An old-age robot system based on multi-dimensional sensing intelligent monitoring

Abstract

The application discloses a kind of old-age robot systems based on multi-dimensional sensing wisdom monitoring, specifically related to multi-dimensional sensing and extended reality technical field, including data construction module, feature formation module, difference identification module, counter-evidence generation module and intervention output module, data construction module is gathered by heart rate sensor, blood oxygen sensor, respiratory frequency sensor, gait capture device and environment detection device corresponding heart rate signal, blood oxygen signal, breathing signal, gait signal and environment signal of old people, all signals collected are executed correction and alignment under unified time reference, and output fusion data;The application is fused by multi-dimensional sensing signal and the feeling of old people's self-description under unified time reference, difference identification and multi-sensory counter-evidence generation, and the result is output in extended reality interface, to solve the problem that abnormal signal is neglected and interface and real feeling mismatch.

Description

Technical Field

[0001] This invention relates to the field of multidimensional sensing and extended reality technology, and more specifically, to an elderly care robot system based on multidimensional sensing intelligent monitoring. Background Technology

[0002] In current smart elderly care practices, elderly care robots are increasingly relying on multi-dimensional sensing and extended reality technologies to complete health monitoring and risk warning. The operating logic of existing systems is roughly as follows: first, data on the elderly’s body and environment are collected through various sensors; then, after preprocessing and fusion, the complex data is simplified into a few intuitive indicators; finally, the indicators are presented in the extended reality interface in the form of virtual overlay, so that family members or caregivers can quickly obtain the elderly’s overall health status in the virtual scene.

[0003] However, this process harbors risks that accumulate layer by layer. First, to avoid too many false alarms, data is often smoothed or set with high thresholds during the calculation phase, and some brief but important abnormal signals may be ignored as noise. Second, the presentation of extended reality emphasizes immersion and authority, and users are likely to believe the stable state displayed on the interface, ignoring the elderly person's own subjective feelings. Furthermore, when the virtual interface shows everything is normal, but the elderly person experiences chest tightness, palpitations, or slight gait abnormalities, caregivers are more likely to believe the interface's conclusions rather than the patient's self-description.

[0004] A step-by-step analysis reveals that the root of the problem lies not in the sensors themselves, but in the fact that the augmented reality interface over-determines uncertain data. This approach masks transient but potentially fatal abnormal signals, creating a mismatch between the elderly’s real feelings and the virtual presentation. Therefore, augmented reality reinforces the cognitive bias of interface priority in health monitoring, causing clinicians and nurses to rely more on virtual images when faced with conflicts, thus delaying the identification and handling of real health crises. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an elderly care robot system based on multi-dimensional sensing intelligent monitoring. This system integrates multi-dimensional sensing signals with the elderly’s self-reported feelings under a unified time reference, identifies differences, generates multi-sensory counter-evidence, and outputs the results in an extended reality interface. This addresses the problems of ignoring abnormal signals and mismatch between the interface and real feelings in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an elderly care robot system based on multi-dimensional sensing intelligent monitoring, comprising a data construction module, a feature formation module, a difference recognition module, a counter-evidence generation module, and an intervention output module;

[0007] The data construction module collects the heart rate, blood oxygen, respiratory rate, gait, and environmental signals of the elderly through heart rate sensors, blood oxygen sensors, respiratory rate sensors, gait capture devices, and environmental detection devices. All collected signals are corrected and aligned under a unified time reference, and fused data is output.

[0008] The feature formation module performs noise suppression and time segmentation on the fused data, performs difference operations on adjacent time slices and combines them with weight calculations to output a state vector;

[0009] The difference recognition module converts the elderly’s self-reported physical sensations into sensation data through voice recognition and interactive input. It encodes the sensation data into a sensation vector and compares it element by element with the state vector to identify the features that differ from each other and count their frequency of occurrence, and then outputs the difference data.

[0010] The disproven generation module extracts the differential data into multi-sensory disproven parameters, generates a first disproven layer of virtual images reflecting gait abnormalities, a second disproven layer of tactile signals reflecting respiratory load, and a third disproven layer of audio signals reflecting cardiac arrhythmia. The three types of disproven layers are then superimposed in the extended reality interface to output the disproven structure.

[0011] The intervention output module transmits the counter-evidence structure to nurses and telemedicine personnel, solves the intervention priority based on the difference in strength, and outputs intervention instructions, enabling nurses to perform review and treatment according to the counter-evidence structure.

[0012] In a preferred embodiment, the data construction module further includes attaching a unified timestamp to the collected heart rate signal, blood oxygen signal, respiratory signal, gait signal and environmental signal and constructing an original signal matrix;

[0013] The average number of sampled values ​​of various signals is calculated within the reference time window of the original signal matrix. The difference between the average number of sampled values ​​and the preset benchmark is calculated to form the average offset. The average offset is then subtracted from each sampled value to output the correction matrix.

[0014] The timestamp of the correction matrix is ​​compared with the unified time reference to solve the drift vector of various signals on the target time axis. When the drift is less than the threshold, interpolation is performed to complete it. When the drift is greater than the threshold, sliding window phase alignment is performed. During the alignment process, the weighting factor is dynamically adjusted according to the drift magnitude, and the alignment matrix is ​​output.

[0015] Synchronization discrimination is performed on the alignment matrix within a unified reference time window. First, the time interval difference between corresponding sampling points of heart rate signal, blood oxygen signal, respiratory signal, gait signal and environmental signal on a unified target time axis is calculated. When the time interval difference is less than a preset threshold, the fusion operation is performed directly. When the time interval difference is greater than or equal to the preset threshold, time slice interpolation correction is performed on the corresponding signals and they are re-weighted and fused, and the fused data is output.

[0016] In a preferred embodiment, the feature forming module further includes acquiring heart rate signal, blood oxygen signal, respiratory signal, gait signal and environmental signal from the fused data at a fixed sampling interval under a unified time reference, and arranging the various signals in chronological order to construct a first feature matrix;

[0017] The first feature matrix is ​​divided into continuous time segments, and the sum of squares of various signals is performed in each time segment to calculate the energy value. At the same time, the upper and lower limits of the values ​​of each type of signal in the time segment are solved and the difference is calculated to form the second feature matrix. Signals that exceed the preset threshold in the second feature matrix are marked as abnormal signals.

[0018] The second feature matrix is ​​subjected to difference calculation between adjacent time segments to solve the ratio of the difference of the average value of each type of signal in adjacent time segments to the length of the time segment, forming a difference matrix. In the difference matrix, signals that exceed the preset threshold are given a correction weight, while signals that do not exceed the preset threshold are kept with the original weight.

[0019] The various signals of the difference matrix are weighted and integrated to construct a weighted matrix. When the rate of change in the weighted matrix continuously exceeds the preset stability threshold, a state vector is generated. When the rate of change in the weighted matrix does not continuously exceed the preset stability threshold, the rate of change of the weighted matrix is ​​continuously counted within a preset number of time segments, and a state vector is generated when the cumulative result meets the continuity condition.

[0020] In a preferred embodiment, the difference recognition module further includes converting the elderly person's self-reported physical sensations into sensation data through voice recognition processing and interactive input processing, aligning them with the timestamps of the fused data under a unified time reference, and encoding the sensation data to form a sensation matrix.

[0021] The perception matrix is ​​compared with multiple state vectors one by one in the same time segment to solve the numerical difference at each corresponding position. All numerical differences are used to construct the first difference matrix, where the rows of the first difference matrix represent time segments and the columns represent heart rate signal, blood oxygen signal, respiratory signal, gait signal and environmental signal.

[0022] Perform difference continuity determination on the first difference matrix between adjacent time segments:

[0023] When the numerical differences in the same column exceed the preset difference threshold in consecutive time segments, a continuous difference marker is formed; when the numerical differences exceed the preset difference threshold only in a single time segment, an instantaneous difference marker is formed, and the marker results are constructed into a second difference matrix.

[0024] In a preferred embodiment, the difference recognition module further includes comparing the second difference matrix with the rate of change of the state vector within the same time segment:

[0025] When the rate of change corresponding to the difference marker is lower than the lower limit, it is determined to be an invalid difference; when the rate of change corresponding to the difference marker is higher than the upper limit, it is determined to be a valid difference. The comparison results are then used to construct a third difference matrix.

[0026] Perform consistency statistics on the third difference matrix across the heart rate signal column, blood oxygen signal column, respiratory signal column, gait signal column, and environmental signal column:

[0027] When multiple columns simultaneously yield valid differences within the same time segment, difference data is generated and output. The difference data includes the signal category, time segment, and cumulative frequency of occurrence of the valid differences.

[0028] If multiple columns fail to resolve valid differences simultaneously within the same time segment, the statistical results for that time segment are recorded as not meeting the conditions, and consistency statistics are continued to be performed and output in subsequent time segments.

[0029] In a preferred embodiment, the counter-evidence generation module further includes classifying the difference data according to gait signal category, respiratory signal category and heart rate signal category, and performing normalization processing on the difference data of each category under a unified time reference to construct multi-sensory counter-evidence parameters;

[0030] The data of gait signal categories in the multi-sensory counter-evidence parameters are subjected to trajectory fitting operation within a time segment to generate a continuous gait curve, and the curve is mapped to virtual image data to form the first counter-evidence layer reflecting gait abnormalities.

[0031] The respiratory signal category data in the multi-sensory counter-evidence parameters are subjected to frequency decomposition operation within a time segment to solve the respiratory fluctuation intensity, and the intensity is converted into a tactile feedback signal to form a second counter-evidence layer reflecting respiratory load.

[0032] The heart rate signal data of the multi-sensory disproving parameter are subjected to rhythm recognition operation within a time segment to solve the heart rhythm fluctuation pattern, and the pattern is converted into an audio signal to form a third disproving layer reflecting heart rhythm abnormalities.

[0033] In a preferred embodiment, the counter-evidence generation module further includes aligning the virtual image data of the first counter-evidence layer, the tactile feedback data of the second counter-evidence layer, and the audio signal data of the third counter-evidence layer according to the timestamps of a unified time segment in the extended reality interface, and superimposing them in the same display area, and then performing the following multi-level judgments in sequence:

[0034] When the numerical difference of the gait signal category in the multi-sensory counter-evidence parameters in the current time segment is less than the preset gait difference threshold, and the numerical difference of the respiratory signal category in the multi-sensory counter-evidence parameters in the current time segment is less than the preset respiratory difference threshold, and the numerical difference of the heart rate signal category in the multi-sensory counter-evidence parameters in the current time segment is less than the preset heart rate difference threshold, the counter-evidence structure is directly output.

[0035] Otherwise, if the numerical difference of the gait signal category in the multi-sensory counter-evidence parameters in the current time segment is greater than the preset gait difference threshold, and the numerical difference of the respiratory signal category in the multi-sensory counter-evidence parameters in the current time segment is greater than the preset respiratory difference threshold, interpolation correction is performed and the counter-evidence structure is output after the correction is completed.

[0036] Otherwise, the multi-sensory counter-evidence parameters of the current time segment are marked as multi-sensory counter-evidence parameters that do not meet the conditions, and the process of aligning with the timestamps of the unified time segment and superimposing them in the same display area continues in subsequent time segments.

[0037] In a preferred embodiment, the intervention output module further includes decomposing the counter-evidence structure into gait difference data, respiratory difference data and heart rate difference data under a unified time reference, and performing amplitude calculation on the three types of difference data in each time segment to construct difference intensity data;

[0038] The cumulative statistical analysis of the difference intensity data is performed between time segments to obtain the cumulative difference intensity result. The cumulative difference intensity result is then compared with the gait intervention threshold, respiratory intervention threshold, and heart rate intervention threshold to form the intervention priority.

[0039] The intervention priorities are combined in a hierarchical manner within a unified time segment to generate intervention instructions, and hierarchical markers are added to the intervention instructions, including high priority and medium priority;

[0040] Intervention instructions are simultaneously transmitted to both nursing staff terminals and telemedicine staff terminals. When a high-priority marker is present in the intervention instruction, the telemedicine staff terminal is triggered to perform the treatment. When a medium-priority marker is present in the intervention instruction, the nursing staff terminal is triggered to perform the review and the review result is fed back to the telemedicine staff terminal.

[0041] The technical effects and advantages of this invention are as follows:

[0042] This solution constructs a difference matrix by combining the perception matrix and the state vector, and combines numerical difference and rate of change in consistency statistics to reveal transient anomalies that are masked by smoothing, thus solving the problem of mismatch between real perception caused by the priority of extended reality interface.

[0043] This scheme constructs the original signal matrix, correction matrix, and alignment matrix step by step, and controls the alignment method with drift vector and time interval difference to ensure that heart rate, blood oxygen, respiration, gait and environmental signals are synchronously fused under a unified time reference, thereby reducing time misalignment and misjudgment.

[0044] This scheme marks abnormal signals using the first and second feature matrices, and then generates a state vector using the continuity conditions of the difference matrix and the weighting matrix, suppressing sporadic noise and single-point fluctuations, making the detection results more reliable.

[0045] This solution categorizes differential data into three types: gait, respiration, and heart rate. These data are then converted into virtual images, tactile feedback, and audio signals, and displayed in an extended reality interface with timestamps aligned and overlaid. This allows caregivers to visually see, touch, and hear abnormalities.

[0046] This solution decomposes the counter-evidence structure into gait, respiration, and heart rate difference data, calculates the difference intensity and compares it with the intervention threshold to form an intervention priority, and then generates a graded intervention instruction to ensure that high priority triggers remote medical care, while medium priority is reviewed by nursing staff, thus realizing tiered intervention management. Attached Figure Description

[0047] Figure 1 This is a system module diagram of the present invention.

[0048] Figure 2 This is a flowchart illustrating the data construction process of this invention.

[0049] Figure 3 The flowchart illustrates the feature formation process of this invention.

[0050] Figure 4 This is a flowchart of the difference recognition process of the present invention.

[0051] Figure 5 This is a flowchart of the process for generating counter-evidence in this invention.

[0052] Figure 6 This is a flowchart of the intervention output of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Refer to the instruction manual appendix Figure 1-6 An embodiment of the present invention provides an elderly care robot system based on multi-dimensional sensing intelligent monitoring, comprising a data construction module, a feature formation module, a difference recognition module, a counter-evidence generation module, and an intervention output module.

[0055] The data construction module collects the heart rate, blood oxygen, respiratory rate, gait, and environmental signals of the elderly through heart rate sensors, blood oxygen sensors, respiratory rate sensors, gait capture devices, and environmental detection devices. All collected signals are corrected and aligned under a unified time reference, and fused data is output.

[0056] The feature formation module performs noise suppression and time segmentation on the fused data, performs difference operations on adjacent time slices and combines them with weight calculations to output a state vector;

[0057] The difference recognition module converts the elderly’s self-reported physical sensations into sensation data through voice recognition and interactive input. It encodes the sensation data into a sensation vector and compares it element by element with the state vector to identify the features that differ from each other and count their frequency of occurrence, and then outputs the difference data.

[0058] The disproven generation module extracts the differential data into multi-sensory disproven parameters, generates a first disproven layer of virtual images reflecting gait abnormalities, a second disproven layer of tactile signals reflecting respiratory load, and a third disproven layer of audio signals reflecting cardiac arrhythmia. The three types of disproven layers are then superimposed in the extended reality interface to output the disproven structure.

[0059] The intervention output module transmits the counter-evidence structure to nurses and telemedicine personnel, solves the intervention priority based on the difference in strength, and outputs intervention instructions, enabling nurses to perform review and treatment according to the counter-evidence structure.

[0060] The data construction module also includes attaching a unified timestamp to the collected heart rate signal, blood oxygen signal, respiratory signal, gait signal and environmental signal and constructing the original signal matrix;

[0061] The average number of sampled values ​​of various signals is calculated within the reference time window of the original signal matrix. The difference between the average number of sampled values ​​and the preset benchmark is calculated to form the average offset. The average offset is then subtracted from each sampled value to output the correction matrix.

[0062] The timestamp of the correction matrix is ​​compared with the unified time reference to solve the drift vector of various signals on the target time axis. When the drift is less than the threshold, interpolation is performed to complete it. When the drift is greater than the threshold, sliding window phase alignment is performed. During the alignment process, the weighting factor is dynamically adjusted according to the drift magnitude, and the alignment matrix is ​​output.

[0063] Synchronization discrimination is performed on the alignment matrix within a unified reference time window. First, the time interval difference between corresponding sampling points of heart rate signal, blood oxygen signal, respiratory signal, gait signal and environmental signal on a unified target time axis is calculated. When the time interval difference is less than a preset threshold, the fusion operation is performed directly. When the time interval difference is greater than or equal to the preset threshold, time slice interpolation correction is performed on the corresponding signals and they are re-weighted and fused to output the fused data.

[0064] It should be noted that by progressively correcting the raw sensor data, signals from different sources are aligned on the same time axis. Specifically, a data structure such as the original signal matrix, correction matrix, drift vector, and alignment matrix is ​​used. The sampled values ​​are compared with the reference, the offset is solved, and the time difference is adjusted by interpolation or sliding window. This ensures that signals such as heart rate, blood oxygen, respiration, gait, and environment are synchronized under the same time reference, avoiding errors caused by data asynchrony and making the fused data more reliable in subsequent processing.

[0065] The feature formation module also includes acquiring heart rate, blood oxygen, respiratory, gait and environmental signals from the fused data at fixed sampling intervals under a unified time reference, and arranging the various signals in chronological order to construct the first feature matrix;

[0066] The first feature matrix is ​​divided into continuous time segments, and the sum of squares of various signals is performed in each time segment to calculate the energy value. At the same time, the upper and lower limits of the values ​​of each type of signal in the time segment are solved and the difference is calculated to form the second feature matrix. Signals that exceed the preset threshold in the second feature matrix are marked as abnormal signals.

[0067] The second feature matrix is ​​subjected to difference calculation between adjacent time segments to solve the ratio of the difference of the average value of each type of signal in adjacent time segments to the length of the time segment, forming a difference matrix. In the difference matrix, signals that exceed the preset threshold are given a correction weight, while signals that do not exceed the preset threshold are kept with the original weight.

[0068] The various signals of the difference matrix are weighted and integrated to construct a weighted matrix. When the rate of change in the weighted matrix continuously exceeds the preset stability threshold, a state vector is generated. When the rate of change in the weighted matrix does not continuously exceed the preset stability threshold, the rate of change of the weighted matrix is ​​continuously counted within a preset number of time segments, and a state vector is generated when the cumulative result meets the continuity condition.

[0069] It should be noted that by further refining features based on the aligned data, and by constructing the first feature matrix, the second feature matrix, the difference matrix, and the weighting matrix in a step-by-step manner, the original signal is transformed into energy values, upper and lower limit differences, anomaly markers, etc. Then, differential operations and weighted corrections are performed, so that the generation of the state vector is not just a simple data processing, but has multiple discrimination conditions, which can eliminate single-point anomalies and improve the stability and accuracy of the system in recognizing the state of the elderly.

[0070] The difference recognition module also includes converting the elderly’s self-reported physical sensations into sensation data through voice recognition and interactive input processing, aligning them with the timestamps of the fused data under a unified time reference, and encoding the sensation data to form a sensation matrix.

[0071] The perception matrix is ​​compared with multiple state vectors one by one in the same time segment to solve the numerical difference at each corresponding position. All numerical differences are used to construct the first difference matrix, where the rows of the first difference matrix represent time segments and the columns represent heart rate signal, blood oxygen signal, respiratory signal, gait signal and environmental signal.

[0072] Perform difference continuity determination on the first difference matrix between adjacent time segments:

[0073] When the numerical differences in the same column exceed the preset difference threshold in consecutive time segments, a continuous difference marker is formed; when the numerical differences exceed the preset difference threshold only in a single time segment, an instantaneous difference marker is formed, and the marker results are constructed into a second difference matrix.

[0074] In this process, the system accurately compares the elderly’s subjective feelings with the objective data measured by the system. By using a feeling matrix, a first difference matrix, and a second difference matrix, the feeling data obtained from speech recognition and interactive input are aligned to a unified time segment. Then, it is compared with the state vector one by one, the difference is calculated, and continuous differences and instantaneous differences are distinguished. This allows the system to combine subjective symptoms and objective signals for judgment, rather than looking at one aspect in isolation, thus making the basis for diagnosis more comprehensive.

[0075] The difference recognition module also includes comparing the rate of change of the second difference matrix with that of the state vector within the same time segment:

[0076] When the rate of change corresponding to the difference marker is lower than the lower limit, it is determined to be an invalid difference; when the rate of change corresponding to the difference marker is higher than the upper limit, it is determined to be a valid difference. The comparison results are then used to construct a third difference matrix.

[0077] The third difference matrix is ​​used to perform consistency statistics among the heart rate signal column, blood oxygen signal column, respiratory signal column, gait signal column, and environmental signal column. It should be noted that consistency statistics refer to the statistics on whether multiple columns simultaneously meet the valid difference criteria within the same time segment.

[0078] When multiple columns simultaneously yield valid differences within the same time segment, difference data is generated and output. The difference data includes the signal category, time segment, and cumulative frequency of occurrence of the valid differences.

[0079] When multiple columns fail to yield valid differences simultaneously within the same time segment, the statistical results for that time segment are recorded as not meeting the conditions, and consistency statistics are continued to be performed and output in subsequent time segments.

[0080] In this process, based on the previous difference discrimination, it is further determined whether these differences are truly valid anomalies. By comparing the rate of change of the second difference matrix, the third difference matrix, and the state vector, invalid differences are filtered out, and only true anomalies are retained. Then, consistency statistics are performed on each column of heart rate, blood oxygen, respiration, gait, and environment to see if multiple signals are abnormal at the same time. This avoids the misjudgment of a single signal's random fluctuation as a serious problem, ensuring that difference data is only output when multiple signals are consistently abnormal, thereby improving reliability.

[0081] The disproving evidence generation module also includes classifying the differential data according to gait signal category, respiratory signal category and heart rate signal category, and performing normalization processing on the differential data of each category under a unified time reference to construct multi-sensory disproving evidence parameters;

[0082] The data of gait signal categories in the multi-sensory counter-evidence parameters are subjected to trajectory fitting operation within a time segment to generate a continuous gait curve, and the curve is mapped to virtual image data to form the first counter-evidence layer reflecting gait abnormalities.

[0083] The respiratory signal category data in the multi-sensory counter-evidence parameters are subjected to frequency decomposition operation within a time segment to solve the respiratory fluctuation intensity, and the intensity is converted into a tactile feedback signal to form a second counter-evidence layer reflecting respiratory load.

[0084] The heart rate signal category data of the multi-sensory counter-evidence parameter are subjected to rhythm recognition operation within a time segment to solve the heart rhythm fluctuation pattern, and the pattern is converted into an audio signal to form a third counter-evidence layer reflecting heart rhythm abnormalities.

[0085] Furthermore, by transforming the differential data into a multimodal layer of counter-evidence, and by first classifying and normalizing it to obtain multisensory counter-evidence parameters, three types of counter-evidence layers are generated: the first counter-evidence layer is a virtual image of gait, the second counter-evidence layer is tactile feedback of breathing, and the third counter-evidence layer is an audio signal of heart rate. This transforms abstract data into an intuitive representation that can be directly seen, touched, and heard on the extended reality interface, allowing nurses and doctors to understand the problem more quickly.

[0086] The counter-evidence generation module also includes aligning the virtual image data of the first counter-evidence layer, the tactile feedback data of the second counter-evidence layer, and the audio signal data of the third counter-evidence layer according to the timestamps of a unified time segment in the extended reality interface, and overlaying them in the same display area, and then performing the following multi-level judgments in sequence:

[0087] When the numerical difference of the gait signal category in the multi-sensory counter-evidence parameters in the current time segment is less than the preset gait difference threshold, and the numerical difference of the respiratory signal category in the multi-sensory counter-evidence parameters in the current time segment is less than the preset respiratory difference threshold, and the numerical difference of the heart rate signal category in the multi-sensory counter-evidence parameters in the current time segment is less than the preset heart rate difference threshold, the counter-evidence structure is directly output.

[0088] Otherwise, if the numerical difference of the gait signal category in the multi-sensory counter-evidence parameters in the current time segment is greater than the preset gait difference threshold, and the numerical difference of the respiratory signal category in the multi-sensory counter-evidence parameters in the current time segment is greater than the preset respiratory difference threshold, interpolation correction is performed and the counter-evidence structure is output after the correction is completed.

[0089] Otherwise, mark the multi-sensory counter-evidence parameters of the current time segment as multi-sensory counter-evidence parameters that do not meet the conditions, and continue to execute the process of aligning with the timestamps of the unified time segment and superimposing them in the same display area in subsequent time segments;

[0090] It should be noted that by aligning and overlaying the three types of counter-evidence layers in the extended reality interface, and then performing layered judgments based on the numerical differences and thresholds of the multi-sensory counter-evidence parameters, a multi-level logic is set up: if all differences are below the threshold, the counter-evidence structure is directly output; if some exceed the threshold, interpolation correction is performed before output; if the conditions are not met at all, it is marked and enters the next time segment for further processing. This ensures that the counter-evidence generation can respond quickly, be fault-tolerant and corrective, and avoid the problem of misjudgment caused by a single condition.

[0091] The intervention output module also includes decomposing the counter-evidence structure into gait difference data, respiratory difference data and heart rate difference data under a unified time reference, and performing amplitude calculation on the three types of difference data in each time segment to construct difference intensity data;

[0092] The cumulative statistical analysis of the difference intensity data is performed between time segments to obtain the cumulative difference intensity result. The cumulative difference intensity result is then compared with the gait intervention threshold, respiratory intervention threshold, and heart rate intervention threshold to form the intervention priority.

[0093] The intervention priorities are combined in a hierarchical manner within a unified time segment to generate intervention instructions, and hierarchical markers are added to the intervention instructions, including high priority and medium priority;

[0094] Intervention instructions are simultaneously transmitted to both nursing staff terminals and telemedicine staff terminals. When a high-priority marker is present in the intervention instruction, the telemedicine staff terminal is triggered to perform the treatment. When a medium-priority marker is present in the intervention instruction, the nursing staff terminal is triggered to perform the review and the review result is fed back to the telemedicine staff terminal.

[0095] It should be noted that by transforming the counter-evidence structure into intervention instructions with execution priorities, the counter-evidence structure is first decomposed into gait difference data, respiratory difference data, and heart rate difference data to form difference intensity data; then, cumulative statistics are performed to solve for the intervention priority; then, intervention instructions with graded labels are generated; finally, the intervention instructions are transmitted to nursing staff and remote medical staff. High-priority instructions directly trigger medical treatment, while medium-priority instructions require review by nursing staff first. This makes the intervention mechanism no longer a one-size-fits-all approach, but a tiered management system, improving the response speed in emergencies and ensuring careful handling in ordinary situations.

[0096] In actual operation, this system first adds a unified timestamp to signals from various sensors such as heart rate, blood oxygen, respiration, gait, and environment, and puts them into the original signal matrix. Next, it checks the differences between these signals and the reference baseline, calculates and corrects the average offset, and obtains a correction matrix. Then, the system analyzes the time drift of different signals to obtain the drift vector, and selects either interpolation or sliding window alignment based on this result to generate an alignment matrix. Finally, it checks the time interval difference of various signals on a unified time axis. If the error is small, it is directly fused; if the error is large, it is corrected first and then fused, ultimately obtaining stable and reliable fused data.

[0097] This process ensures that all signals are aligned and compared, preventing them from contradicting each other. The system then extracts features from the fused data. First, it arranges the signals into a first feature matrix. Then, it calculates the energy value and maximum / minimum difference in each time segment to form a second feature matrix, marking signals exceeding a threshold as abnormal. Next, it calculates the difference between adjacent time segments to obtain a difference matrix, assigning corrective weights to signals exceeding the threshold while retaining the original weights for those not exceeding the threshold. These weighted signals are then combined to form a weighted matrix. When the rate of change of the weighted matrix reaches a certain threshold within a continuous time period, a state vector is output.

[0098] Next, the difference recognition module converts the elderly person's expressed feelings into feeling data through voice recognition and interactive input. This data is then arranged into a feeling matrix under a unified time reference and compared one or more state vectors to obtain the first difference matrix. The system then determines whether the difference is persistent or occasional, thus forming the second difference matrix. Subsequently, it compares the second difference matrix with the rate of change of the state vectors to generate the third difference matrix and performs consistency statistics among five types of signals: heart rate, blood oxygen, respiration, gait, and environment. Only when multiple signals simultaneously show valid differences in the same time segment will the difference data be output, and the category, time segment, and cumulative frequency of occurrence will be recorded to ensure that the output abnormal information is the result of multi-dimensional joint verification, rather than being triggered by a single accidental anomaly.

[0099] After obtaining reliable difference data, the disproving evidence generation module classifies and normalizes this data according to three categories: gait, respiration, and heart rate, to obtain multi-sensory disproving evidence parameters. Then, it sequentially generates three different disproving evidence layers: fitting gait data into a virtual image to form the first disproving evidence layer; converting respiration data into tactile feedback signals to form the second disproving evidence layer; and converting heart rate data into audio signals to form the third disproving evidence layer. These three disproving evidence layers are then aligned with timestamps in the extended reality interface and superimposed on the same display area. The system performs multi-level judgments: if all values ​​are below a threshold, the disproving evidence structure is directly output; if some values ​​exceed the threshold, they are corrected before output; if none are met, it is marked as not meeting the conditions. The system continues to make judgments in the next time segment. Finally, the intervention output module decomposes the counter-evidence structure back into gait difference data, respiratory difference data, and heart rate difference data, calculates the difference intensity data for each time segment, and accumulates them to obtain the cumulative difference intensity result. These results are compared with the intervention thresholds for gait, respiration, and heart rate, respectively, to form the intervention priority. The system combines the priority and the counter-evidence structure to generate intervention instructions with high or medium priority markers. Finally, the intervention instructions are sent to both the nursing staff terminal and the remote medical staff terminal: high priority instructions will directly trigger the remote medical staff to take immediate action, while medium priority instructions require the nursing staff to review them first and then feed back the review results to the remote medical staff.

[0100] In practical applications, this system solves two key problems. First, the elderly person's self-reported feelings may not match the interface display. For example, the elderly person may say they have chest tightness or unsteady gait, but the interface may show everything as normal. Through the mechanism of perception matrix, difference matrix, and consistency statistics, the system can compare subjective and objective data within the same time segment. It also requires multiple signals to be abnormal at the same time before outputting difference data, thereby reducing false alarms. Second, some abnormalities are transient and cross-modal, such as a short-term increase in respiratory load accompanied by mild gait abnormalities. Through multi-sensory verification parameters, these abnormalities are projected as visual images, tactile feedback, and audible prompts, and presented in an aligned and superimposed manner in the extended reality interface. This allows caregivers to see, feel, and hear the abnormalities simultaneously within the same time segment, and then determine the specific treatment path based on the intervention priority and the hierarchical mechanism of intervention instructions.

[0101] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-dimensional sensing-based intelligent monitoring system for the elderly, characterized in that, The system comprises a data construction module, a feature forming module, a difference identification module, a counter-evidence generation module, and an intervention output module. The data construction module collects the heart rate signal, blood oxygen signal, breathing signal, gait signal, and environmental signal of the elderly through a heart rate sensor, blood oxygen sensor, breathing frequency sensor, gait capture device, and environment detection device, performs correction and alignment of all collected signals under a unified time reference, and outputs fused data. The feature forming module performs noise suppression and time segmentation processing on the fused data, performs difference operation on adjacent time slices, and combines weight calculation to output a state vector. The difference identification module converts the body feelings described by the elderly into feeling data through voice recognition and interactive input, encodes the feeling data into a feeling vector, and compares it with the state vector element by element to identify the features with differences and count the occurrence frequency, and outputs difference data. In the difference identification module, the body feelings described by the elderly are converted into feeling data through voice recognition processing and interactive input processing, and are aligned with the time stamp of the fused data under a unified time reference, and the feeling data is encoded into a feeling matrix. The feeling matrix and multiple state vectors are compared one by one within the same time segment, and the numerical difference of each corresponding position is solved, and all numerical differences are constructed into a first difference matrix, wherein the rows of the first difference matrix represent time segments, and the columns represent heart rate signals, blood oxygen signals, breathing signals, gait signals, and environmental signals. The first difference matrix performs difference continuity discrimination between adjacent time segments: When the numerical difference in the same column in the continuous time segment exceeds the preset difference threshold, a continuous difference marker is formed; when the numerical difference in only a single time segment exceeds the preset difference threshold, an instantaneous difference marker is formed, and the marker result is constructed into a second difference matrix. In the difference identification module, the change rate of the second difference matrix and the state vector in the same time segment is compared: When the change rate corresponding to the difference marker is lower than the lower limit, it is solved as invalid difference, and when the change rate corresponding to the difference marker is higher than the upper limit, it is solved as valid difference, and the comparison result is constructed into a third difference matrix. The third difference matrix performs consistency statistics between the heart rate signal column, blood oxygen signal column, breathing signal column, gait signal column, and environmental signal column: When multiple columns simultaneously solve valid differences in the same time segment, difference data is generated and output, and the difference data includes the signal category corresponding to the valid difference, the time segment, and the cumulative occurrence frequency; When multiple columns do not simultaneously solve valid differences in the same time segment, the statistical result of the time segment is recorded as not meeting the condition, and the consistency statistics is continued in the subsequent time segment and the statistical result is output The counter-evidence generation module solves the difference data into multi-sensory counter-evidence parameters, generates a first counter-evidence layer reflecting gait abnormalities through virtual images, generates a second counter-evidence layer reflecting respiratory load through tactile signals, generates a third counter-evidence layer reflecting heart rhythm abnormalities through audio signals, and superimposes the three types of counter-evidence layers in the extended reality interface to output a counter-evidence structure. The intervention output module outputs intervention instructions by delivering the counter-evidence structure to the nursing staff and the remote medical staff, solving the intervention priority based on the difference intensity, and outputting the intervention instructions, so that the nursing staff performs review and disposal according to the counter-evidence structure. 2.The multi-dimensional sensor-based smart monitoring system for the elderly according to claim 1, wherein: In the data construction module, the collected heart rate signal, blood oxygen signal, respiration signal, gait signal and environment signal are additionally attached with a unified timestamp and constructed into an original signal matrix. The original signal matrix is used to calculate the average of the sampling values of various signals in the reference time window, to solve the average offset formed by the difference between the average and the preset reference, and to subtract the average offset from each sampling value to output a corrected matrix. The timestamp of the corrected matrix is compared with the unified time reference to solve the drift vector of various signals on the target time axis, to perform interpolation padding when the drift is less than a threshold, to perform sliding window phase alignment when the drift is greater than the threshold, and to dynamically adjust the weighting factor during the alignment process according to the drift size, and to output an aligned matrix. The aligned matrix is used to perform synchronism discrimination in the unified reference time window, to first calculate the time interval difference between the corresponding sampling points of the heart rate signal, blood oxygen signal, respiration signal, gait signal and environment signal on the unified target time axis, to directly perform fusion operation when the time interval difference is less than a preset threshold, and to perform time slice interpolation correction and re-weighted fusion on the corresponding signals when the time interval difference is greater than or equal to the preset threshold, and to output fused data. 3.The multi-dimensional sensor-based smart monitoring system for the elderly according to claim 2, wherein: In the feature formation module, the fused data is used to obtain the signals of the heart rate signal, blood oxygen signal, respiration signal, gait signal and environment signal at a fixed sampling interval under the unified time reference, and to arrange and construct a first feature matrix in time sequence. The first feature matrix is divided into consecutive time slices, and the square sum of various signals is calculated in each time slice to count the energy value, and the upper limit and lower limit of each signal in the time slice are solved and the difference is calculated to form a second feature matrix, and the signals exceeding the preset threshold in the second feature matrix are marked as abnormal signals. The second feature matrix is used to perform difference calculation between adjacent time slices to solve the ratio of the difference between the average values of various signals in adjacent time slices to the length of the time slice, form a difference matrix, and assign a correction weight to the signals exceeding the preset threshold in the difference matrix, and keep the original weight of the signals not exceeding the preset threshold. The signals of the difference matrix are integrated to construct a weighted matrix, and a state vector is generated when the change rate in the weighted matrix continuously exceeds a preset stability threshold, and the change rate of the weighted matrix is continuously counted in a preset number of time slices when the change rate in the weighted matrix continuously does not exceed the preset stability threshold, and a state vector is generated when the cumulative result meets the continuity condition. 4.The multi-dimensional sensor-based smart monitoring system for the elderly according to claim 3, wherein: In the counter-evidence generation module, the difference data is classified according to the gait signal category, respiration signal category and heart rate signal category, and normalized processing is performed on the difference data of each category under the unified time reference to construct multi-sensory counter-evidence parameters. The data of the gait signal category in the multi-sensory counter-evidence parameter is subjected to a trajectory fitting operation within a time segment to generate a continuous gait curve, and the curve is mapped to virtual image data to form a first counter-evidence layer reflecting gait abnormalities; The data of the respiratory signal category in the multi-sensory counter-evidence parameter is subjected to a frequency decomposition operation within a time segment to solve the respiratory fluctuation intensity, and the intensity is converted into a tactile feedback signal to form a second counter-evidence layer reflecting respiratory load; The data of the heart rate signal category in the multi-sensory counter-evidence parameter is subjected to a rhythm recognition operation within a time segment to solve the heart rate fluctuation pattern, and the pattern is converted into an audio signal to form a third counter-evidence layer reflecting heart rate abnormalities. 5.The multi-dimensional sensor-based smart monitoring system for the elderly according to claim 4, wherein: In the counter-evidence generation module, the virtual image data of the first counter-evidence layer, the tactile feedback data of the second counter-evidence layer, and the audio signal data of the third counter-evidence layer are aligned according to the time stamp of the unified time segment in the extended reality interface, and are superimposed and presented in the same display area. Then, the following multi-level judgment is performed in sequence: When the numerical difference of the gait signal category in the multi-sensory counter-evidence parameter in the current time segment is less than the preset gait difference threshold, and the numerical difference of the respiratory signal category in the multi-sensory counter-evidence parameter in the current time segment is less than the preset respiratory difference threshold, and the numerical difference of the heart rate signal category in the multi-sensory counter-evidence parameter in the current time segment is less than the preset heart rate difference threshold, the counter-evidence structure is directly outputted; Otherwise, when the numerical difference of the gait signal category in the multi-sensory counter-evidence parameter in the current time segment is greater than the preset gait difference threshold, and the numerical difference of the respiratory signal category in the multi-sensory counter-evidence parameter in the current time segment is greater than the preset respiratory difference threshold, interpolation correction is performed and the counter-evidence structure is outputted after the correction is completed; Otherwise, the multi-sensory counter-evidence parameters in the current time segment are marked as multi-sensory counter-evidence parameters that do not meet the conditions, and the process of aligning according to the time stamp of the unified time segment and superimposing and presenting in the same display area is continued in the subsequent time segments. 6.The multi-dimensional sensor-based smart monitoring system for the elderly according to claim 5, wherein: In the intervention output module, the counter-evidence structure is decomposed into gait difference data, respiratory difference data, and heart rate difference data under a unified time reference, and amplitude calculation is performed on the three types of difference data in each time segment to construct difference intensity data; The difference intensity data is subjected to cumulative statistics between time segments to solve the cumulative difference intensity result, and the cumulative difference intensity result is compared with the gait intervention threshold, the respiratory intervention threshold, and the heart rate intervention threshold one by one to form an intervention priority; The intervention priority is subjected to hierarchical combination within a unified time segment to generate an intervention instruction, and a hierarchical marker is added to the intervention instruction, including high priority and medium priority; The intervention instruction is transmitted to the nursing staff terminal and the remote medical staff terminal at the same time. When there is a high priority marker in the intervention instruction, the remote medical staff terminal is triggered to perform disposal. When there is a medium priority marker in the intervention instruction, the nursing staff terminal is triggered to perform review, and the review result is fed back to the remote medical staff terminal.

Images