask; a value of 1 is used when a component is valid, and a value of 0 is used when a component is missing. Calculation is only performed when the number of valid components reaches a preset lower limit. Otherwise, a retransmission is requested; after successful calculation, the L grids with the smallest distance and their adjacent grids are selected as the coarse positioning range of the mobile inspection terminal.

[0064] The regional fingerprint database records the grid coordinate range, gateway number, transmit power, channel, acquisition time, mean and standard deviation of each component. When a gateway or antenna moves, the spatial layout changes significantly, or the continuous detection results of the reference terminal exceed the drift threshold, the affected grid is recalibrated.

[0065] In practice, the detection area first needs to be gridded. For example, a 100m x 100m detection area can be divided into 2500 grids of 2m x 2m each. At the center of each grid, a calibration terminal with the same hardware configuration as the online inspection terminal sends no fewer than 50 test data packets. Each gateway records the received signal strength, signal-to-noise ratio, data packet reception rate, and other parameters, and calculates the statistical characteristics of these parameters.

[0066] For received signal strength, assuming a gateway receives 50 frames of data at grid 12 with signal strengths of -71dBm, -70dBm, -72dBm, -69dBm, and -73dBm, the calculated mean is -71dBm and the standard deviation is 2dB. The signal-to-noise ratio (SNR) is similarly calculated using its mean and standard deviation. If clock synchronization is implemented between gateways, the time difference of arrival can also be calculated.

[0067] During the online detection phase, when the mobile inspection terminal sends data packets, each gateway performs statistical analysis using the same parameter standards. For example, gateway A receives a signal strength of -70dBm, gateway B receives -80dBm, and gateway C receives -75dBm. These real-time observations are compared with the calibration values ​​of each grid to calculate the standardized distance.

[0068] It is important to note that due to environmental interference, obstructions, and other factors, some gateways may be unable to receive data packets. In such cases, a valid masking mechanism is used to exclude missing components, preventing the overall positioning results from being affected by the data loss of individual gateways. Grid distance calculation is only performed when the number of valid components reaches a preset threshold; otherwise, the terminal is requested to retransmit.

[0069] Hazard credibility field and candidate instance generation.

[0070] like Figure 2 As shown, within the coarse localization range, the spatial response, category feature matching score, and hazard existence confidence of each location to be judged are calculated for each observation point. The spatial response of the i-th observation point to location x is:

[0071]

[0072] In the formula, Indicates spatial response; x represents the position to be determined; Indicates the coordinates of the i-th observation point; Indicates the spatial decay scale; Indicates visibility in the range of 0 to 1; Indicates the effective field of view; the indicator function is located at x. The value is 1 if the location is within range, and 0 otherwise. The closer the location is to the observation point and the higher the visibility, the greater the spatial response.

[0073] The standardized features at location x are compared with the reference template for the c-th type of hazard to obtain the feature matching score:

[0074]

[0075] In the formula, This represents the matching score of the i-th observation point for location x belonging to the c-th type of hidden danger; This represents the standardized feature vector at position x. This is a reference template representing a Class C hazard. The closer the two are, the better. The closer it is to 1.

[0076] Spatial response and feature matching results are weighted according to observation confidence:

[0077]

[0078] In the formula, This indicates the confidence level of location x belonging to the c-th type of hazard; n represents the number of valid observation points; This represents the confidence weight for the i-th observation point. Output only when the sum of the weights is greater than 0. .

[0079] Comparison of the most recent K frames grid by grid With threshold Retain at least q frames that meet the threshold. Divide the retained grid into independent connected components using either a 2D eight-neighborhood or a 3D twenty-six-neighborhood, and calculate the initial representative coordinates based on the average confidence level across multiple frames.

[0080]

[0081] In the formula, This represents the average confidence level of the most recent K frames; This represents the r-th candidate instance region of the c-th type of hazard; This represents the confidence level of position x at the k-th time in the past (i.e., time tk). This represents the initial representative coordinates of the instance. Stable coordinates can be obtained even when the current frame is not yet detected due to the use of multi-frame averages; if the weight sum is 0, the geometric center of the connected component is used.

[0082] In practice, assume that a potential temperature anomaly is detected in an industrial park. The coarse location range is defined as grid 12 and its eight adjacent grids. For a certain location point x=(25,30,1.5) (unit: meters) in grid 12, there are two observation points: observation point 1 is located at (20,25,1.2), and observation point 2 is located at (30,35,1.8).

[0083] For observation point 1, the distance to position x is calculated to be 5.83 meters. Assume... =3 meters, visibility =0.8, and position x is within the effective field of view of observation point 1, then =exp(-(5.83²) / (2×3²))×0.8×1≈0.43.

[0084] For observation point 2, the distance is 7.07 meters. =4 meters, visibility =0.6, then =exp(-(7.07²) / (2×4²))×0.6≈0.32.

[0085] Assuming the temperature feature vector at location x is M(x) = [0.8, 0.3, 0.6], and the reference template for potential temperature anomalies is [0.7, 0.4, 0.5], then... =exp(-0.5×((0.8-0.7)²+(0.3-0.4)²+(0.6-0.5)²))≈0.86, ≈0.82.

[0086] Assume the confidence weights of the two observation points are respectively =0.9, =0.8, then P(x)=(0.9×0.43×0.86+0.8×0.32×0.82) / (0.9+0.8)≈0.31.

[0087] Statistical analysis was performed on the most recent 10 frames (K=10), assuming that the confidence level of position x exceeds a threshold in 8 frames. If the value is 0.25 (q=8), then the position is retained. Through connected component analysis, adjacent retained grids are divided into independent candidate instances, and their initial representative coordinates are calculated.

[0088] Time alignment, data attribution determination, and precise location.

[0089] After multi-source observations reach the edge processing node, the time of each observation source is first corrected:

[0090]

[0091] In the formula, Represents the original timestamp; This indicates the correction amount obtained through timing or clock models; Indicates the correction timestamp; Indicates a unified reference time; This represents the propagation time difference, with a positive or negative sign. When comparing, take the absolute value: when | If the calibration limit is exceeded, the observation is discarded; if it is not exceeded, the terminal linear velocity and angular velocity are used to complete the first-order position and attitude compensation, and the uncertainties of the clock, linear velocity and angular velocity are added to the observation covariance.

[0092] like Figure 3 As shown, for each observation that has completed time correction, its scores relative to each candidate instance in terms of time, space, category, and event identifier are calculated, and then weighted:

[0093]

[0094] In the formula, This represents the association score of the i-th observation belonging to the r-th candidate instance; , , and All values ​​are normalized to 0 to 1, representing temporal proximity, spatial proximity, category similarity, and event identifier consistency, respectively. , , and The weights are non-negative and their sum is 1. Only if the maximum... Not lower than the association threshold Only when the time is right will the observation be assigned to the corresponding instance.

[0095] In practice, assume there are two candidate instances: instance A is located at (25, 30, 1.5), and instance B is located at (45, 50, 2.0). A new observation is obtained, with an original timestamp of t = 1000 seconds, which is corrected using a clock. =0.5 seconds, then correct the timestamp. =1000.5 seconds. Standard reference time. =1001 seconds, then the time difference =0.5 seconds, which does not exceed the calibration limit (e.g., 2 seconds).

[0096] Assuming the linear velocity of the inspection terminal is v = 1.5 m / s and the angular velocity is ω = 0.1 radians / s, the observations can be compensated to the reference time. Simultaneously, clock correction errors, velocity errors, etc., are propagated into the observation covariance.

[0097] When calculating the association score, assume the spatial coordinates of the observation are (26, 31, 1.6), the distance to instance A is 1.73 meters, and the distance to instance B is 21.9 meters. Set the spatial proximity score. =0.9, =0.1. Time proximity score = =0.95. Category similarity score. =0.85, =0.3. Event identifier consistency score = =1.0.

[0098] Set weights =0.2, =0.4, =0.3, =0.1, then =0.2×0.95+0.4×0.9+0.3×0.85+0.1×1.0=0.89, =0.2×0.95+0.4×0.1+0.3×0.3+0.1×1.0=0.42.

[0099] If the association threshold If the value is 0.6, the observation is assigned to instance A, not instance B. This association mechanism ensures the correct attribution of observation data and avoids the erroneous merging of observations with different potential risks.

[0100] like Figure 4 As shown, the ranging and positioning unit uses a UWB positioning unit, the attitude observation unit uses an inertial measurement unit, and the relative observation sensors include visual sensors or millimeter-wave sensors. UWB provides distance constraints for the inspection terminal, the inertial measurement unit provides attitude constraints, and the visual or millimeter-wave sensors provide the direction, distance, or image position of the potential hazard relative to the terminal. Sensor rotation and translation extrinsic parameters are calibrated after deployment or maintenance. Ranging noise is measured using... This indicates that the same symbol should be avoided when using the same symbol as the subsequent dimensionless residual.

[0101]

[0102] In the formula, Indicates the first Distance measurement values ​​of each UWB anchor point; Indicates the location of the inspection terminal; Indicates the coordinates of the anchor point; This indicates the distance measurement error.

[0103] Visual and millimeter-wave observations are first transformed to a unified coordinate system using installation extrinsic parameters. Then, the residuals of ranging, attitude, vision, and millimeter-wave observations are weighted and summed according to their respective covariances and confidence levels to form a weighted objective function.

[0104]

[0105] In the formula, , , and The costs for ranging, attitude, vision, and millimeter-wave residuals are represented in that order. , , and Non-negative fusion weights; For the total cost; , and These represent the obtained terminal position, terminal attitude, and coordinates of the hazard representative, respectively. Constraints during the solution process. ,Right now It must be a valid three-dimensional rotation matrix that preserves the length and orientation relationship; arbitrary numerical matrices are not acceptable. Millimeter-wave amplitude is only considered when the scattering reference value and penetration thickness have been obtained through calibration or field measurement. Otherwise, only distance and angle observations are used.

[0106] Perform observability and uncertainty checks before outputting the positioning results:

[0107]

[0108] In the formula, This represents the information matrix after removing the numerical regularization term; Indicate the joint state dimension; Indicates the maximum number of allowed conditions; Indicates the covariance of the location of potential hazards; Indicates positional uncertainty; This indicates the upper limit of allowable conditions. The information matrix may be less than rank, the condition number may be too large, or... Exceed Instead of outputting the final coordinates, it triggers a turn, supplementary distance measurement, or remeasurement.

[0109] After the inspection is passed, a fixed location information region is formed by using the coordinates and position covariance of the hidden danger representative. Then, the candidate instance region is expanded according to the possible displacement during the equipment action error, the instruction delay period, and the minimum isolation margin. The closed operation scale is calculated from small to large to obtain a connected region that contains the optimized coordinates and does not merge with adjacent instances as a continuous spatial envelope.

[0110] In practice, assuming there are 4 UWB anchor points with coordinates as follows: =(0,0,0) =(50,0,0) =(0,50,0) =(50,50,0) (unit: meters). The distance values ​​are respectively =35.4 meters =22.4 meters =25.0 meters =15.8 meters, and the distance measurement error covariance is 0.1 meters².

[0111] The inertial measurement unit provides attitude parameters of 5° pitch, 2° roll, and 30° yaw, with an attitude error covariance of 0.01 radians². The visual sensor detects the hazard in the image at (u,v)=(320,240), with a focal length f=500 pixels and an image resolution of 640×480.

[0112] Sensor mounting extrinsic parameters: translation of the vision sensor relative to the terminal =(0.1,0,0.2) meters, rotate This is a 10° rotation around the X-axis. The extrinsic parameters of the millimeter-wave sensor are... =(0,0.1,0.15) meters, rotate This is a rotation of -5° around the Y-axis.

[0113] The terminal location is obtained through joint optimization. =(25,30,1.2) meters, terminal attitude The coordinates of the potential hazard are determined by rotating the object 30° around the Z-axis. =(25,30,1.5) meters, location covariance =diag(0.04,0.04,0.09) m².

[0114] Calculate position uncertainty = =0.42 meters, less than the upper limit of allowable distance. =1.0 meter. Information matrix The rank is 9 (joint state dimension), and the condition number is 25, which is less than 9. =100, which satisfies the observability condition.

[0115] Based on the representative coordinates (25, 30, 1.5) and location covariance of the potential hazard, a 95% confidence ellipsoid is formed. Considering the equipment action error of 0.3 meters, the displacement during the command delay of 0.5 meters, and the safety margin of 0.5 meters, the candidate instance region is expanded and then closed, resulting in a continuous spatial envelope of an ellipsoid centered at (25, 30, 1.5), with a major axis of 2.5 meters, a minor axis of 2.0 meters, and a height axis of 1.8 meters.

[0116] Risk classification and technical measures.

[0117] First, normalize the location uncertainty, then calculate the comprehensive risk score:

[0118]

[0119] In the formula, This represents the normalized positional uncertainty; It is obtained from the previous equation; S represents the maximum allowable limit; S represents the overall risk score. Indicates the credibility of the maximum potential hazard within the candidate region; Indicates the severity of the consequences; Indicates the degree of occlusion; Indicates the ability to handle situations; , , and For non-negative weights that sum to 1, , is the non-negative disposal capacity deduction coefficient; S is the risk score in the range of 0 to 100.

[0120] When the location uncertainty or occlusion exceeds the corresponding threshold, the system first issues a data acquisition command to increase the sampling frequency, change the observation direction, or supplement the ranging; when the risk score reaches the linkage threshold, it issues a disposal command to the pre-authorized isolation, ventilation, or audible and visual alarm devices. Each command includes at least the event number, a monotonically increasing sequence number, a one-time random number, the target spatial envelope, control parameters, and a validity period.

[0121] like Figure 5 As shown, the task closed-loop state is obtained by multiplying five binary flags:

[0122]

[0123] In the formula, This represents the closed-loop state of the j-th task; For identification purposes, only fields that do not change with the action, such as event number, instance number, device number, and instruction binding random number, are compared. As a location marker, check whether the coordinates of the evidence to be processed or re-measured are located within the target envelope; Use the action indicator to check the deviation between the feedback quantity and the control quantity; As an indicator of effectiveness, check whether the technical indicators have reached the improvement threshold; As a marker of evidence integrity, the system checks that the serial number is unused, the random number matches, the feedback is not expired, and the signature is valid. All five items must be 1. Set the value to 1 and allow cancellation; keep the task running if any item is 0.

[0124] After loop closure, only data that passes signature verification, has conflict-free review results, and meets the required sample size are used to update observation weights, category reference templates, and confidence thresholds. The updated model is first checked on an independent validation set, and if performance degrades, it reverts to the previous version. This process uses controlled incremental updates and omits the learning rate formula, which is irrelevant to the core judgment of this invention.

[0125] In practical implementation, it is assumed that =0.42 meters, =1.0 meter, then . =0.82, =0.90, =0.60, =0.30. Set weight. =0.25, =0.20, =0.30, =0.25, =0.10, then S=100×(0.25×0.82+0.20×0.42+0.30×0.90+0.25×0.60-0.10×0.30)=66.9.

[0126] If the risk linkage threshold is 60, then S=66.9>60, triggering a response command. The command includes: event number EVT-2024-001, sequence number SEQ-001, random number RAND-12345678, target spatial envelope centered at (25,30,1.5), ellipsoid parameters (2.5,2.0,1.8), control parameters of starting ventilation equipment at 80% power, and validity period of 300 seconds.

[0127] After the process is completed, the system receives feedback from the device and retest data. Assume an identity identifier. =1 (Event ID, Instance ID, Device ID, and Random Number all match), Location Flag =1 (The retested coordinates (25.2, 30.1, 1.4) are within the target envelope), action marker =1 (Actual power of ventilation equipment is 78%, and the deviation from the control value of 80% is within the allowable range), effect indicator =1 (Temperature drops from 45℃ to 38℃, reaching the improvement threshold), Evidence integrity flag =1 (Serial number not used, random number consistent, feedback time within validity period, signature verification passed).

[0128] but =1×1×1×1×1=1, allowing cancellation. If any flag is 0, the task remains active, requiring reprocessing or supplementary evidence.

[0129] Simplify digital calculation and verification methods.

[0130] The demonstration uses only the received signal strength components from three LoRa gateways. The real-time observation vector is (-70, -80, -75) dBm; the fingerprint mean of grid 12 is (-71, -79, -77) dBm, and the calibration standard deviation is (2, 2, 4) dB; the fingerprint mean of grid 21 is (-65, -86, -76) dBm, and the calibration standard deviation is (2, 3, 2) dB; the fingerprint mean of grid 13 is (-75, -77, -70) dBm, and the calibration standard deviation is (2, 2, 2) dB. All three components are valid, therefore the number of valid components is 3.

[0131] For grid 12, the real-time value is subtracted from the fingerprint mean and divided by the standard deviation to obtain the standardized difference (0.50, -0.50, 0.50). The squared average of the three differences is 0.25, and the square root gives a distance of 0.50. For grid 21, the standardized difference is also obtained (-2.50, 2.00, 0.50), with a squared average of 3.50, and the square root gives a distance of 1.87. For grid 13, the standardized difference is obtained (2.50, -1.50, -2.50), with a squared average of approximately 4.92, and the square root gives a distance of 2.22. Therefore, grid 12 and its adjacent grids are selected as the coarse positioning range for the terminal. Here, the area where the inspection terminal is located is positioned, and the grid is not directly used as the location of potential hazards.

[0132] Within this range, the confidence level, spatial response, and feature matching score of the two valid observation points are taken as (0.90, 0.90, 0.90) and (0.80, 0.80, 0.80), respectively. Substituting these values ​​into the formula yields... It is approximately 0.730. Let... =0.37m =1m、 =0.82、 =0.90、 =0.60、 =0.30, take =0.25、 =0.20、 =0.30、 =0.25 and =0.10, where the sum of the four weighted coefficients is 1, then S=66.9. If all five closed-loop flags are 1, then =1. The above calculations are summarized as follows:

[0133] =(0.90×0.90×0.90+0.80×0.80×0.80) / (0.90+0.80)≈0.730;

[0134] S=100(0.25 × 0.82 + 0.20 × 0.37 + 0.30 × 0.90 + 0.25 × 0.60 -0.10 × 0.30)=66.9;

[0135] =1×1×1×1×1=1.

[0136] The numerical values ​​in the formula are only used to illustrate the order of operations and are not considered as experimental performance results. In actual verification, the average positioning error, 95th percentile error, recall rate, false alarm rate, false association rate, and false cancellation rate were statistically analyzed under the same detection threshold. The complete method was compared with a scheme that did not perform time error propagation, did not use installation extrinsic parameters, and did not perform association gating using the same data partitioning and hardware conditions.

[0137] Example 2

[0138] This embodiment provides an intelligent identification, location, and closed-loop handling system for safety hazard points, including a data acquisition and source discrimination module, a mobile terminal coarse positioning module, a fixed node observation source coordinate module, a hazard credibility field and instance decomposition module, a clock correction and error propagation module, a multi-source observation association module, a ranging and positioning and attitude observation module, a relative observation sensor module, an external parameter calibration and coordinate transformation module, a joint positioning and observability determination module, a continuous spatial envelope module, a risk and control command module, an equipment feedback and anti-replay verification module, and an evidence review and update module.

[0139] The modules work collaboratively as follows: The data acquisition and source discrimination module collects environmental, equipment, image, attitude, and inspection event data and distinguishes between two types of event sources: fixed nodes and mobile terminals; the fixed node observation source coordinate module uses the registered coordinates of fixed nodes only as observation source coordinates and does not directly identify them as hazard coordinates; the mobile terminal coarse positioning module compares the real-time signal characteristics formed by multiple wireless receiving nodes within the same event time window with the regional fingerprint to determine the terminal coarse positioning range; the hazard credibility field and instance splitting module forms a hazard existence credibility field within the coarse positioning range and establishes different hazard state quantities for different connected domains, registering them as mutually independent candidate hazard instances; the clock correction and error propagation module corrects the multi-source observation timestamps and adds clock and motion uncertainties to the effective observation covariance; the multi-source observation association module calculates association scores based on temporal proximity, spatial proximity, category feature similarity, and event identifier consistency, outputting only observations that reach the association threshold to the corresponding instance, while observations that do not reach the threshold enter the supplementary collection queue; the ranging, positioning, and attitude observation modules are provided with ranging and attitude constraints by the UWB positioning unit and the inertial measurement unit, respectively, and the relative observation sensor module provides information through a visual sensor or millimeter-wave sensor. The system provides observations of the direction, distance, or amplitude of the hazard relative to the terminal; the extrinsic parameter calibration and coordinate transformation module maintains the version, residuals, and covariance of the sensor's installed extrinsic parameters and transforms the relative observations to the global coordinate system; the joint positioning and observability determination module jointly solves for the terminal position, terminal attitude, and representative coordinates of the hazard, and outputs the positioning result when the information matrix meets the full rank and condition number thresholds; the continuous spatial envelope module performs expansion and closure operations on the fixed position information region, candidate instance region, and safety margin to generate a connected control range containing optimized coordinates; the risk and control command module determines the hazard credibility, location uncertainty, and risk based on the risk level, location uncertainty, and risk level. The risk attribute generates acquisition control instructions or disposal control instructions with event number, instruction sequence number, one-time random number, target space envelope, control parameters and validity period; the device feedback and anti-replay verification module receives device response, execution status and retest data, and outputs five closed-loop flags of identity, location, action, effect and evidence integrity based on identity digest, envelope inclusion, action deviation, technical indicator improvement and cryptographic verification results; the evidence review and update module cancels the item only when all five closed-loop flags are 1, and updates the observation weight, category reference template and credibility judgment threshold using the verification passed and conflict-free review samples.

[0140] In practical applications, this invention can be widely used for the identification and handling of safety hazards in various complex scenarios. For example, in petrochemical industrial parks, it can monitor safety hazards such as leak points, high-temperature areas, and abnormal vibrations in real time; during power facility inspections, it can identify problems such as equipment overheating, insulation damage, and structural deformation; and at construction sites, it can detect risks such as tower crane tilting, foundation pit deformation, and personnel violations.

[0141] During system deployment, the layout of fixed sensor nodes and mobile inspection terminals is first set according to the characteristics of the scenario. Fixed nodes include environmental sensors, fixed cameras, gas detectors, etc., while mobile terminals include inspection robots, drones, handheld devices, etc. Each device is configured with a unique identifier, and its installation location, detection range, field of view, and other parameters are recorded.

[0142] When an abnormal event is detected, the system automatically initiates the identification and location process. First, the source of the event is determined. If it's a fixed node, its installation location is used as the starting point for observation; if it's a mobile terminal, its approximate location range is determined through wireless fingerprinting. Then, within the coarse location range, a hazard confidence field is generated through multi-sensor data fusion, and independent candidate hazard instances are identified through connected component analysis.

[0143] For each candidate instance, the system performs precise temporal and spatial correlation to ensure correct attribution of observation data. A joint localization algorithm simultaneously solves for the pose of the inspection terminal and the location of potential hazards, and evaluates the observability and uncertainty of the localization results. Only localization results that meet quality requirements are output; otherwise, the system automatically triggers supplementary observations.

[0144] Once the location is determined, the system calculates a comprehensive risk score based on the reliability of the hazard, location accuracy, and risk attributes, and takes corresponding measures according to the risk level. The response instructions include complete technical parameters and safety verification information to ensure the reliability and traceability of the instructions.

[0145] After the processing is completed, the system verifies the effectiveness of the processing through five technical evidence verification mechanisms. Only when all verification conditions are met will the item be closed, ensuring the authenticity and effectiveness of the closed-loop processing. At the same time, the system will continuously optimize model parameters based on the verification results to improve recognition accuracy and processing efficiency.

[0146] This invention, through a systematic technical architecture and a rigorous quality control mechanism, enables intelligent identification, precise location, and closed-loop handling of safety hazards in complex scenarios, providing reliable technical support for industrial safety management.

[0147] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for intelligent identification, location, and closed-loop handling of security hazard points in complex scenarios, characterized in that, Includes the following steps: S1, data acquisition, event identification, and low-power transmission; S2. Event source diversion and coarse regional localization; S3, Generation of Hidden Danger Credibility Field and Candidate Instances; S4, time alignment, data attribution determination, and precise positioning; S5. Risk Classification and Technical Response; S6. Feedback verification and model update.

2. The method according to claim 1, characterized in that, The specific steps of step S1 are as follows: collecting environmental, equipment, image, posture or inspection event data from fixed sensor nodes or mobile inspection terminals, combining scene, area, time window, terminal, hazard category or status field into an event identifier, and uploading it to the edge processing node through a low-power wireless communication network; The low-power wireless communication network is a LoRa network; during the deployment phase, the calibration terminal sequentially sends multiple frames of data packets in each spatial grid, records the received signal strength, signal-to-noise ratio, data packet reception rate, and arrival time difference under clock synchronization conditions according to the gateway number, and saves the mean and standard deviation of each component; during online detection, the mobile inspection terminal sends data packets in the same way, and each gateway forms a real-time observation vector in a fixed order. For each valid component, calculate the difference between the real-time value and the grid mean, normalize it using the standard deviation of the component, and then average and square-take the square root of the squared normalized difference to obtain the distance between each grid. Select the first L grids and their adjacent grids in ascending order of distance as the coarse positioning range. If the number of valid components is insufficient, request retransmission.

3. The method according to claim 1, characterized in that, Step S2 specifically involves: determining the observation source coordinates and coarse positioning range based on the event source; when the event originates from a fixed sensing node, using the valid installation coordinates in the device registry as the observation source coordinates, and forming a coarse positioning area based on the effective detection radius and field of view direction; when the event originates from a mobile inspection terminal, comparing the real-time signal characteristics formed by multiple wireless receiving nodes within the same event time window with the area fingerprint to determine the terminal's coarse positioning range.

4. The method according to claim 1, characterized in that, The specific steps of step S3 are as follows: In the coarse positioning area or within the coarse positioning range, a hazard existence credibility field is formed based on the observation position uncertainty, observation direction, effective field of view, occlusion visibility and category feature matching results. A candidate location set is obtained through multi-frame thresholding and connected component clustering. Different connected components are registered as candidate hazard instances and the initial representative coordinates of each instance are calculated. Step S3 calculates the spatial response, category feature matching score, and hazard existence confidence level sequentially using the following three formulas: ； in: Indicates spatial response, The location to be determined. Let i be the coordinates of the i-th observation point. For spatial attenuation scale, For visibility, For an effective field of view, Only in the effective field Non-zero inside; This represents the matching score of the i-th observation point for location x belonging to the c-th type of hazard. To standardize the feature vector, This is a reference template for class c; This indicates the confidence level of location x belonging to the c-th type of hazard; n represents the number of valid observation points. To measure the confidence level weight; Measure how consistent the representation of source i is with that of category c at x; Output only when the sum of the weights involved in the calculation is greater than 0. .

5. The method according to claim 4, characterized in that, Step S3 performs threshold comparison on each grid at the most recent K detection times, and retains the grids that have reached the threshold at least q times; Perform two-dimensional eight-neighbor or three-dimensional twenty-six-neighbor connected component partitioning on the retained grid and delete connected components with too small area or volume. Register each remaining connected component as an independent candidate instance. Calculate the weighted centroid of each instance using the average confidence of the most recent K frames as the weight. When the weight sum is 0, use the geometric center of the connected component instead.

6. The method according to claim 1, characterized in that, The specific steps of step S4 are as follows: correcting the timestamps of multi-source observations, using the terminal linear velocity and angular velocity to compensate asynchronous observations within the allowed time difference to a unified reference time, propagating clock correction error, linear velocity error and angular velocity error to the effective observation covariance, and assigning observations that have reached the association threshold to the corresponding candidate instances based on the corrected time difference, spatial distance, category feature similarity and event identifier consistency. By combining ranging constraints, attitude constraints, projection of the hazard relative to the terminal, orientation, distance or amplitude constraints, and sensor installation extrinsic parameters, the terminal position, terminal attitude, hazard representative coordinates and position covariance are jointly solved for each candidate instance. After removing the numerical regularization term, the rank and condition number of the joint information matrix are judged. The positioning result is output only when the observability threshold is met. The positioning information ellipsoid and the candidate instance extended region are closed, and the connected region containing the coordinates of the hidden danger is extracted as the continuous spatial envelope. Step S4: First, correct the time according to the following formula: ； in: This is the original timestamp. This is the clock correction amount. To correct the timestamp, For reference time, The propagation time difference is positive or negative; when | If the observation exceeds the calibration limit, discard the observation; otherwise, use the terminal linear velocity and angular velocity to convert the observation to the reference time, and add the additional covariance formed by the clock, linear velocity and angular velocity errors to the original observation covariance.

7. The method according to claim 6, characterized in that, Step S4 converts temporal proximity, spatial proximity, category feature similarity, and event identifier consistency into scores ranging from 0 to 1, and calculates the association score between the i-th observation and the r-th candidate instance according to the following formula: ； in: For correlation scoring, , , and The scores are, in order, time, space, category, and event identifier. , , and The weights are non-negative and sum to 1; only the maximum score not lower than the association threshold is considered. The observations are assigned to the corresponding instances, and the remaining observations enter the supplementary collection queue.

8. The method according to claim 6, characterized in that, The ranging constraint in step S4 is provided by the ranging and positioning unit, which is a UWB positioning unit; the attitude constraint is provided by the attitude observation unit, which is an inertial measurement unit, and the relative observation sensor includes a visual sensor or a millimeter-wave sensor; the UWB ranging model is: ； in: For the first Distance values ​​at each anchor point For the location of the inspection terminal, For anchor point coordinates, For ranging error; visual or millimeter-wave sensors transform relative observations to the global coordinate system through calibrated rotation and translation extrinsic parameters, and then weight and sum the ranging, attitude, visual and millimeter-wave residuals according to their respective covariance and confidence level, and jointly solve the terminal position, terminal attitude and hazard representative coordinates; Step S4 checks whether the information matrix is ​​full rank and the condition number does not exceed the calibration upper limit after removing the numerical regularization term, and calculates the position uncertainty according to the following formula: ； in: For the covariance of the location of the hidden danger, The positional uncertainty; only when the observability condition is satisfied and If the position does not exceed the allowed upper limit, output the positioning result. Then, perform expansion and closure operations on the positioning information region, candidate instance region, and safety distance to obtain a connected region containing optimized coordinates that does not merge with adjacent instances as a continuous spatial envelope.

9. The method according to claim 1, characterized in that, Specifically, step S5 involves generating a collection control command or a disposal control command with an event number, instruction sequence number, one-time random number, target space envelope, control parameters, and validity period based on the credibility of the hidden danger, the uncertainty of the location, and the risk attributes. Step S5 calculates the risk score according to the following formula: ； in: To normalize the position uncertainty, S represents the maximum allowable limit; S represents the overall risk score. To maximize the credibility of the greatest potential risks, As for the severity of the consequences, To the degree of obstruction, To improve the ability to handle [the situation], , , and For non-negative weights that sum to 1, It is a non-negative processing capacity deduction coefficient; when the location uncertainty or occlusion degree is too high, a supplementary sampling instruction is issued; when S reaches the linkage threshold, a processing instruction with event number, sequence number, one-time random number, target space envelope, control parameters and validity period is issued.

10. The method according to claim 1, characterized in that, The specific steps of step S6 are as follows: receiving device feedback and retest data, calculating five binary flags for identity, location, action, effect and evidence integrity based on the consistency of identity digest before and after processing, whether the evidence coordinates after processing are within the spatial envelope, the deviation between telemetry and control vectors fed back by the device, the changes in technical indicators before and after processing and the anti-replay signature verification results. The item is cancelled only when all five flags are 1, and the observation weight, category reference template and credibility judgment threshold are updated using the verification samples that have passed the verification and have no conflicts. Step S6 calculates the closed-loop state of the j-th task according to the following formula: ； in: , , , and The indicators are, in order: identity, location, action, effect, and evidence integrity, all of which are binary variables. The identity indicator only compares fields that do not change with the disposal, such as event number, instance number, device number, and instruction binding random number. The location indicator checks whether the evidence coordinates are within the spatial envelope. The action indicator compares the control quantity and the feedback quantity. The effect indicator compares the technical indicators before and after the disposal. The evidence indicator verifies that the serial number, random number, validity period, and digital signature were not used. An item is cancelled only when all five indicators are 1, and the observation weight, category template, and judgment threshold are updated only using samples with valid signatures, no conflicts after review, and the required number.

11. A smart system for identifying, locating, and handling safety hazard points, characterized in that: The system includes modules for data acquisition and source identification, mobile terminal coarse positioning, fixed node observation source coordinates, hazard confidence field and instance decomposition, clock correction and error propagation, multi-source observation correlation, ranging and attitude observation, relative observation sensor, extrinsic parameter calibration and coordinate transformation, joint positioning and observability determination, continuous spatial envelope, risk and control command, equipment feedback and anti-replay verification, and evidence review and update. The data acquisition and source identification module uses the registered coordinates of fixed nodes only as observation source coordinates. The instance decomposition module establishes different hazard state quantities for different connected domains. The error propagation module incorporates clock and motion uncertainties into the effective observation covariance. The multi-source observation correlation module only outputs observations that reach the correlation threshold to the corresponding instance. The joint positioning and observability determination module outputs the positioning result when the information matrix meets the threshold. The continuous spatial envelope module generates a connectivity control range containing optimized coordinates; the device feedback and anti-replay verification module outputs five closed-loop flags based on identity digest, envelope inclusion, action deviation, technical indicator improvement, and cryptographic verification results.

Images