Safety monitoring system based on multi-source state perception

By using a multi-source state perception system, the perception reliability is calibrated and an individual state baseline is constructed to identify complex risk events. This solves the problem of insufficient reliability of equipment operating status and perception data in existing technologies, and enables accurate identification and dynamic feedback of low-intensity anomalies.

CN122715325BActive Publication Date: 2026-11-10SHANDONG WUHUI INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202611207101.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-08-11
Publication Date
2026-11-10
Estimated Expiration
2046-08-11

AI Technical Summary

Technical Problem

Existing elderly care safety monitoring systems are inadequate in assessing equipment operating status and the reliability of sensor data. They cannot adapt to individual differences, struggle to identify low-intensity anomalies and perform time-series correlation analysis, leading to false alarms or missed alarms, and lack a dynamic feedback mechanism.

Method used

The safety monitoring system based on multi-source state perception acquires the activity status of the monitored object and the operating status of the sensing equipment through the state acquisition module, calibrates the perception credibility using the trusted calibration module, constructs an individual state baseline by combining historical activity patterns, identifies complex risk events through the risk identification module, and performs hierarchical monitoring and perception credibility updates through the monitoring and handling module.

Benefits of technology

It improves the reliability of monitoring data and the accuracy of anomaly judgment, and can identify complex risk events formed by the continuous superposition of multiple low-intensity anomalies, reduce false alarm interference and enhance the system's continuous adaptability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122715325B_ABST
    Figure CN122715325B_ABST
Patent Text Reader

Abstract

This invention discloses a safety monitoring system based on multi-source state perception, relating to the field of safety monitoring technology. The system acquires the activity status of the monitored object and the operating status of the sensing equipment, forms an activity status sequence through time-space correlation calibration, determines the perception reliability based on data continuity, signal stability, and cross-source consistency, and corrects the activity status deviation by combining the individual state baseline. Based on the continuous evolution of position, posture, and static state, it identifies composite risk events formed by the superposition of low-intensity anomalies, and implements multi-source complementary verification and graded monitoring and handling according to risk confidence. The perception reliability and individual state baseline are dynamically updated using the event confirmation results, thereby improving the accuracy and adaptability of elderly care safety risk identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology, specifically to a safety monitoring system based on multi-source state perception. Background Technology

[0002] Existing elderly care safety monitoring systems typically utilize positioning devices or wearable devices to collect the location, posture, and activity information of monitored individuals, and identify abnormal states such as falls and prolonged periods of inactivity based on fixed thresholds. However, different sensing devices may be affected by factors such as signal obstruction, device disconnection, data loss, and location drift. Current technologies generally lack a joint evaluation of device operating status and the reliability of sensing data, making it easy to directly use low-quality sensing data for risk assessment. Furthermore, fixed monitoring thresholds are difficult to adapt to individual differences among monitored individuals in terms of daily routines, activity levels, and regional usage habits, easily leading to false alarms or missed alarms.

[0003] Furthermore, existing technologies primarily identify strong abnormal events such as single falls or significant periods of stillness. They lack time-series correlation analysis and multi-dimensional composite judgment mechanisms for low-intensity anomalies—such as slow positional shifts, gradual changes in posture, and continuously increasing periods of stillness—which, while insufficient to trigger an alarm on a single occurrence, may accumulate into safety risks. Once a risk is triggered, an alarm is typically generated directly, failing to utilize different sensing devices for complementary verification based on the risk's reliability, and also making it difficult to continuously adjust device reliability and individual activity baselines using event confirmation results. Therefore, there is an urgent need for an elderly care safety monitoring system that can simultaneously address the reliability calibration of sensing data, adapt to individual differences, identify composite risks, and provide dynamic feedback updates. Summary of the Invention

[0004] This invention addresses the shortcomings of existing technologies by proposing a safety monitoring system based on multi-source state perception.

[0005] The technical solution to achieve the purpose of this invention is as follows: A safety monitoring system based on multi-source state awareness includes: The status acquisition module acquires the activity status of the monitored object and the operating status of the sensing device, and forms an activity status sequence through time alignment. The trusted calibration module determines the perceived trustworthiness based on the operating status of the sensing device, constructs an individual state baseline by combining historical activity patterns, corrects the deviation of the activity state sequence, and forms a trusted state feature set. The risk identification module analyzes the temporal evolution relationship of position, attitude and static state based on the trusted state feature set, identifies composite risk events formed by the continuous superposition of low-intensity anomalies, and determines the risk confidence level. The monitoring and handling module calls the corresponding sensing devices for complementary verification based on the risk confidence level, generates and executes hierarchical monitoring instructions based on the verification results, and updates the sensing confidence level and individual status baseline.

[0006] Furthermore, the determination of perceived credibility includes: The activity status sequence is split according to the device identifier of the sensing device, and the activity status data and operation status data under the continuous time slice node are sorted according to the standard collection timestamp to form a single device time-series status dataset. Within the sliding statistical window, data continuity is determined based on the number of valid data frames and the theoretical number of data frames; signal stability is determined based on the dispersion of the received signal strength indication value; and cross-source consistency is determined based on the cross-source deviation of similar activity state data within the same time slice node. A weighted summation method is used to fuse data continuity, signal stability, and cross-source consistency to obtain the perception reliability bound to the sensing device identifier.

[0007] Furthermore, the construction of individual status baselines and the correction of deviations include: Retrieve historical activity status data within a preset baseline period and remove data for time periods corresponding to abnormal events; Based on the statistical baseline location distribution interval of time period and monitored area, the baseline value of the probability of body posture and the baseline threshold of static duration, an individual status baseline is formed; The real-time location coordinates, body posture, and duration of stillness are compared with the individual status baseline of the corresponding time period and the corresponding monitoring area to obtain the original deviation of location, original deviation of posture, and original deviation of stillness duration. The original deviations are corrected by using the perception credibility of the corresponding sensing device to form a credible state feature set.

[0008] Further identification of complex risk events includes: A sliding analysis window is used to traverse the trusted state feature set, and the data in the window is mapped to state nodes according to the standard collection timestamp. Calculate the position displacement, attitude transition type, and static duration increment of adjacent state nodes, and construct the active state evolution relationship with state nodes as vertices and state transition features as edges; The comprehensive deviation is obtained by integrating the corrected deviations of position dimension, attitude dimension and static duration dimension, and low-intensity abnormal nodes are marked to form anomaly accumulation features; Composite risk events are identified based on the proportion of low-intensity abnormal nodes and the number of dimensions involved in the anomalies, and the risk type, occurrence area, and risk confidence level of the composite risk events are determined.

[0009] Furthermore, the determination of the risk type, the area of ​​occurrence, and the confidence level of the risk includes: Based on the combination of the most prevalent abnormal dimensions within the sliding analysis window, the composite risk event is identified as a sedentary-to-get-up risk, a slow fall risk, or a regional stagnation risk. The monitoring area with the highest concentration of low-intensity abnormal nodes was identified as the occurrence area; The significance of the anomaly is determined by the proportion of low-intensity anomaly nodes, and the anomaly dimension coefficient is determined by the number of dimensions involved in the anomaly. The risk confidence level is obtained by integrating the proportion of low-intensity abnormal nodes, the abnormal dimension coefficient, and the perception credibility of the sensing devices corresponding to the abnormal nodes.

[0010] Furthermore, the formation of the activity state sequence includes: Acquire real-time location coordinates, body posture, duration of stillness, and operating status of sensing devices, and add local acquisition timestamps and monitoring area identifiers to the collected data; Extract the activity status field and the running status field according to the unified data field specification to form a standardized status dataset; The local sampling clock offset is determined based on the issuance time of the synchronous acquisition command and the local clock time of the sensing device, the initial timestamp is corrected, and the status data is collected to the time slice node. Spatial grouping is performed according to the monitoring area identification, and repeated location data and location data with overlapping sensing coverage are calibrated. The calibrated time slice nodes are spliced ​​together in time slice order to form an activity state sequence.

[0011] Furthermore, the acquisition of real-time location coordinates, body posture, and duration of stillness includes: The beacon is located via Bluetooth and broadcasts its identification and fixed coordinates. The wearable monitoring wristband scans the broadcast signal and records the received signal strength indicator value. The Bluetooth gateway then forwards the scan results. The distance between the wearable monitoring wristband and the Bluetooth positioning beacon is calculated based on the logarithmic distance path loss model, and the real-time location coordinates are calculated based on the distance. Acceleration and angular velocity data are collected, and a complementary filtering algorithm is used to obtain pitch and roll angles. Body posture is determined based on the attitude angle threshold range. The cumulative duration of static posture without category switching is calculated to obtain the duration of stillness.

[0012] Furthermore, based on the risk confidence level, corresponding sensing devices are invoked for complementary verification, including: Risk levels are categorized based on risk confidence levels, and complementary review strategies are matched for different risk levels. In low-risk areas, the Bluetooth gateway in the affected area is invoked to adjust the scanning frequency and Bluetooth beacon matching range; Under the medium-risk level, a complementary review strategy corresponding to the low-risk level is implemented, and high-precision collection commands are sent to the wearable monitoring wristband to obtain data on body posture and duration of stillness. Under high-risk levels, a full review strategy is implemented, triggering vibration alerts on the wearable monitoring wristband to collect the response feedback status of the monitored individual.

[0013] Furthermore, the determination of the review results and the generation of graded monitoring instructions include: The activity status data and operational status data during the review period are standardized, time-aligned, and spatially calibrated to form a review status sequence. Calculate the overall deviation of each time slice node in the verification state sequence; Based on the trend of changes in the overall deviation and the consistency between the combination of abnormal dimensions and the original composite risk event, it is determined whether the event is confirmed, not confirmed, or escalated. Based on the determination of the authenticity of the event and the risk level, risk relief instructions, mild attention instructions, early warning and handling instructions, or emergency alarm instructions are generated and sent to the corresponding monitoring terminals for execution.

[0014] Further updates to perceived credibility and individual state baselines include: When an event is confirmed and the verification data of the corresponding sensing device is consistent with the event judgment result, the sensing credibility is updated using a positive adjustment formula. When an event is confirmed as a false alarm and the false alarm is caused by data deviation of the corresponding sensing device, the sensing confidence level is updated using a negative adjustment formula; Write the updated perception credibility into the device status database, mark the data corresponding to the time period of real abnormal events as abnormal data, and mark the data corresponding to the time period of false alarm events as normal data; Based on the updated historical dataset, the baseline characteristic parameters for each time period and each monitored area are recalculated to complete the dynamic iteration of the individual status baseline.

[0015] Compared with the prior art, the advantages of this invention are as follows: 1. Based on the data continuity, signal stability and cross-source consistency of sensing devices, the perception credibility is calculated, and the individual status baseline is constructed by combining the historical activity patterns of the monitored objects. The perception credibility is used to correct the deviation of the real-time activity status from the individual status baseline. This can reduce the impact of equipment failure, signal interference and individual behavior differences on risk identification results, and improve the credibility of monitoring status data and the accuracy of anomaly judgment. 2. By associating the location changes, posture changes, and static continuous states of the monitored objects over a continuous period of time, the system constructs the evolution relationship of activity states, identifies composite risk events formed by the continuous superposition of multiple low-intensity anomalies, and calls on sensing devices to implement complementary verification and graded monitoring and handling based on the risk confidence level. At the same time, the system uses the event confirmation results to update the perception confidence level and individual state baseline, which can promptly detect progressive safety risks, reduce false alarm interference, and enhance the system's continuous adaptability. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the safety monitoring system based on multi-source state perception in this invention; Figure 2 This is a schematic diagram of the activity state sequence construction process in this invention; Figure 3 This is a schematic diagram of the process of generating a trusted state feature set in this invention; Figure 4 This is a schematic diagram of the process for identifying and classifying and monitoring complex risk events in this invention. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0018] This invention discloses a safety monitoring system based on multi-source state awareness, comprising: The status acquisition module obtains the activity status of the monitored object and the operating status of each sensing device, and performs correlation calibration on the activity status according to the acquisition time and the monitored area to form an activity status sequence.

[0019] The trusted calibration module calculates data continuity, signal stability, and cross-source consistency based on the operating status of the sensing devices to determine the trusted level of each sensing device. It also constructs an individual status baseline by combining the historical activity patterns of the monitored object and corrects the deviation of the real-time activity status from the individual status baseline based on the trusted level, thus forming a trusted status feature set.

[0020] The risk identification module associates the location changes, posture changes, and static continuous states of the monitored object with a set of reliable state features over a continuous period of time, constructs the evolution relationship of the activity state, identifies composite risk events formed by the continuous superposition of multiple low-intensity anomalies, and determines the risk type, occurrence area, and risk confidence level of the composite risk event.

[0021] The monitoring and handling module calls the corresponding sensing devices for complementary verification based on the risk confidence level, generates and executes hierarchical monitoring instructions based on the verification results, and uses the event confirmation results to update the perception confidence level and individual status baseline.

[0022] refer to Figure 1 , Figure 1 This is a schematic diagram of a safety monitoring system based on multi-source state awareness.

[0023] Figure 1 The composition and data transmission relationships among the multi-source sensing devices, status acquisition module, trusted calibration module, risk identification module, monitoring and handling module, and monitoring terminal are shown. The feedback and update process of event confirmation results on the perception trustworthiness and individual status baseline is also shown.

[0024] It should be noted that, Figure 1 This is only a schematic diagram to help understand the system composition, module connection relationships and data flow, and does not represent the precise working state in actual operation.

[0025] Furthermore, the status acquisition module acquires the activity status of the monitored object and the operating status of each sensing device, and performs correlation calibration on the activity status according to the acquisition time and the monitored area to form an activity status sequence, including: Following a pre-set 200ms uniform sampling period, synchronous data collection commands are sent to all sensing devices deployed in each monitoring area, collecting two types of data in parallel. This sampling period is set based on the shortest duration of daily human movements in elderly care scenarios, ensuring complete capture of key actions such as getting up and falling, while also considering the power consumption control of the sensing devices. The sensing devices in this solution include Bluetooth positioning beacons and Bluetooth gateways deployed in each monitoring area, as well as wearable monitoring wristbands worn by the monitored individuals.

[0026] Of the two types of data collected, the first type is the activity status data of the monitored object, which specifically includes real-time location coordinates, body posture, and duration of stillness.

[0027] The real-time location coordinates are obtained through a combination of Bluetooth positioning beacons and a wearable monitoring wristband: Bluetooth positioning beacons are deployed at fixed points in each monitored area, continuously broadcasting Bluetooth signals with their unique identifiers and fixed coordinates; the wearable monitoring wristband scans the broadcast signals of surrounding Bluetooth positioning beacons during the sampling period, reads the received signal strength indication value of each signal and records the corresponding beacon identifier; the wristband uploads the scan results to the Bluetooth gateway, which then forwards them to the status acquisition module. The status acquisition module calculates the distance between the wristband and each beacon based on a logarithmic distance path loss model. The calculation formula is: Among them, reference distance Received power at a reference distance of 1m Take -59dBm, The received signal strength indication value of the beacon actually scanned by the wristband, and the path loss index. The value is set to 2.4. This parameter is set based on the typical wireless propagation characteristics of indoor home environments, which can ensure that the indoor positioning error is controlled within 1m. The three beacons with the highest signal strength are selected, and the trilateration method is used to calculate the planar coordinates of the monitored object in the monitored area, which are the real-time position coordinates.

[0028] Body posture and duration of stillness are collected by a wearable monitoring wristband: the wristband has a built-in three-axis accelerometer and a three-axis gyroscope, which simultaneously collect three-axis acceleration data and three-axis angular velocity data of the wrist; a complementary filtering algorithm is used to fuse the two types of data to obtain the pitch angle and roll angle of the wristband relative to the horizontal plane. In the complementary filtering, the weight of the accelerometer is set to 0.02 and the weight of the gyroscope is set to 0.98. This parameter is set according to the low-frequency characteristics of human daily movements, which can take into account both the calculation accuracy of static posture and the response speed of dynamic movements; based on the preset posture angle threshold range, the pitch angle and roll angle are mapped to the corresponding body postures, specifically including three static postures (standing, sitting, and lying down) and two dynamic postures (walking and falling); the specific judgment rule is: when the absolute value of the pitch angle is ≤15° and the absolute value of the roll angle is ≤10°, and the pitch angle is ≤30° for three consecutive days, the pitch angle and roll angle are ≤15°. When the acceleration amplitude fluctuation within a sampling period is ≤0.1g, it is determined to be a standing posture; when the pitch angle is in the range of 30°~60°, the absolute value of the roll angle is ≤20°, and the acceleration amplitude fluctuation is ≤0.1g, it is determined to be a sitting posture; when the pitch angle is in the range of 75°~105°, it is determined to be a lying posture; when the acceleration amplitude fluctuation is ≥0.3g and the average angular velocity is ≥25° / s within 5 consecutive sampling periods, it is determined to be a walking posture; when an instantaneous acceleration peak is detected to be ≥1.8g and the subsequent posture switches to lying down, it is determined to be a fall posture; where g is the acceleration due to gravity; the above thresholds are set according to the wrist angle distribution pattern of the daily posture of the adult human body, which can cover the accurate identification of most daily postures in the elderly care scenario; the cumulative time during which the body posture does not change category within a continuous sampling period is the static duration.

[0029] The second category is the data on the operating status of sensing devices, which specifically includes the device's online status, received signal strength, power supply voltage, and local clock count. This data is directly read by the Bluetooth positioning beacon, Bluetooth gateway, and wearable monitoring wristband through their own system status interfaces.

[0030] Each completed data collection is uploaded to the status collection module after the corresponding sensing device adds a local collection time stamp and the identification of the monitoring area.

[0031] The heterogeneous raw data uploaded by different types of sensing devices is normalized and converted into a format. Based on the preset unified data field specification, the activity status field and the running status field are extracted, and a globally unique data identifier is assigned to each parsed status data.

[0032] All data local acquisition timestamps are uniformly mapped to initial timestamps based on the system's global reference clock. The original monitoring area identifier is retained as the spatial dimension attribute of the data. The original data is then structured and processed to output a standardized state dataset containing all state data.

[0033] First, the local sampling clock offset of each sensing device is calculated: when the synchronization acquisition command is issued, the issuance time corresponding to the system's global reference clock is recorded. After receiving the command, the sensing device records the corresponding local clock time and reports it along with the operating status data. The single clock offset value is calculated by the difference between the issuance time and the local clock time. The arithmetic mean of the clock offset values ​​of three consecutive measurements is taken as the effective local sampling clock offset of the device. This calculation method is set based on the random characteristics of network transmission fluctuations. Multiple averaging can effectively reduce the random error of a single measurement and improve the accuracy of offset calculation.

[0034] Based on the obtained local sampling clock offset, the initial timestamp of each data in the standardized state dataset is corrected for deviation. The correction method is to sum the initial timestamp with the local sampling clock offset of the corresponding sensing device to obtain the standard acquisition timestamp, thereby eliminating the local clock drift and synchronization error of different sensing devices.

[0035] The entire set of activity status data is divided into time slices according to a preset fixed time step of 1 second. All activity status data falling within the same time slice range, along with their corresponding operational status data, are aggregated to the corresponding time slice node, achieving unified alignment of all sensor source data in the time dimension. The 1-second time step is set based on the granularity of state changes in daily human activities, ensuring the continuity of the state sequence while avoiding redundant calculations caused by excessively fine data granularity.

[0036] Based on the monitoring area identifier carried by each activity status data, the full status data within the same time slice node is spatially grouped to obtain a status data subset corresponding to each monitoring area; each status data subset contains the activity status data and operation status data of all sensing devices within the corresponding monitoring area in the current time slice.

[0037] For location data of the same monitored object collected by multiple Bluetooth gateways within the same monitoring area, a distance threshold of 0.5m is used. The Euclidean distance between location coordinates is calculated to determine if they are from the same sampling point. Duplicate sampling data with a distance less than the location distance threshold are discarded, retaining only the single data point with the stronger received signal. The 0.5m threshold is set based on the typical error range of indoor Bluetooth positioning, accurately distinguishing different sampling points and avoiding duplicate data interference with subsequent analysis.

[0038] For location data where sensing coverage overlaps at the boundary of the monitored area, spatial fusion is performed based on the beacon installation location and the received signal strength. Calibration is then performed using data from sensing devices that are closer to the sampling point and have higher signal strength as a benchmark, thus completing the spatial dimension correlation calibration.

[0039] Following the chronological order of time slices, all time-aligned and spatially calibrated time slice nodes are sequentially concatenated. The status data within each time slice node is split according to data attributes, generating an active status subset and a running status subset. The active status subset consists of calibrated active status data collected by all sensing devices within the corresponding monitoring area during that time slice, while the running status subset consists of running status data synchronously reported by the corresponding sensing devices. The data within these two subsets are correlated one-to-one using data identifiers. Each time slice node synchronously encapsulates the temporal attributes of the corresponding collection period and the spatial attributes of the corresponding monitoring area. Finally, a sequentially arranged active status sequence with both temporal and spatial attributes is formed and output to the trusted calibration module.

[0040] refer to Figure 2 , Figure 2 A schematic diagram illustrating the process of constructing an activity state sequence.

[0041] Figure 2 The process is shown in the illustration of how data collected from positioning beacons, Bluetooth gateways, and monitoring wristbands are sequentially acquired, standardized, time-aligned, and calibrated to form an activity state sequence according to time slice nodes.

[0042] It should be noted that, Figure 2 This is merely a schematic diagram to help understand the processing flow of multi-source state data and the data organization method of activity state sequences, and does not represent the precise working state in actual operation.

[0043] Furthermore, the trusted calibration module determines the perceived trustworthiness based on the operating status of the sensing device, constructs an individual state baseline by combining historical activity patterns, corrects deviations in the activity state sequence, and forms a trusted state feature set, including: The system receives the activity status sequence output by the status acquisition module and parses the encapsulated activity status subset and running status subset at each time slice node.

[0044] Data within the two subsets are dimensionally split according to the device identifier of the sensing device. All data from the same sensing device under continuous time slice nodes are sorted and collected according to the standard collection timestamp to generate a single-device time-series status dataset for the corresponding sensing device. Each single-device time-series status dataset contains a one-to-one sequence of activity status data and a sequence of running status data, and the data are bound and matched through a globally unique data identifier.

[0045] A sliding statistical window with a length of 30 seconds is set up to traverse the time-series status dataset of a single device along the time axis with a step size of 1 second, providing a unified statistical unit for subsequent credibility index calculation. The 30-second window length is set according to the granularity of changes in human activity status in the elderly care scenario, taking into account both the real-time performance and statistical stability of credibility assessment, and avoiding jumps in credibility results caused by fluctuations in single-frame data.

[0046] For each sliding statistical window, the number of valid data frames and the theoretical number of data frames in the single device time-series status dataset within the statistical window are calculated. Valid data frames refer to status data frames with complete data identification, continuous standard acquisition timestamps, and non-empty fields. The theoretical number of data frames is the ratio of the sliding window length to the time step, i.e., 30 frames.

[0047] The data continuity of the sensing device in the current window is calculated using the ratio method. The calculation formula is as follows: , in, The value represents the data continuity and ranges from [0,1]. The number of valid data frames within the window. This represents the theoretical number of data frames within the window.

[0048] Data continuity is used to characterize the integrity of data transmission from sensing devices. A higher value indicates a lower packet loss rate and better data acquisition continuity.

[0049] Extract all received signal strength indication values ​​within the corresponding sliding window from the operating status data sequence, and calculate the standard deviation of this set of values. The unit is dBm.

[0050] The signal stability of the sensing device in the current window is obtained based on the signal strength standard deviation mapping, and the calculation formula is as follows: , in, The value is for signal stability and ranges from [0,1]. The permissible threshold for signal fluctuation is set at 10 dBm. This threshold is set based on the normal propagation characteristics of indoor Bluetooth signals. Under normal circumstances, the fluctuation of Bluetooth signal strength usually does not exceed 5 dBm. 10 dBm corresponds to the critical state of severe signal obstruction or interference. If this threshold is exceeded, the signal is considered to be completely unstable.

[0051] Signal stability is used to characterize the wireless communication quality of sensing devices. The higher the value, the more stable the signal transmission and the less the data is affected by environmental interference.

[0052] For the same activity status data of the same time slice node and the same monitored object, all valid data items output by different sensing devices are extracted; the location dimension is selected from the real-time location coordinates calculated by each sensing source, and the body posture and static duration dimensions are based on the data of wearable monitoring wristband.

[0053] For the location dimension, calculate the average of the pairwise Euclidean distances between all valid real-time location coordinates. The cross-source consistency degree of the location dimension is obtained based on the average distance mapping, and the calculation formula is as follows: , in, The allowable threshold for position deviation is set at 1.5m. This threshold is set based on the maximum reasonable error range of the indoor Bluetooth positioning system. If the deviation exceeds 1.5m, it is determined that there is a significant discrepancy between the multi-source position data, and the consistency fails.

[0054] For body posture and static duration dimensions with only a single sensing source, the cross-source consistency is set to a preset benchmark value of 0.8. This benchmark value is set based on the conventional accuracy level of single sensor posture detection and can be dynamically adjusted through event confirmation results.

[0055] The arithmetic mean of the cross-source consistency scores across all dimensions is taken as the cross-source consistency score for that sensing device. The value range is [0,1].

[0056] A weighted summation method is used to fuse data continuity, signal stability, and cross-source consistency to obtain the perception reliability of the corresponding sensing device. The calculation formula is: , in, , , These are the weighting coefficients for the three indicators, respectively, set to 0.4, 0.3, and 0.3, with a sum of 1. The weight allocation is based on the degree of influence of each indicator on data reliability: data continuity is the fundamental prerequisite for data reliability and is assigned the highest weight; signal stability and cross-source consistency reflect data reliability from the dimensions of equipment operating status and multi-source cross-validation, respectively, and are assigned balanced weights.

[0057] The calculated perception credibility is bound to the device identifier of the corresponding perception device and transmitted synchronously with the status data to complete the quantitative evaluation of the credibility of each perception device.

[0058] The system retrieves historical activity data of the monitored individual within a preset baseline period, removing data corresponding to confirmed abnormal events. The baseline period is preset to 7 days, based on the weekly cycle of human daily routines, which can fully cover the individual's entire cycle of daily activities, including living, working, and resting, while also having a moderate dynamic update capability to adapt to the slow changes in the elderly's routines.

[0059] The daily time is divided into 24 consecutive 1-hour time slots according to the natural day sequence. Each time slot corresponds to a fixed hourly time interval, and the intervals are continuous and non-overlapping. For each time slot, all valid historical activity status data within that time slot are spatially grouped according to the monitoring area identifier, resulting in a subset of historical data for each monitoring area. Each subset of historical data contains multiple valid historical data frames sorted by standard collection timestamps. Each data frame carries three core attributes: real-time location coordinates, body posture, and duration of stillness. Feature statistics are performed on each subset of historical data for each time slot to generate three baseline feature parameters for the corresponding time slot and monitoring area. The specific generation process is as follows: The first is the baseline location distribution interval. Real-time location coordinates of all valid data frames in the historical data subset for that period are extracted to form a location coordinate sample set. The centroid coordinates of the location coordinate sample set are calculated, where the x-coordinate of the centroid is the arithmetic mean of the x-coordinates of all samples, and the y-coordinate is the arithmetic mean of the y-coordinates of all samples. The Euclidean distance from each location sample point to the centroid is calculated point by point, forming a distance sample set. The 95th percentile value of the distance sample set is taken as the baseline distribution radius. The 95th percentile is set based on the location fluctuation characteristics of daily activities, which can cover the vast majority of normal activity points while excluding interference from a very small number of accidental offset points. A circular baseline location distribution interval is formed with the centroid as the center and the baseline distribution radius as the range. This interval represents the normal activity range of the monitored object in the corresponding time period and area; exceeding this interval is considered a deviation in location dimension.

[0060] Secondly, there are the baseline probabilities of each body posture. All valid data frames in the historical data subset for that period are categorized and counted according to body posture type. The number of data frames for each of the four common postures—standing, sitting, lying down, and walking—is counted. Falling postures, being considered abnormal events, are excluded from the valid historical data, and their default baseline probability value is 0. The number of frames for each posture is divided by the total number of valid data frames in the historical data subset for that period to obtain the baseline probability value for that posture. The sum of the baseline probability values ​​for all postures is 1. This baseline value represents the distribution pattern of the monitored subject's normal postures within the corresponding time period and area. If the proportion of a certain posture deviates significantly from the baseline value, it is determined that an anomaly has occurred in the posture dimension.

[0061] Thirdly, there is the baseline threshold for the duration of stillness. The historical data subset for that period is traversed in chronological order of the standard data collection timestamps. All continuous stillness segments are extracted, defined as time segments within consecutive data frames where the body posture category does not change. The total duration of each continuous stillness segment is recorded, forming a stillness duration sample set. The 90th percentile value of this stillness duration sample set is taken as the baseline threshold for the duration of stillness in that period and region. The 90th percentile is set based on the distribution characteristics of stillness duration in elderly care scenarios, covering the vast majority of normal stillness behaviors. Stillness durations exceeding this threshold significantly deviate from daily patterns, indicating anomalies. This threshold represents the upper limit of normal duration for a monitored subject to maintain the same posture within the corresponding period and region; exceeding this threshold is considered a deviation in the stillness duration dimension.

[0062] The baseline characteristic parameters of all time periods and all monitored areas are integrated to form an individual status baseline that is dynamically adapted according to time period. This baseline is stored in a local baseline library and is automatically updated with the historical data of the past 7 days every morning to achieve dynamic iteration of the baseline.

[0063] The real-time activity status data of the current period is compared with the individual status baseline of the corresponding period and the corresponding monitoring area, and the original deviation of three dimensions is calculated respectively: the original deviation of position is the ratio of the Euclidean distance between the center of the distribution interval of the real-time position and the baseline position to the interval radius; the original deviation of attitude is 1 minus the occurrence probability baseline value of the real-time attitude; the original deviation of stillness duration is the ratio of the difference between the real-time stillness duration and the stillness duration baseline threshold.

[0064] The original deviations of each dimension are weighted and corrected using perceived credibility to obtain the corrected deviations. The correction formula is as follows: , in, For the corrected deviation, This represents the original deviation. This correction method can reduce the impact of low-reliability data on state deviation judgment and avoid misjudgments of deviation due to equipment failure or signal interference.

[0065] The corrected deviations of each dimension, along with the corresponding standard collection timestamps, monitoring area identifiers, sensing device identifiers, and sensing credibility, are structured and encapsulated, arranged in chronological order to form a set of credible state features, and then output to the risk identification module.

[0066] refer to Figure 3 , Figure 3 This is a schematic diagram of the process of generating a trustworthy state feature set.

[0067] Figure 3This paper demonstrates the process of determining perceived credibility based on data continuity, signal stability, and cross-source consistency; constructing an individual state baseline based on historical activity data; and using perceived credibility to correct the deviation of the real-time state from the individual state baseline, thereby forming a credible state feature set.

[0068] It should be noted that, Figure 3 This is merely a schematic diagram to help understand the processing relationship between perceived credibility, individual state baseline, state deviation, and credible state feature set, and does not represent the precise working state in actual operation.

[0069] Furthermore, the risk identification module analyzes the temporal evolution relationship of position, attitude, and static state based on a reliable state feature set, identifies composite risk events formed by the continuous superposition of low-intensity anomalies, and determines the risk confidence level, including: The system receives the trusted state feature set output by the trusted calibration module, and analyzes the corrected deviation of the encapsulated position dimension, posture dimension, and static duration dimension node by node, as well as the corresponding bound standard acquisition timestamp, real-time position coordinates, body posture, static duration, monitoring area identifier, sensing device identifier, and sensing trustworthiness. All data items are matched one-to-one through globally unique data identifiers, and are consistent with the data structure in the trusted state feature set generation stage.

[0070] A sliding analysis window with a length of 300 seconds is set, and the set of credible state features is traversed along the time axis with a step size of 60 seconds. Each sliding analysis window contains 300 consecutive time slice nodes, which serve as the basic analysis unit for composite risk identification. The 300-second window length is set according to the evolution cycle of typical abnormal behaviors in elderly care scenarios, which can fully cover the gradual change process of states such as sitting up and slow fall prodromal period, while avoiding the introduction of interference from irrelevant daily activities due to an excessively large window; the 60-second step size takes into account the response speed and computational efficiency of risk identification.

[0071] For each sliding analysis window, all time slice nodes within the window are mapped to state nodes according to the order of standard acquisition timestamps. Each state node carries the real-time position coordinates, body posture, duration of stillness, and deviation after three-dimensional correction at the corresponding moment.

[0072] Calculate the state transition features between adjacent state nodes, specifically including position displacement, attitude transition type, and static duration increment; where position displacement is the Euclidean distance between the real-time position coordinates of adjacent nodes, attitude transition type is the category switching relationship of the body attitude of adjacent nodes, and static duration increment is the difference in the static duration of adjacent nodes.

[0073] Using state nodes as vertices and state transition features as edges, a continuous temporal activity state evolution link is constructed to form an activity state evolution relationship, which is used to characterize the state change process and magnitude of the monitored object within the analysis window.

[0074] First, calculate the overall deviation of each state node, and then use a weighted summation method to fuse the deviations after three-dimensional correction. The calculation formula is as follows: , in, For the overall deviation, , , These are the corrected deviations for the position dimension, attitude dimension, and duration of stillness dimension, respectively. , , These are the weight coefficients for the corresponding dimensions, respectively, set to 0.3, 0.4, and 0.3, with a sum of 1. The weight allocation is based on the contribution of each dimension to safety risks in the elderly care scenario: body posture and duration of stillness are directly related to core risks such as falls and lingering, and are assigned higher weights; the location dimension helps determine regional risks and is assigned balanced weights.

[0075] The thresholds for low-intensity anomalies are set at 0.3 and for strong anomalies at 0.7. The thresholds are set according to the deviation of the individual's baseline: a comprehensive deviation of less than 0.3 corresponds to normal daily activity fluctuations, and a deviation in the range of [0.3, 0.7) corresponds to low-intensity anomalies, which are minor anomalies that are not enough to trigger a risk alarm at any given moment; a deviation of 0.7 or greater corresponds to strong anomalies, which are significant single anomaly events.

[0076] Traverse all state nodes within the sliding analysis window, mark nodes that meet the low-intensity anomaly conditions, and count the total number of low-intensity anomaly nodes, the maximum length of consecutive low-intensity anomaly segments, and the number and combination types of dimensions involved in the anomaly to form an anomaly cumulative feature.

[0077] A composite risk assessment condition is set. When the sliding analysis window simultaneously meets two core assessment conditions, a composite risk event is determined to exist within that window. The first condition is that the proportion of low-intensity abnormal nodes within the window is ≥40%, meaning the ratio of the number of low-intensity abnormal nodes to the total number of nodes within the window is not less than 0.4. This threshold is set based on the proportion of abnormal fluctuations in normal activities in elderly care scenarios. The proportion of low-intensity abnormal fluctuations in normal daily activities is usually less than 20%, and 40% corresponds to a persistent cumulative effect of abnormal states. The second condition is that the low-intensity anomalies within the window involve at least two different dimensions, meaning the anomaly is not a single-dimensional, accidental fluctuation, but rather the combined effect of multiple-dimensional anomalies.

[0078] When a compound risk event is identified, its core attributes are determined. Risk types are mapped based on the combination of anomalies with the highest percentage within the window, specifically categorized into three types: sedentary-to-stand risk, slow fall risk, and regional stagnation risk. Specifically, sedentary-to-stand risk is characterized by a predominant deviation in static duration accompanied by an abnormal shift from a sitting to a standing posture; slow fall risk is characterized by a predominant deviation in posture accompanied by a continuously decreasing displacement; and regional stagnation risk is characterized by a predominant deviation in position accompanied by a continuously increasing static duration. The occurrence area is identified as the monitored area with the highest concentration of low-intensity anomaly nodes within the window.

[0079] Based on the cumulative characteristics of anomalies and perceived credibility, the risk confidence level corresponding to a composite risk event is calculated using the following formula: , in, The risk confidence level has a value range of [0,1]. The percentage of low-intensity abnormal nodes within the window; These are the coefficients for outlier dimensions, with values ​​of 0.5 for one dimension, 0.8 for two dimensions, and 1.0 for three dimensions. This is the arithmetic mean of the perception reliability of the sensing devices corresponding to all abnormal nodes within the window.

[0080] This confidence calculation method integrates the significance of anomalies, the degree of multi-dimensional superposition, and the credibility of the data source, which can effectively reduce the probability of false alarms caused by low-confidence data.

[0081] The risk type, occurrence area, and risk confidence level of the complex risk event, along with the start and end timestamps of the corresponding analysis window, are structured and encapsulated, and then output to the monitoring and handling module.

[0082] Furthermore, the monitoring and handling module invokes corresponding sensing devices for complementary verification based on the risk confidence level, generates and executes tiered monitoring instructions based on the verification results, and uses the event confirmation results to update the sensing confidence level and individual status baseline, including: Receive composite risk events output by the risk identification module, and analyze the risk type, occurrence area, risk confidence level, and start and end timestamps of the analysis window.

[0083] Two confidence level thresholds are set at 0.6 and 0.8, respectively, dividing the risk confidence into three levels: a risk confidence level in the range [0.4, 0.6) corresponds to low risk, [0.6, 0.8) corresponds to medium risk, and [0.8, 1.0] corresponds to high risk. These thresholds are set based on the alarm tolerance requirements of elderly care monitoring scenarios. The low level focuses on reducing false alarm interference, the medium level balances response speed and accuracy, and the high level prioritizes safety.

[0084] Based on the risk level, a complementary verification strategy is matched accordingly: At a low risk level, all Bluetooth gateways within the affected area are invoked to increase the scanning frequency, adjusting the regular scanning cycle to 100ms, while simultaneously expanding the Bluetooth beacon matching range and increasing the number of beacons involved in the positioning calculation; the verification duration is set to 60s. At a medium risk level, based on the low-risk verification strategy, a high-precision acquisition command is sent to the wearable monitoring wristband worn by the monitored individual, adjusting the posture sampling cycle from 200ms to 100ms, and simultaneously acquiring higher temporal resolution body posture and static duration data; the verification duration is set to 30s. At a high risk level, a full verification strategy is executed. Based on the medium-risk verification strategy, the vibration alert of the wearable monitoring wristband is triggered, and the response feedback status of the monitored individual is collected; the verification duration is set to 15s.

[0085] According to the matching verification strategy, verification acquisition instructions are issued to the corresponding sensing devices to obtain activity status data and operation status data within the verification period; the verification data is subjected to standardization processing, time alignment and spatial calibration operations consistent with the status acquisition module to generate a verification status sequence and calculate the comprehensive deviation degree corresponding to each time slice node within the verification period.

[0086] The authenticity of an event is determined based on the review status sequence and the trend of changes in the comprehensive deviation. If the comprehensive deviation remains above the low-intensity anomaly threshold of 0.3 during the review period, and the combination of anomaly dimensions is consistent with the original composite risk event, the event is considered confirmed. If the comprehensive deviation falls below 0.3 during the review period, or the number of anomaly dimensions decreases to one or less, the event is considered invalid and is a false alarm. If the comprehensive deviation rises above the strong anomaly threshold of 0.7 during the review period, the event is considered escalated.

[0087] Based on the determination of the authenticity of the event and the corresponding risk level, a corresponding level of monitoring instruction is generated and issued for execution.

[0088] If the event is determined to be invalid, a risk clearance instruction is generated to terminate the current review process, restore all participating sensing devices to their normal sampling working state, and simultaneously record the associated sensing devices, triggering conditions, and time period of this false alarm event.

[0089] If a low-risk event is confirmed, a mild attention instruction is generated, and an abnormal activity alert is pushed to the monitoring terminal, marking the area where the event occurred, the risk type, and the start time. The monitoring device continuously monitors the activity status of the monitored person for the next 10 minutes, and maintains the verification and sampling status until the status returns to the normal range.

[0090] If a medium-risk event is confirmed, an early warning and response instruction is generated, and an early warning message is pushed to the monitoring personnel's terminal, including the event location, risk type, duration, and real-time status data. At the same time, a gentle vibration reminder is triggered on the wearable monitoring wristband to prompt the monitored person to adjust their state and pay attention to safety.

[0091] If a high-risk event is confirmed or the event escalates, an emergency alarm command is generated, and alarm information is simultaneously pushed to the monitoring personnel and preset emergency contacts, along with the location of the event, real-time status data and risk level. At the same time, the monitoring terminal is linked to initiate a voice call to help the monitoring personnel quickly confirm the on-site situation and, if necessary, trigger the emergency rescue process.

[0092] The perception reliability of the corresponding sensing device is updated based on the event confirmation result. If the event is confirmed and the verification data of the corresponding sensing device is consistent with the event judgment result, the perception reliability is updated using a positive adjustment formula: , in, For the updated perceived credibility, To assess the perceived credibility before the update. To adjust the step size in the positive direction, we take 0.05.

[0093] If the event is confirmed as a false alarm, and the false alarm is directly caused by the data deviation of the corresponding sensing device, the sensing confidence level is updated using a negative adjustment formula: , in, The adjustment step size is set to 0.1 for negative adjustment.

[0094] This adjustment mechanism is set according to the reward and punishment logic of device performance. Positive adjustments are made in small increments to avoid excessive fluctuations in credibility, while negative adjustments are made in large increments to quickly reduce the weight of unreliable devices. The step size balances the stability of credibility and the efficiency of adaptive adjustment. The updated perceived credibility is synchronously written into the device state database for subsequent credibility calculations and state deviation corrections.

[0095] The historical dataset corresponding to the individual status baseline is updated synchronously: for confirmed genuine abnormal events, the data for the corresponding time period is marked as abnormal data and continues to be excluded from the baseline statistical sample; for confirmed false alarm events, the data for the corresponding time period is marked as normal data and included in the baseline statistical sample. In the daily baseline rolling update phase at midnight, the baseline characteristic parameters for each time period and each monitored area are recalculated using the updated historical dataset to complete the dynamic iteration of the individual status baseline.

[0096] refer to Figure 4 , Figure 4 This is a schematic diagram of the process for identifying and classifying complex risk events.

[0097] Figure 4 This paper demonstrates the process of constructing activity state evolution relationships based on a set of credible state features, identifying complex risk events formed by the accumulation of low-intensity anomalies, determining risk attributes and risk levels, and completing event authenticity determination, generation of hierarchical monitoring instructions, and feedback updates through multi-source complementary verification.

[0098] It should be noted that, Figure 4 This is merely a schematic diagram to help understand the process relationships between the identification of complex risk events, risk classification, complementary review, monitoring and handling, and feedback updates. It does not represent the precise working status in actual operation.

[0099] This invention discloses a safety monitoring system based on multi-source state perception. By performing spatiotemporal correlation calibration on the activity state of the monitored object and the operating state of the sensing device, and combining the perception credibility with the individual state baseline to form a credible state feature set, the system identifies composite risk events formed by the superposition of low-intensity anomalies based on the continuous evolution of multi-dimensional activity states. Through complementary verification, hierarchical monitoring and feedback update mechanisms, the system achieves accurate identification and adaptive monitoring of progressive safety risks in elderly care scenarios.

[0100] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

[0101] If the technical solution disclosed herein involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution disclosed herein involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information. Alternatively, on the personal information processing device, the personal information processing rules are clearly indicated by signs / information, and authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information. The personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

Claims

1. A safety monitoring system based on multi-source state perception, characterized in that, include: The status acquisition module acquires the activity status of the monitored object and the operating status of the sensing device, and forms an activity status sequence through time alignment. The trusted calibration module determines the perceived trustworthiness based on the operating status of the sensing device, constructs an individual state baseline by combining historical activity patterns, corrects the deviation of the activity state sequence, and forms a trusted state feature set. The risk identification module analyzes the temporal evolution relationship of position, attitude and static state based on the trusted state feature set, identifies composite risk events formed by the continuous superposition of low-intensity anomalies, and determines the risk confidence level. The monitoring and handling module calls the corresponding sensing devices for complementary verification based on the risk confidence level, generates and executes hierarchical monitoring instructions based on the verification results, and updates the sensing confidence level and individual status baseline. The determination of perceived credibility includes: The activity status sequence is split according to the device identifier of the sensing device, and the activity status data and operation status data under the continuous time slice node are sorted according to the standard collection timestamp to form a single device time-series status dataset. Within the sliding statistical window, data continuity is determined based on the number of valid data frames and the theoretical number of data frames; signal stability is determined based on the dispersion of the received signal strength indication value; and cross-source consistency is determined based on the cross-source deviation of similar activity state data within the same time slice node. A weighted summation method is used to fuse data continuity, signal stability, and cross-source consistency to obtain the perception reliability bound to the device identifier of the sensing device; The updates to perceived credibility and individual state baselines include: When an event is confirmed and the verification data from the corresponding sensing device matches the event determination result, the sensing reliability is updated using a positive adjustment formula: , in, For the updated perceived credibility, To assess the perceived credibility before the update. To adjust the step size positively, we set it to 0.05; When an event is confirmed as a false alarm and the false alarm is caused by data deviation of the corresponding sensing device, the sensing confidence level is updated using a negative adjustment formula: , in, The negative adjustment step size is set to 0.1; Write the updated perception credibility into the device status database, mark the data corresponding to the time period of real abnormal events as abnormal data, and mark the data corresponding to the time period of false alarm events as normal data; Based on the updated historical dataset, the baseline characteristic parameters for each time period and each monitored area are recalculated to complete the dynamic iteration of the individual status baseline.

2. The safety monitoring system based on multi-source state perception as described in claim 1, characterized in that, The construction of individual status baselines and the correction of deviations include: Retrieve historical activity status data within a preset baseline period and remove data for time periods corresponding to abnormal events; Based on the statistical baseline location distribution interval of time period and monitored area, the baseline value of the probability of body posture and the baseline threshold of static duration, an individual status baseline is formed; The real-time location coordinates, body posture, and duration of stillness are compared with the individual status baseline of the corresponding time period and the corresponding monitoring area to obtain the original deviation of location, original deviation of posture, and original deviation of stillness duration. The original deviations are corrected by using the perception credibility of the corresponding sensing device to form a credible state feature set.

3. The safety monitoring system based on multi-source state perception as described in claim 1, characterized in that, The identification of complex risk events includes: A sliding analysis window is used to traverse the trusted state feature set, and the data in the window is mapped to state nodes according to the standard collection timestamp. Calculate the position displacement, attitude transition type, and static duration increment of adjacent state nodes, and construct the active state evolution relationship with state nodes as vertices and state transition features as edges; The comprehensive deviation is obtained by integrating the corrected deviations of position dimension, attitude dimension and static duration dimension, and low-intensity abnormal nodes are marked to form anomaly accumulation features; Composite risk events are identified based on the proportion of low-intensity abnormal nodes and the number of dimensions involved in the anomalies, and the risk type, occurrence area, and risk confidence level of the composite risk events are determined.

4. The safety monitoring system based on multi-source state awareness as described in claim 3, characterized in that, The determination of risk type, location of occurrence, and risk confidence level includes: Based on the combination of the most prevalent abnormal dimensions within the sliding analysis window, the composite risk event is identified as a sedentary-to-get-up risk, a slow fall risk, or a regional stagnation risk. The monitoring area with the highest concentration of low-intensity abnormal nodes was identified as the occurrence area; The significance of the anomaly is determined by the proportion of low-intensity anomaly nodes, and the anomaly dimension coefficient is determined by the number of dimensions involved in the anomaly. The risk confidence level is obtained by integrating the proportion of low-intensity abnormal nodes, the abnormality dimension coefficient, and the perception credibility of the sensing devices corresponding to low-intensity abnormal nodes.

5. The safety monitoring system based on multi-source state perception as described in claim 1, characterized in that, The formation of the activity state sequence includes: Acquire real-time location coordinates, body posture, duration of stillness, and operating status of sensing devices, and add local acquisition timestamps and monitoring area identifiers to the collected data; Extract the activity status field and the running status field according to the unified data field specification to form a standardized status dataset; The local sampling clock offset is determined based on the issuance time of the synchronous acquisition command and the local clock time of the sensing device, the initial timestamp is corrected, and the status data is collected to the time slice node. Spatial grouping is performed according to the monitoring area identification, and repeated location data and location data with overlapping sensing coverage are calibrated. The calibrated time slice nodes are spliced ​​together in time slice order to form an activity state sequence.

6. The safety monitoring system based on multi-source state awareness as described in claim 5, characterized in that, Real-time location coordinates, body posture, and duration of stillness are acquired, including: The beacon is located via Bluetooth and broadcasts its identification and fixed coordinates. The wearable monitoring wristband scans the broadcast signal and records the received signal strength indicator value. The Bluetooth gateway then forwards the scan results. The distance between the wearable monitoring wristband and the Bluetooth positioning beacon is calculated based on the logarithmic distance path loss model, and the real-time location coordinates are calculated based on the distance. Acceleration and angular velocity data are collected, and a complementary filtering algorithm is used to obtain pitch and roll angles. Body posture is determined based on the attitude angle threshold range. The cumulative duration of static posture without category switching is calculated to obtain the duration of stillness.

7. The safety monitoring system based on multi-source state perception as described in claim 1, characterized in that, Based on the risk confidence level, corresponding sensing devices are invoked for complementary verification, including: Risk levels are categorized based on risk confidence levels, and complementary review strategies are matched for different risk levels. In low-risk areas, the Bluetooth gateway in the affected area is invoked to increase the scanning frequency and expand the Bluetooth beacon matching range; Under the medium-risk level, a complementary review strategy corresponding to the low-risk level is implemented, and high-precision collection commands are sent to the wearable monitoring wristband to obtain data on body posture and duration of stillness. Under high-risk levels, a full review strategy is implemented. Based on the complementary review strategy for medium-risk levels, the vibration of the wearable monitoring wristband is triggered to collect the response feedback status of the monitored object.

8. The safety monitoring system based on multi-source state perception as described in claim 1, characterized in that, The determination of the review results and the generation of graded monitoring instructions include: The activity status data and operational status data during the review period are standardized, time-aligned, and spatially calibrated to form a review status sequence. Calculate the overall deviation of each time slice node in the verification state sequence; Based on the trend of changes in the overall deviation and the consistency between the combination of abnormal dimensions and the original composite risk event, it is determined whether the event is confirmed, not confirmed, or escalated. Based on the determination of the authenticity of the event and the risk level, generate risk relief instructions, mild attention instructions, early warning and handling instructions, or emergency alarm instructions, and issue them to the corresponding monitoring terminals for execution.

Citation Information

Patent Citations

  • Remote fall detection and determination system using an image collection device

    KR102364424B1

  • KR20260100242A