A non-contact health monitoring device, method, equipment and medium for recuperation

Abstract

This application provides a non-contact health monitoring device, method, equipment, and medium, belonging to the field of health monitoring technology. It collects Wi-Fi amplitude data and calculates Shannon entropy to form an instantaneous vitality index sequence. The instantaneous vitality index sequence is processed through a sliding time window to determine behavioral feature vectors and activity intensity indices. Normal behavior boundaries are defined using behavioral feature vectors from historical normal periods, and a baseline lower limit for activity intensity is determined based on the activity intensity indices from historical normal periods. If the behavioral feature vector of the current window is outside the normal behavior boundary, or the activity intensity index of the current window is lower than the baseline lower limit, it is determined to be an abnormal window. When both conditions are met simultaneously, it is determined to be a high-risk rhythmic abnormality window. This allows for graded early warning monitoring of the activities of elderly people living alone. Using the scheme of this application, it is possible to effectively identify environmental steady-state channel jitter and transient channel disturbances caused by human behavior.

Description

Technical Field

[0001] This application relates to the field of health monitoring technology, and more specifically, to a non-contact health monitoring device, method, equipment, and medium. Background Technology

[0002] Health monitoring refers to the process of dynamically quantifying and semantically mapping an individual's vital sign data stream and daily behavior patterns by using multi-source heterogeneous data acquisition methods such as wearable sensors, portable medical devices, and mobile terminals, combined with signal filtering and preprocessing, physiological feature parameter extraction algorithms, time series trend modeling, and risk stratification assessment frameworks. This process transforms fragmented vital sign readings into a health status profile with temporal correlation and clinical relevance.

[0003] In existing non-contact health monitoring processes, activity detection often relies on coarse-grained signal strength indicators or single-threshold motion sensors. This makes it difficult to effectively distinguish between steady-state channel jitter caused by environmental factors and transient channel disturbances caused by human behavior. For example, traditional Wi-Fi signal strength indicators can only detect very low-amplitude signal fluctuations when elderly people living alone are sitting quietly reading or taking a deep nap. The range of these fluctuations highly overlaps with the stationary state of the channel when an elderly person suddenly falls and loses mobility, resulting in a large number of false disturbances and missed detections that do not reflect the actual safety situation. Therefore, effectively identifying steady-state channel jitter caused by environmental factors and transient channel disturbances caused by human behavior has become a challenge for the industry. Summary of the Invention

[0004] This application provides a non-contact health monitoring device, method, equipment, and medium that can effectively identify steady-state channel jitter caused by environmental factors and transient channel disturbances caused by human behavior.

[0005] Firstly, this application provides a non-contact health monitoring method for elderly care, applied to a non-contact health monitoring device for monitoring abnormal activities of elderly people living alone based on non-contact sensing. The method includes the following steps: Amplitude data of Wi-Fi channel status information in the living area is collected, preprocessed to extract transient amplitude components, and Shannon entropy is calculated based on the distribution of the transient amplitude components of all subcarriers within a short time window to form a transient vitality index sequence. The instantaneous vitality index sequence is processed by a sliding time window to determine a behavioral feature vector that includes numerical dispersion, distribution symmetry and its deviation from the time series baseline, and to determine the proportion of vitality index frames that exceed the adaptive transient noise floor within the window, which is used as the activity intensity index. A single-classification model is trained using behavioral feature vectors from historical normal periods to define the boundaries of normal behavior, and the lower limit of the activity intensity benchmark is determined based on the time-segmented statistical distribution of the activity intensity index during historical normal periods. During real-time monitoring, if the behavior feature vector of the current window is outside the normal behavior boundary, or the activity intensity index of the current window is lower than the activity intensity benchmark lower limit, the corresponding window is determined to be an abnormal window. If both conditions are met at the same time, the corresponding window is determined to be a high-risk rhythm abnormal window. Based on the number of consecutive abnormal windows and the weight of abnormal types, the activities of elderly people living alone are monitored and warned in a tiered manner.

[0006] In some embodiments, processing the instantaneous vitality index sequence through a sliding time window to determine a behavioral feature vector that includes numerical dispersion, distribution symmetry, and its deviation from a time series benchmark specifically includes: The instantaneous vitality index values ​​of all frames within a preset fixed-duration sliding window are used to form a vitality sample set within the window. Calculate the statistical distribution characteristics of the live sample set within the window, wherein the statistical distribution characteristics include the standard deviation reflecting the numerical dispersion and the skewness reflecting the symmetry of the distribution; Calculate the deviation between the current window's statistical distribution characteristics and the time series benchmark of the corresponding period's historical normal mean; The standard deviation, the skewness, and the bias are combined to form a corresponding behavioral feature vector.

[0007] In some embodiments, determining the percentage of activity index frames exceeding the adaptive transient noise floor within the window, as the activity intensity index, specifically includes: Calculate the bottom percentile statistics of the live sample set within the current sliding window and determine them as the adaptive transient noise base for that window; Traverse the set of vitality samples within the window, determine whether the instantaneous vitality index value of each frame is greater than the adaptive transient noise floor, and count the results. Divide the count result by the total number of frames contained in the sliding window to obtain the activity intensity index of the corresponding window.

[0008] In some embodiments, training a single-classification model using behavioral feature vectors from historical normal periods to define the boundaries of normal behavior specifically includes: Obtain the behavioral feature vectors corresponding to all windows that match the current window time period identifier within the historical normal time period, and construct a training sample set of normal behavior by time period; The sample points in the time-segmented normal behavior training sample set are mapped to a high-dimensional feature space by a kernel function preset in the single classification model, and the minimum hypersphere containing the vast majority of sample points is obtained by solving the problem. The normal behavioral boundaries of elderly people living alone for a given period of time are obtained by fitting a closed surface in a high-dimensional space defined by the coordinates of the center of the minimum hypersphere and the radius parameters.

[0009] In some embodiments, during real-time monitoring, if the behavior feature vector of the current window is outside the normal behavior boundary, or the activity intensity index of the current window is lower than the activity intensity benchmark lower limit, then the corresponding window is determined to be an abnormal window. Determining the corresponding window as a high-risk rhythmic abnormal window when both conditions are met specifically includes: Obtain the behavior feature vector and activity intensity index corresponding to the current sliding window; The behavioral feature vector is input into the decision function corresponding to the normal behavior boundary of the corresponding time period to obtain a first judgment result that characterizes the relative positional relationship between the behavioral feature vector and the normal behavior boundary. The activity intensity index is compared with the lower limit of the baseline for the corresponding time period to obtain a second judgment result indicating whether the activity intensity index is lower than the statistical baseline level; When the first judgment result indicates that the behavior feature vector is outside the normal behavior boundary, or when the second judgment result indicates that the activity intensity index is lower than the activity intensity benchmark lower limit, the corresponding window is determined to be an abnormal window; When the first judgment result indicates that the behavioral feature vector is outside the normal behavioral boundary, and the second judgment result indicates that the activity intensity index is lower than the lower limit of the activity intensity benchmark, the corresponding window is determined to be a high-risk rhythm abnormality window.

[0010] In some embodiments, the hierarchical early warning monitoring of the activities of elderly people living alone, based on the number of consecutive abnormal windows and the weight of the abnormality type, specifically includes: The number of consecutive windows marked as anomalous before the current time is counted, and the anomaly type weight of each window is accumulated to obtain a weighted consecutive anomaly count value; The weighted continuous anomaly count value is compared with the preset multi-level early warning threshold to determine the current corresponding early warning level; Based on the aforementioned warning level, a corresponding warning response strategy is selected to conduct tiered warning monitoring of the activities of elderly people living alone.

[0011] In some embodiments, the amplitude data refers to fine-grained physical layer measurements that reflect the degree of amplitude attenuation of the signal after multipath propagation during the transmission of the wireless signal between the transmitter and receiver.

[0012] Secondly, this application provides a non-contact health monitoring device, which includes an activity anomaly monitoring unit, the activity anomaly monitoring unit comprising: The acquisition module is used to collect amplitude data of Wi-Fi channel status information in the living area, preprocess it to extract transient amplitude components, and calculate Shannon entropy based on the distribution of the transient amplitude components of all subcarriers within a short time window to form a transient vitality index sequence. The processing module is used to process the instantaneous vitality index sequence through a sliding time window, determine the behavioral feature vector containing numerical dispersion, distribution symmetry and its deviation from the time series reference, and determine the proportion of vitality index frames exceeding the adaptive transient noise base within the window as the activity intensity index; The processing module is also used to train a single-classification model using behavioral feature vectors from historical normal periods to define the boundaries of normal behavior, and to determine the lower limit of the activity intensity benchmark based on the time-segmented statistical distribution of the activity intensity index during historical normal periods. The processing module is also used to determine that if the behavior feature vector of the current window is outside the normal behavior boundary or the activity intensity index of the current window is lower than the activity intensity benchmark lower limit during real-time monitoring, the corresponding window is determined to be an abnormal window. When both conditions are met at the same time, the corresponding window is determined to be a high-risk rhythm abnormal window. The execution module is used to perform hierarchical early warning monitoring of the activities of elderly people living alone based on the number of consecutive abnormal windows and the weight of the abnormal type.

[0013] Thirdly, this application provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described non-contact health monitoring method.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described non-contact health monitoring method.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The non-contact health monitoring device, method, equipment, and medium provided in this application collect fine-grained amplitude data of Wi-Fi channel state information, perform steady-state channel component suppression preprocessing on the amplitude data, calculate the Shannon entropy of the probability distribution of transient amplitude values ​​of all subcarriers within a short sampling window, and construct a transient vitality index sequence. This transforms the original signal amplitude fluctuations, susceptible to multipath fading interference, into a statistical entropy change characterization reflecting the complexity of transient channel disturbances. Through the quantitative measurement of the disorder of signal distribution using Shannon entropy, a direct mathematical tool and data foundation are provided for distinguishing between environmental steady-state channel jitter and systematic entropy changes caused by actual human activity at the physical layer. Subsequently... Using a fixed-duration sliding window, the standard deviation, skewness, and temporal matching deviation value of the instantaneous vitality index sequence are extracted to construct a behavioral feature vector. The percentage of frames exceeding the adaptive transient noise floor within the window is then used as the activity intensity index. This achieves a multi-dimensional spatial mapping of the elderly's activity rhythms and comparison with historical benchmarks. The standard deviation characterizes the dispersion of entropy change amplitude, the skewness characterizes the morphological shift of the entropy change distribution, and the temporal matching deviation value characterizes the deviation of the behavioral pattern from the historical normal rhythm. This separates simple low-amplitude fluctuations from true long-term static states in the feature space. Simultaneously, the activity intensity index quantifies the percentage of frames exceeding the noise floor within the window. The effective activity component ratio of the baseline is determined. Then, a single-class support vector machine is trained using behavioral feature vectors matched with historical normal time periods to fit the normal behavior boundaries of elderly people living alone across different time periods. Simultaneously, the mean and standard deviation of the activity intensity index across different time periods in historical normal time periods are determined. This process abstracts the individualized daily rhythms and activity habits of the elderly from massive monitoring data into a tight decision boundary and statistical baseline. The single-class support vector machine accurately delineates the normal behavior envelope in the feature space. In real-time monitoring, if the behavioral feature vector of the current window is outside the normal behavior boundary, or the activity intensity index is lower than the corresponding time period baseline lower limit, it is determined to be an abnormal window. When both conditions are met simultaneously, a high-risk rhythm anomaly window is identified. This hierarchical judgment mechanism allows the system to tolerate regular low-activity routines. When elderly people are taking a normal nap or sitting quietly, their behavioral feature vectors still fall within the boundaries of the single-class support vector machine for the corresponding time period, and even if the activity intensity index is low, a high-risk alarm will not be triggered. The system has a full-dimensional quantitative perception capability for rhythm deviations and abnormal activity levels. When the behavioral pattern deviates significantly from the historical norm at the statistical feature level, or the activity intensity is systematically lower than the historical low threshold, the corresponding level of abnormal warning can be triggered. In summary, this scheme can effectively identify environmental steady-state channel jitter and transient channel disturbances caused by human behavior. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a non-contact health monitoring method according to some embodiments of this application; Figure 2This is a flowchart illustrating the process of determining behavioral feature vectors according to some embodiments of this application; Figure 3 This is a schematic flowchart illustrating the process of determining an activity intensity index according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of an activity anomaly monitoring unit according to some embodiments of this application; Figure 5 This is an internal structural diagram of a computer device for implementing a non-contact health monitoring method according to some embodiments of this application. Detailed Implementation

[0017] To better understand the technical solutions in this embodiment, the technical solutions in this embodiment will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0018] refer to Figure 1 The figure is a flowchart illustrating a non-contact health monitoring method according to some embodiments of this application. The non-contact health monitoring method mainly includes the following steps: In step 101, amplitude data of Wi-Fi channel status information in the living area is collected, preprocessed to extract transient amplitude components, and Shannon entropy is calculated based on the distribution of the transient amplitude components of all subcarriers within a short time window to form a transient vitality index sequence.

[0019] In practice, at least one pair of Wi-Fi transceivers can be deployed in the living area of ​​an elderly person living alone. The transmitting end continuously sends wireless signals, and the receiving end acquires channel state information data packets at a preset sampling frequency. The channel state information data packets contain channel response information of each subcarrier under orthogonal frequency division multiplexing (OFDM) technology. The channel response information of each subcarrier contains amplitude and phase components. Steady-state channel component suppression preprocessing is performed on the acquired amplitude data. The least mean square algorithm is used to filter out the steady-state multipath components that remain unchanged for more than 100 consecutive frames in the amplitude data, while retaining the transient amplitude components that change with human activity. For the channel state information data within each short sampling window, all usable subcarriers are extracted. The transient amplitude values ​​of each subcarrier are divided by the sum of the transient amplitude values ​​of all subcarriers within the window to obtain the normalized probability value of each subcarrier within the window. Based on the obtained set of normalized probability values, the Shannon entropy of the set is calculated. The Shannon entropy is calculated as follows: for each normalized probability value in the set, the product of the probability value and its logarithm to the base of the natural constant is calculated, and the sum of all the products is taken as the negative number. The resulting value is the instantaneous vitality index corresponding to the channel state information data of that window. The instantaneous vitality indices corresponding to each short-time sampling window during the acquisition process are arranged in the order of acquisition time to form the instantaneous vitality index sequence.

[0020] It should be noted that the amplitude data mentioned in this application refers to fine-grained physical layer measurements reflecting the degree of amplitude attenuation after multipath propagation of the wireless signal during its propagation between the transmitter and receiver. Each frame of data contains amplitude responses corresponding to multiple subcarriers, and their variations are modulated by human activity affecting the indoor wireless propagation environment, thus indirectly characterizing human activity in space. The Shannon entropy is used to quantify the uniformity of the transient amplitude distribution of all subcarriers within a short sampling window. A higher entropy value indicates a more uniform distribution of transient amplitudes across subcarriers, corresponding to a more complex transient multipath effect in the indoor wireless propagation environment at that moment, reflecting more intense instantaneous activity of elderly people living alone. A lower entropy value indicates a more concentrated amplitude distribution on a few subcarriers, corresponding to a more stable indoor wireless propagation environment at that moment, reflecting that elderly people living alone are in a static or low-activity state. The instantaneous vitality index sequence is... The Shannon entropy values ​​calculated from channel state information data of multiple consecutive short-time sampling windows are arranged in chronological order to form a one-dimensional numerical sequence. The fluctuations in the values ​​in the sequence directly correspond to the dynamic fluctuations in the activity level of elderly people living alone. The Wi-Fi transceiver can be a commercial wireless network card that supports channel state information derivation, such as an Intel 5300 series network card or a compatible IoT development board. Both the transmitter and receiver are deployed in the living area where the elderly person lives alone. The operating frequency band can be 2.4 GHz or 5 GHz. The number of subcarriers depends on the specific wireless network card model and operating bandwidth configuration. This application does not impose a unique limitation on the specific hardware implementation method and parameter configuration. The duration of the short-time sampling window is set to 50ms-200ms to cover the smallest time scale of human micro-movements, ensuring that transient channel changes caused by human activity can be captured.

[0021] In step 102, the instantaneous vitality index sequence is processed by a sliding time window to determine a behavioral feature vector that includes numerical dispersion, distribution symmetry and its deviation from the time series reference, and to determine the proportion of vitality index frames exceeding the adaptive transient noise base within the window, which is used as the activity intensity index.

[0022] In some embodiments, reference Figure 2 As shown in the figure, this is a flowchart illustrating the determination of behavioral feature vectors according to some embodiments of this application. The determination of behavioral feature vectors, including numerical dispersion, distribution symmetry, and deviation from the time series benchmark, by processing the instantaneous vitality index sequence through a sliding time window can be achieved through the following steps: First, in step 1021, the instantaneous vitality index values ​​of all frames within a preset fixed-duration sliding window are obtained to form a vitality sample set within the window; Then, in step 1022, the statistical distribution characteristics of the live sample set within the window are calculated, the statistical distribution characteristics including the standard deviation reflecting the numerical dispersion and the skewness reflecting the symmetry of the distribution; Subsequently, in step 1023, the deviation between the current window statistical distribution characteristics and the time series benchmark of the historical normal mean for the corresponding time period is calculated; Finally, in step 1024, the standard deviation, the skewness, and the deviation are combined into a corresponding behavioral feature vector.

[0023] In specific implementation, obtaining the instantaneous vitality index values ​​of all frames within a preset fixed-duration sliding window to form a vitality sample set within the window can be achieved in the following way: For example, using the starting point of the instantaneous vitality index sequence as a reference, a rectangular window with a time span equal to the preset fixed duration is used. The rectangular window is moved sequentially along the time axis according to a preset sliding step size. After each movement, the instantaneous vitality index values ​​corresponding to all frames within the time interval covered by the rectangular window are extracted as an independent vitality sample set within the window. The fixed duration is determined based on the average duration of a single typical behavioral event in the daily life activities of elderly people living alone, so that the vitality data within a single window can completely cover the change process of an activity event from start to finish. The sliding step size balances the timeliness of the monitoring response with the consumption of system computing resources. The smaller the step size, the larger the overlap between adjacent windows, resulting in more detailed capture of behavioral changes; the larger the step size, the smaller the data processing volume and the higher the overall system efficiency. This application does not impose a unique limitation on the specific value of the fixed duration and the specific value of the sliding step size.

[0024] It should be noted that the set of vitality samples within the window mentioned in this application refers to a one-dimensional numerical set consisting of the instantaneous vitality index values ​​corresponding to all frames within the time interval covered by a preset fixed-duration sliding window.

[0025] In specific implementation, the statistical distribution characteristics of the vitality sample set within the window are calculated. These characteristics include the standard deviation reflecting numerical dispersion and the skewness reflecting distribution symmetry. This can be achieved in the following way: First, all instantaneous vitality index values ​​contained in the vitality sample set within the window are accumulated. The accumulated result is divided by the number of elements in the vitality sample set within the window to obtain the vitality mean within the window. Then, for each instantaneous vitality index value in the vitality sample set within the window, the difference between that instantaneous vitality index value and the vitality mean within the window is calculated. The squared difference is then accumulated, and the accumulated result is divided by the number of elements to obtain the variance. Finally, the square root of the variance is calculated, and the result is the mean. The standard deviation reflects the dispersion of the vitality sample set within the window. Simultaneously, for each instantaneous vitality index value in the vitality sample set within the window, the cube of the difference between the instantaneous vitality index value and the mean vitality value within the window is calculated. All cube values ​​are summed and divided by the number of elements. This result is then divided by the cube of the standard deviation. The resulting value is the skewness reflecting the symmetry of the distribution. In a preferred embodiment, the deviation of the time series benchmark is calculated as follows: the basic feature vector composed of the standard deviation and skewness of the current window is subtracted from the mean vector of the basic feature vectors of all windows in the corresponding historical normal time period. The magnitude of this difference is taken as the deviation of the time series benchmark. This application does not limit this calculation.

[0026] It should be noted that the standard deviation mentioned in this application refers to the statistical characteristic value obtained by measuring the dispersion of all instantaneous vitality index values ​​in the vitality sample set within the window, which is used to quantify the fluctuation range of the instantaneous activity level of elderly people living alone within the window period; the skewness refers to the statistical characteristic value obtained by measuring the distribution asymmetry of all instantaneous vitality index values ​​in the vitality sample set within the window, which is used to characterize the skewed characteristics of the high-activity state and low-activity state of elderly people living alone in the time distribution within the window period; the deviation of the time series benchmark refers to the statistical value that quantifies the degree of deviation between the basic characteristics of the current window and the normal characteristics of the same period in history, which is used to characterize the degree of matching between the behavior pattern and the daily rhythm.

[0027] In specific implementation, combining the standard deviation, skewness, and deviation into a corresponding behavioral feature vector can be achieved in the following way: for example, the calculated standard deviation is used as the first dimension component of the three-dimensional vector, the calculated skewness is used as the second dimension component of the three-dimensional vector, and the calculated deviation is used as the third dimension component of the three-dimensional vector. A three-dimensional vector is constructed according to a fixed combination order of the first dimension being the fluctuation dimension, the second dimension being the symmetry dimension, and the third dimension being the rhythm matching dimension, and this vector is determined as the behavioral feature vector corresponding to the current sliding window. As a preferred embodiment, before combining the standard deviation, skewness, and temporal matching deviation into a behavioral feature vector, the standard deviation, skewness, and temporal matching deviation can be normalized respectively to eliminate the influence of the difference in the dimensions of the three on the distance metric in the subsequent training of the single-class support vector machine model.

[0028] It should be noted that the behavioral feature vector mentioned in this application refers to a three-dimensional numerical vector formed by combining the standard deviation, skewness, and deviation of the same sliding window in a preset order. It is used to jointly describe the behavioral rhythm pattern of elderly people living alone within the window period from three dimensions: the intensity of activity fluctuations, the symmetry of activity distribution, and the degree of rhythm matching.

[0029] In some embodiments, determining the percentage of activity index frames exceeding the adaptive transient noise floor within the window, as an activity intensity index, can be achieved through the following steps: Calculate the bottom percentile statistics of the live sample set within the current sliding window and determine them as the adaptive transient noise base for that window; Traverse the set of vitality samples within the window, determine whether the instantaneous vitality index value of each frame is greater than the adaptive transient noise floor, and count the results. Divide the count result by the total number of frames contained in the sliding window to obtain the activity intensity index of the corresponding window.

[0030] In specific implementation, calculating the bottom percentile statistical value of the active sample set within the current sliding window and determining it as the adaptive transient noise base of the window can be achieved in the following way: For example, the entire 24 hours are pre-divided into multiple continuous time intervals with a fixed time granularity, and each time interval is assigned a unique time interval identifier; for any current sliding window, the midpoint between its start and end times is taken as the timestamp of the window, the time interval into which the timestamp falls is found, and the time interval identifier of the time interval is obtained; all values ​​in the active sample set within the current window are arranged in ascending order, and the value located at the preset bottom percentile level is taken as the adaptive transient noise base of the window; the bottom percentile level is set to 5%-15% to characterize the background noise level when there is no human activity within the window.

[0031] It should be noted that the adaptive transient noise floor mentioned in this application refers to a reference value determined based on the bottom percentile statistical value of the active sample set within the current sliding window, which is used to define the boundary level between the entropy change caused by effective human activity within the window and the background noise.

[0032] In specific implementation, traversing the set of vitality samples within the window and determining whether the instantaneous vitality index value of each frame is greater than the adaptive transient noise base, and counting accordingly, can be achieved in the following way: For example, initialize a counter variable with a count value of zero; sequentially read each instantaneous vitality index value in the set of vitality samples within the window in chronological order; for each read instantaneous vitality index value, compare its value with the adaptive transient noise base; if the comparison result is that the instantaneous vitality index value is greater than the adaptive transient noise base, increment the current value of the counter variable by one; if the comparison result is that the instantaneous vitality index value is less than or equal to the adaptive transient noise base, keep the value of the counter variable unchanged; after all instantaneous vitality index values ​​in the set of vitality samples within the window have completed the above comparison and update operations, the final value of the counter variable is the number of frames in the window whose instantaneous vitality index value exceeds the adaptive transient noise base; in other embodiments, a segmented counting method can also be used for parallel processing to improve computational efficiency in large-scale data scenarios. This application does not limit the specific loop implementation form of the traversal execution.

[0033] For specific implementation, refer to Figure 3 As shown in the figure, this is a flowchart illustrating the process of determining the activity intensity index in some embodiments of this application. Dividing the count result by the total number of frames contained in the sliding window to obtain the activity intensity index of the corresponding window can be achieved in the following way: for example, obtaining the total number of frames contained in the current sliding window, where the total number of frames is equal to the number of elements in the activity sample set within the window; using the count result obtained in the aforementioned steps as the dividend and the total number of frames as the divisor, performing floating-point division, and the resulting quotient is the activity intensity index of the corresponding current window; as a preferred embodiment, when calculating the activity intensity index, if the total number of frames is lower than a preset minimum effective frame count threshold, the activity intensity index of the window can be directly set to zero or marked as an invalid window to avoid statistical distortion due to insufficient sample size within the window.

[0034] It should be noted that the activity intensity index mentioned in this application refers to the ratio of the number of frames in the current sliding window whose instantaneous vitality index value exceeds the adaptive transient noise floor of the corresponding time period to the total number of frames in the window. It is used to quantify the proportion of the duration of high vitality of elderly people living alone in the window period and to reflect the overall activity intensity level of the period.

[0035] In step 103, a single-classification model is trained using behavioral feature vectors from historical normal periods to define the boundaries of normal behavior, and the lower limit of the activity intensity benchmark is determined based on the time-segmented statistical distribution of the activity intensity index during historical normal periods.

[0036] In some embodiments, training a single-classification model using behavioral feature vectors from historical normal periods to define the boundaries of normal behavior can be achieved through the following steps: Obtain the behavioral feature vectors corresponding to all windows that match the current window time period identifier within the historical normal time period, and construct a training sample set of normal behavior by time period; The sample points in the time-segmented normal behavior training sample set are mapped to a high-dimensional feature space by a kernel function preset in the single classification model, and the minimum hypersphere containing the vast majority of sample points is obtained by solving the problem. The normal behavioral boundaries of elderly people living alone for a given period of time are obtained by fitting a closed surface in a high-dimensional space defined by the coordinates of the center of the minimum hypersphere and the radius parameters.

[0037] In specific implementation, obtaining the behavioral feature vectors corresponding to all windows within a historical normal time period to form a normal behavior training sample set can be achieved in the following way: First, a time interval for establishing a baseline of normal behavior is defined as the historical normal time period. The historical normal time period should cover the daily routine of an elderly person living alone for several consecutive days in a state of physical and mental health, and no confirmed abnormal events or system maintenance operations have occurred during this period. Then, all windows generated by sliding windows of a preset fixed duration within the historical normal time period are traversed, and the behavioral feature vectors corresponding to all windows are grouped according to the time period identifier, with each time period identifier corresponding to a set of behavioral feature vectors. Finally, the vector set corresponding to each time period identifier is deduplicated and its integrity is verified. Invalid vectors with missing dimensions due to data loss or insufficient window sample size are removed, and the set of behavioral feature vectors that pass the verification is used as the time-segmented normal behavior training sample set corresponding to that time period identifier. As a preferred embodiment, the selection of the historical normal time period can be dynamically adjusted in combination with the daily routine of the elderly person living alone, for example, by rolling updates on a weekly cycle, so that the time-segmented normal behavior training sample set can adapt to the seasonal or gradual changes in the elderly person's activity habits.

[0038] It should be noted that the time-segmented normal behavior training sample set mentioned in this application refers to a vector set consisting of behavioral feature vectors corresponding to all sliding windows with the same time segment identifier within a historical normal time segment, after deduplication and integrity verification. This set is used to provide training samples for a single-class support vector machine to describe the rhythmic patterns of normal behavior of elderly people living alone in the corresponding time segment, so as to learn the distribution characteristics of normal behavior.

[0039] It should be noted that the single-class classification model is a single-class support vector machine. Specifically, in its implementation, the sample points in the time-segmented normal behavior training sample set are mapped to a high-dimensional feature space using a preset kernel function in the single-class classification model. The solution to obtain the minimum hypersphere containing the vast majority of sample points can be achieved in the following way: First, select the kernel function type used in the single-class support vector machine. This kernel function is used to calculate the inner product of any two behavioral feature vectors in the high-dimensional feature space. Taking the radial basis function kernel as an example, this kernel function can map the non-linearly distributed sample points in the original three-dimensional feature space to a higher-dimensional feature space, making the sample points that were difficult to enclose with a closed surface in the original space present a more compact clustered form in the high-dimensional space. Then, using each behavioral feature vector in the time-segmented normal behavior training sample set as input, construct the optimization objective of the single-class support vector machine. This optimization objective is to find a hypersphere with the smallest radius in the high-dimensional feature space, and to make this hypersphere... The hypersphere can contain the vast majority of sample points in the training sample set, while allowing a small number of sample points to be located outside the hypersphere to enhance the model's tolerance to minor fluctuations in normal behavior. During the optimization process, the balance between the hypersphere volume and sample coverage is controlled by adjusting a preset regularization parameter. The smaller the value of the regularization parameter, the more compact the hypersphere and the more sensitive it is to anomalies; the larger the value, the more relaxed the hypersphere and the stronger its tolerance to normal fluctuations. Finally, by solving this optimization problem, the implicit representation of the hypersphere's center in the high-dimensional feature space and the value of the hypersphere's radius are obtained. As a preferred embodiment, the type of kernel function and the internal parameters of the kernel function can be selected by performing a grid search on historical normal time period data through cross-validation to select the combination that minimizes the validation error. In other embodiments, linear kernel functions or polynomial kernel functions can also be used to adapt to behavioral feature vectors with different distribution patterns. This application does not impose a unique limitation on the specific selection of the kernel function and the specific value of the regularization parameter.

[0040] It should be noted that the single-classification model described in this application refers to an anomaly detection model that relies solely on a normal behavior training sample set for unsupervised learning. It is used to find a hypersphere boundary in a high-dimensional feature space that can tightly enclose the distribution area of ​​most normal behavior feature vectors, thereby establishing a descriptive model of the normal behavior patterns of elderly people living alone. The minimum hypersphere refers to a spherical geometry with the smallest radius that contains most of the normal behavior training sample points, obtained by a single-class support vector machine in a high-dimensional feature space after kernel function mapping. It is used to define the distribution range of normal behavior feature vectors of elderly people living alone in a high-dimensional space. Its center and radius parameters together constitute the mathematical expression basis of the normal behavior boundary.

[0041] In specific implementation, the normal behavioral boundaries of elderly people living alone for a corresponding time period can be obtained by fitting the closed surface of the high-dimensional space defined by the center coordinates and radius parameters of the minimum hypersphere. This can be achieved in the following way: After training a single-class support vector machine, extract the support vector set and corresponding Lagrange multiplier coefficients stored within the model. The center position of the minimum hypersphere is implicitly expressed in the high-dimensional feature space through the support vectors and the Lagrange multipliers. For any new behavioral feature vector, calculate the distance from the behavioral feature vector to the center in the high-dimensional feature space after kernel function mapping, and compare this distance with the radius of the hypersphere. Based on the above comparison logic, the high-dimensional... The hypersphere equation defined by the center and radius of the sphere in the feature space is transformed into a decision function. When any row of feature vectors is received as input, the decision function outputs a real value. When the output value is greater than zero, it indicates that the mapping point of the feature vector in the high-dimensional feature space is located inside or above the hypersphere. When the output value is less than zero, it indicates that the mapping point is located outside the hypersphere. The closed surface of the hypersphere defined by the above decision function is determined as the normal behavior boundary of elderly people living alone. In other embodiments, the output value of the decision function can also be scaled to make the degree of abnormality more intuitively quantitatively explained. This application does not limit the specific update mechanism of the normal behavior boundary and the post-processing method of the output value.

[0042] It should be noted that the normal behavior boundary mentioned in this application refers to the closed surface corresponding to the smallest hypersphere obtained by training a single-class support vector machine in the original feature space, which is used to distinguish whether the behavioral feature vector of elderly people living alone belongs to a normal rhythm pattern or a potential abnormal pattern during real-time monitoring.

[0043] In specific implementation, determining the lower limit of the activity intensity benchmark based on the time-segmented statistical distribution of the activity intensity index during historical normal periods can be achieved in the following way: First, define the same historical normal period as when training the single-class support vector machine. Traverse each window generated by a sliding window of a preset fixed duration within this historical normal period, and extract the activity intensity index of each window according to the time period identifier, forming a historical activity intensity index sequence corresponding to each time period identifier. Then, accumulate all values ​​in the historical activity intensity index sequence corresponding to each time period identifier, and divide the accumulated result by the number of values ​​contained in the sequence. The result is the mean of the activity intensity index in the historical normal period corresponding to that time period identifier. Simultaneously, for the sequence... For each activity intensity index value in the column, calculate the difference between that value and the mean, square the difference, sum the results, divide the sum by the number of values ​​in the sequence to obtain the variance, and then take the square root of the variance. The result is the standard deviation of the activity intensity index in the historical normal period corresponding to that time period. Finally, calculate the lower limit of the activity intensity benchmark corresponding to each time period based on the mean and standard deviation of the activity intensity index corresponding to that time period, according to a preset multiple parameter. The lower limit of the benchmark is calculated by subtracting the product of the multiple parameter and the standard deviation from the mean. In specific implementation, for example, the multiple parameter can be in the range of 1-3 to ensure that the lower limit of the activity intensity benchmark is not less than 0. This will not be elaborated further here.

[0044] It should be noted that the mean of the activity intensity index mentioned in this application refers to the statistical quantity obtained by arithmetically averaging the activity intensity indices of all sliding windows within a historical normal period. It is used to characterize the average level of the proportion of high-activity time in each period of a normal state for elderly people living alone. The standard deviation of the activity intensity index refers to the statistical quantity obtained by measuring the dispersion of the activity intensity indices of all sliding windows within a historical normal period. It is used to characterize the daily natural fluctuation range of the activity intensity index of elderly people living alone under normal conditions. The lower limit of the benchmark determined by the mean and the standard deviation is used as the fluctuation tolerance boundary when setting the abnormal activity intensity judgment threshold in real-time monitoring.

[0045] In step 104, during real-time monitoring, if the behavior feature vector of the current window is outside the normal behavior boundary, or the activity intensity index of the current window is lower than the activity intensity benchmark lower limit, then the corresponding window is determined to be an abnormal window. If both conditions are met simultaneously, the corresponding window is determined to be a high-risk rhythm abnormal window.

[0046] In some embodiments, during real-time monitoring, if the behavior feature vector of the current window is outside the normal behavior boundary, or the activity intensity index of the current window is lower than the activity intensity benchmark lower limit, then the corresponding window is determined to be an abnormal window. Determining the corresponding window to be a high-risk rhythmic abnormal window when both conditions are met simultaneously can be achieved using the following steps: Obtain the behavior feature vector and activity intensity index corresponding to the current sliding window; The behavioral feature vector is input into the decision function corresponding to the normal behavior boundary of the corresponding time period to obtain a first judgment result that characterizes the relative positional relationship between the behavioral feature vector and the normal behavior boundary. The activity intensity index is compared with the lower limit of the baseline for the corresponding time period to obtain a second judgment result indicating whether the activity intensity index is lower than the statistical baseline level; When the first judgment result indicates that the behavior feature vector is outside the normal behavior boundary, or when the second judgment result indicates that the activity intensity index is lower than the activity intensity benchmark lower limit, the corresponding window is determined to be an abnormal window; When the first judgment result indicates that the behavioral feature vector is outside the normal behavioral boundary, and the second judgment result indicates that the activity intensity index is lower than the lower limit of the activity intensity benchmark, the corresponding window is determined to be a high-risk rhythm abnormality window.

[0047] In specific implementation, the behavior feature vector is input into the decision function corresponding to the normal behavior boundary of the corresponding time period to obtain a first judgment result representing the relative positional relationship between the behavior feature vector and the normal behavior boundary. This can be achieved in the following way, for example: during real-time monitoring, for each newly generated sliding window, the behavior feature vector corresponding to the window is passed as an input parameter to step 103. The system employs a pre-trained single-class support vector machine decision function corresponding to a specific time period. This decision function utilizes the support vector set and corresponding Lagrange multiplier coefficients stored during the training phase to calculate the distance from the behavioral feature vector to the center of the smallest hypersphere in the high-dimensional feature space using a preset kernel function. This distance is then compared numerically with the radius of the smallest hypersphere. If the calculated distance is less than or equal to the radius, the decision function outputs a non-negative value, indicating that the behavioral feature vector is located inside or above the normal behavior boundary, and the first judgment result is set to a normal state. If the calculated distance is greater than the radius, the decision function outputs a negative value, indicating that the behavioral feature vector is located outside the normal behavior boundary, and the first judgment result is set to an abnormal state. In other embodiments, the original value output by the decision function can also be directly used as the continuous confidence representation of the first judgment result. This application does not impose a unique limitation on the specific representation of the first judgment result or the setting method of the buffer tolerance interval.

[0048] It should be noted that the decision function mentioned in this application refers to a mathematical discriminant function constructed based on the center coordinates and radius parameters of the smallest hypersphere, used to output a real value for any input behavioral feature vector to characterize the relative positional relationship of the vector with the normal behavioral boundary in the high-dimensional feature space; the first judgment result refers to the output result obtained after inputting the behavioral feature vector of the current window into the decision function, used to indicate the positional affiliation of the behavioral feature vector with respect to the normal behavioral boundary in the high-dimensional feature space, that is, to determine whether the behavioral pattern of the window is within the normal behavioral rhythm range of elderly people living alone.

[0049] In specific implementation, comparing the activity intensity index with the baseline lower limit of the corresponding time period to obtain a second judgment result indicating whether the activity intensity index is lower than the statistical baseline level can be achieved in the following way: for example, comparing the activity intensity index calculated in the current window with the baseline lower limit of the corresponding time period; if the activity intensity index is less than the baseline lower limit, it indicates that the proportion of effective activity time of elderly people living alone in the current window is significantly lower than the historical average level of their daily period, and the second judgment result is set to an abnormal state; if the activity intensity index is greater than or equal to the baseline lower limit, it indicates that the activity intensity of the current window is within the normal fluctuation range, and the second judgment result is set to a normal state; wherein, the multiplier parameter is used to adjust the sensitivity of the abnormal judgment, the larger the value of the multiplier parameter, the lower the baseline lower limit, and the more stringent the triggering condition for judging that the activity intensity is too low; this application does not limit the specific value of the multiplier parameter.

[0050] It should be noted that the second judgment result in this application refers to the judgment result obtained by comparing the activity intensity index of the current window with the corresponding time period benchmark lower limit determined based on the average and standard deviation of the historical normal time period, which is used to indicate whether the activity intensity level of the window is significantly lower than the normal statistical fluctuation range of the same period in history.

[0051] In specific implementation, when the first judgment result indicates that the behavioral feature vector is outside the normal behavior boundary, or the second judgment result indicates that the activity intensity index is lower than the activity intensity benchmark lower limit, the corresponding window is determined to be an abnormal window. When the first judgment result indicates that the behavioral feature vector is outside the normal behavior boundary, and the second judgment result indicates that the activity intensity index is lower than the activity intensity benchmark lower limit, determining the corresponding window as a high-risk rhythm abnormal window can be achieved in the following way: for example, simultaneously performing the acquisition operations of the first judgment result and the second judgment result on the same sliding window; when the first judgment result is an abnormal state or the second judgment result is an abnormal state... The current window is marked as an abnormal window. If and only if the first judgment result is an abnormal state and the second judgment result is also an abnormal state, the current window is marked as a high-risk rhythm abnormal window, and the window identifier, timestamp, and corresponding abnormal type information of the window are written into an abnormal window record queue. If both the first judgment result and the second judgment result are normal states, the current window is determined to be a normal window and no abnormal marking is triggered. In other embodiments, different confidence weights may be assigned to the first judgment result and the second judgment result respectively, and the abnormal level may be comprehensively determined by weighted fusion. This application does not limit the specific judgment logic of abnormal marking and the verification mechanism of interpretable events.

[0052] It should be noted that the abnormal rhythm described in this application refers to the state in which the current sliding window simultaneously satisfies the condition that the behavioral feature vector is outside the normal behavioral boundary and the activity intensity index is below the lower limit of the statistical baseline level. This is used to characterize that the activity rhythm of elderly people living alone has deviated from their daily normal behavior pattern and the activity level is abnormally low during the window period, which may pose a health or safety risk.

[0053] In specific implementation, when the first judgment result indicates that the behavioral feature vector is outside the normal behavior boundary, and the second judgment result indicates that the activity intensity index is lower than the lower limit of the activity intensity benchmark, the determination of the corresponding window as a high-risk rhythm abnormality window can be implemented in the following way, for example: simultaneously execute the acquisition operations of the first judgment result and the second judgment result for the same sliding window. If and only if the first judgment result is an abnormal state and the second judgment result is also an abnormal state, mark the current window as a high-risk rhythm abnormality window, and write the window identifier, timestamp, and corresponding abnormality type information of the window into an abnormal window record queue, and assign it a risk weight value higher than that of an ordinary abnormal window; if either the first judgment result or the second judgment result is a normal state, then it is processed separately according to the judgment logic of the ordinary abnormal window. This application does not make a unique limitation on the specific judgment logic and weight value method of high-risk abnormality marking.

[0054] It should be noted that the high-risk rhythm abnormality window mentioned in this application refers to the state in which the current sliding window simultaneously satisfies the condition that the behavioral feature vector is outside the normal behavioral boundary and the activity intensity index is lower than the lower limit of the statistical baseline level. It is used to characterize that the activity rhythm of elderly people living alone has significantly deviated from their daily normal behavior pattern and the activity level is abnormally low during the window period, which poses a high health or safety risk.

[0055] In step 105, the activities of elderly people living alone are monitored and warned in a tiered manner based on the number of consecutive abnormal windows and the weight of the abnormal type.

[0056] In some embodiments, the hierarchical early warning monitoring of the activities of elderly people living alone, based on the number of consecutive abnormal windows and the weight of the abnormality type, can be achieved through the following steps: The number of consecutive windows marked as anomalous before the current time is counted, and the anomaly type weight of each window is accumulated to obtain a weighted consecutive anomaly count value; The weighted continuous anomaly count value is compared with the preset multi-level early warning threshold to determine the current corresponding early warning level; Based on the aforementioned warning level, a corresponding warning response strategy is selected to conduct tiered warning monitoring of the activities of elderly people living alone.

[0057] In specific implementation, the number of consecutive windows marked as abnormal before the current time is counted, and the weight of the abnormality type of each window is accumulated to obtain a weighted consecutive abnormality count value. This can be achieved in the following way: for example, maintaining a counter variable to record the weighted consecutive abnormality values, which is set to zero during system initialization; after each window is judged, if the window is marked as a high-risk rhythm abnormality window, the current value of the counter variable is read and a preset high-risk weight value is accumulated; if the window is marked as a normal abnormality window, the current value of the counter variable is read and a preset normal risk weight value is accumulated; if the window is a normal window, regardless of the current value of the counter variable, it is reset to zero to indicate the interruption of the consecutive abnormality sequence, thereby obtaining a weighted consecutive abnormality count value; preferably, the high-risk weight value is set to 2-5, and the normal risk weight value is set to 1 to ensure that high-risk abnormalities can trigger high-level warnings more quickly; in other embodiments, the abnormality ratio within a sliding window can also be used instead of strict continuous counting for abnormality assessment. This application does not limit the specific statistical method for consecutive abnormalities.

[0058] It should be noted that the weighted continuous anomaly count value mentioned in this application refers to the count value obtained by accumulating the number of consecutively marked abnormal windows according to the weight of the abnormality type during real-time monitoring, which is used to measure the duration and risk level of abnormal activity status of elderly people living alone.

[0059] In specific implementation, the weighted continuous anomaly count value is compared with a preset multi-level warning threshold to determine the current warning level. This can be achieved in the following way: Multiple warning thresholds are set in ascending order, each corresponding to a warning level. A higher warning level indicates a longer duration of continuous anomalies and a more severe potential risk. After each update of the weighted continuous anomaly count value, the count value is compared sequentially with the multiple warning thresholds. Starting from the lowest level threshold, if the count value reaches or exceeds a certain level threshold but has not yet reached the next higher level threshold, the current warning level is determined to be that level. If the count value exceeds the highest level threshold, the current warning level is determined to be the highest level. If the count value is lower than the lowest level threshold, it is determined that the warning trigger condition has not been met, and the warning level is set to no warning state.

[0060] In specific implementation, the following methods can be used to conduct graded early warning monitoring of the activities of elderly people living alone by selecting the corresponding early warning response strategy according to the early warning level. For example, a set of corresponding early warning response strategies can be pre-configured for each early warning level. The early warning response strategy includes elements such as notification channels, notification recipients, notification content, and whether manual confirmation is required. When the current early warning level is determined, the system retrieves the early warning response strategy bound to the early warning level from the preset strategy library and executes the corresponding early warning action according to the strategy. For the first early warning level, a text reminder can be pushed to the preset family member's mobile terminal through the application to inform the elderly person that there is a slight abnormality in their current activity rhythm and suggest paying attention. For the second early warning level, the preset family member's contact number can be dialed through the automatic voice call system, and a voice notification containing information on the duration of the abnormality can be played, requesting the family member to actively contact the elderly person to confirm the situation. For the third early warning level, the preset emergency rescue number or community care center can be automatically dialed, and key information such as the elderly person's identity information, address, and continuous abnormal duration can be broadcast. In other embodiments, the monitoring frequency can also be adjusted or auxiliary sensing devices can be used for cross-validation according to different early warning levels. This application does not limit the specific form and execution method of the early warning response strategy.

[0061] It should be noted that the warning level mentioned in this application refers to the response level determined by comparing the weighted continuous abnormal count value with the preset multi-level threshold. It is used to distinguish the urgency of abnormal activity events of elderly people living alone. Different warning levels correspond to different attention and intervention strategies. The warning response strategy refers to a set of specific notification and intervention action plans that are pre-associated with each warning level. It is used to guide the system to perform response behaviors such as information push, communication contact or emergency call to family members or caregivers when the corresponding warning level is triggered, so as to achieve graded and appropriate warning intervention.

[0062] In another aspect, in some embodiments, this application provides a non-contact health monitoring device, including an activity anomaly monitoring unit, as referenced. Figure 4 The figure is a schematic diagram of the structure of an activity anomaly monitoring unit according to some embodiments of this application. The activity anomaly monitoring unit 200 includes: a data acquisition module 201, a processing module 202, and an execution module 203, which are described below: The acquisition module 201 in this application is mainly used to acquire amplitude data of Wi-Fi channel status information in the living area, preprocess it to extract transient amplitude components, and calculate Shannon entropy based on the distribution of the transient amplitude components of all subcarriers in the short time window to form a transient vitality index sequence. Processing module 202, in this application, is mainly used to process the instantaneous vitality index sequence through a sliding time window, determine the behavioral feature vector containing numerical dispersion, distribution symmetry and its deviation from the time series reference, and determine the proportion of vitality index frames exceeding the adaptive transient noise base within the window as the activity intensity index; In addition, the processing module 202 in this application is also used to train a single classification model using behavioral feature vectors from historical normal periods to define the boundaries of normal behavior, and to determine the lower limit of the activity intensity benchmark based on the time-segmented statistical distribution of the activity intensity index in historical normal periods. In addition, the processing module 202 in this application is also used to determine that if the behavior feature vector of the current window is outside the normal behavior boundary or the activity intensity index of the current window is lower than the activity intensity benchmark lower limit during real-time monitoring, the corresponding window is determined to be an abnormal window. When both conditions are met at the same time, the corresponding window is determined to be a high-risk rhythm abnormal window. The execution module 203 in this application is mainly used to perform hierarchical early warning monitoring of the activities of elderly people living alone based on the number of consecutive abnormal windows and the weight of abnormal types.

[0063] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described non-contact health monitoring method.

[0064] In some embodiments, reference Figure 5 This figure is an internal structural diagram of a computer device for implementing a non-contact health monitoring method according to some embodiments of this application. The non-contact health monitoring method in the above embodiments can be implemented through... Figure 5 The computer device shown is used to implement this, and the computer device 300 includes at least one processor 301, a communication bus 302, a memory 303, and at least one communication interface 304.

[0065] The processor 301 may be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of the non-contact health monitoring method in this application.

[0066] The communication bus 302 is used to transmit information between the aforementioned components.

[0067] Memory 303 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 303 may exist independently and be connected to processor 301 via communication bus 302. Memory 303 may also be integrated with processor 301.

[0068] The memory 303 stores program code for executing the solution of this application, and its execution is controlled by the processor 301. The processor 301 executes the program code stored in the memory 303. The program code may include one or more software modules. In the above embodiments, the non-contact health monitoring method can be implemented by the processor 301 and one or more software modules in the program code in the memory 303.

[0069] Communication interface 304 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0070] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0071] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device may be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0072] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described non-contact health monitoring method.

[0073] In summary, the non-contact health monitoring device, method, equipment, and medium disclosed in this application collect amplitude data of Wi-Fi channel status information in the living area, preprocess it to extract transient amplitude components, and calculate Shannon entropy based on the distribution of the transient amplitude components of all subcarriers within a short time window to form a transient vitality index sequence. The transient vitality index sequence is processed through a sliding time window to determine a behavioral feature vector containing numerical dispersion, distribution symmetry, and deviation from the time series reference, and to determine the proportion of vitality index frames exceeding the adaptive transient noise floor within the window, which serves as the activity intensity index. Historical normal time periods are utilized. A single-classification model is trained using behavioral feature vectors to define the boundaries of normal behavior, and the lower limit of activity intensity is determined based on the time-segmented statistical distribution of the activity intensity index during historical normal periods. During real-time monitoring, if the behavioral feature vector of the current window is outside the normal behavior boundary, or the activity intensity index of the current window is lower than the lower limit of the activity intensity benchmark, the corresponding window is determined to be an abnormal window. If both conditions are met simultaneously, the corresponding window is determined to be a high-risk rhythm abnormal window. Based on the number of consecutive abnormal windows and the weight of abnormal types, the activities of elderly people living alone are monitored and warned in a graded manner. This can effectively identify environmental steady-state channel jitter and transient channel disturbances caused by human behavior.

[0074] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0075] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A non-contact health monitoring method, applied to a non-contact health monitoring device for monitoring abnormal activities of elderly people living alone based on non-contact sensing, characterized in that, The method includes the following steps: Amplitude data of Wi-Fi channel status information in the living area is collected, preprocessed to extract transient amplitude components, and Shannon entropy is calculated based on the distribution of the transient amplitude components of all subcarriers within a short time window to form a transient vitality index sequence. The instantaneous vitality index sequence is processed by a sliding time window to determine a behavioral feature vector that includes numerical dispersion, distribution symmetry and deviation from the time series baseline, and to determine the proportion of vitality index frames exceeding the adaptive transient noise floor within the window, which is used as the activity intensity index. A single-classification model is trained using behavioral feature vectors from historical normal periods to define the boundaries of normal behavior, and the lower limit of the activity intensity benchmark is determined based on the time-segmented statistical distribution of the activity intensity index during historical normal periods. During real-time monitoring, if the behavior feature vector of the current window is outside the normal behavior boundary, or the activity intensity index of the current window is lower than the activity intensity benchmark lower limit, the corresponding window is determined to be an abnormal window. If both conditions are met at the same time, the corresponding window is determined to be a high-risk rhythm abnormal window. Based on the number of consecutive abnormal windows and the weight of abnormal types, the activities of elderly people living alone are monitored and warned in a tiered manner.

2. The method as described in claim 1, characterized in that, By processing the instantaneous vitality index sequence through a sliding time window, the behavioral feature vector, which includes numerical dispersion, distribution symmetry, and deviation from the time series benchmark, is determined. Specifically, this includes: The instantaneous vitality index values ​​of all frames within a preset fixed-duration sliding window are used to form a vitality sample set within the window. Calculate the statistical distribution characteristics of the live sample set within the window, wherein the statistical distribution characteristics include the standard deviation reflecting the numerical dispersion and the skewness reflecting the symmetry of the distribution; Calculate the deviation between the current window's statistical distribution characteristics and the time series benchmark of the corresponding period's historical normal mean; The standard deviation, the skewness, and the bias are combined to form a corresponding behavioral feature vector.

3. The method as described in claim 1, characterized in that, Determining the percentage of frames within the window that exceed the adaptive transient noise floor, as the activity intensity index, specifically includes: Calculate the bottom percentile statistics of the live sample set within the current sliding window and determine them as the adaptive transient noise base for that window; Traverse the set of vitality samples within the window, determine whether the instantaneous vitality index value of each frame is greater than the adaptive transient noise floor, and count the results. Divide the count result by the total number of frames contained in the sliding window to obtain the activity intensity index of the corresponding window.

4. The method as described in claim 1, characterized in that, Using behavioral feature vectors from historical normal periods to train a single-classification model to define the boundaries of normal behavior specifically includes: Obtain the behavioral feature vectors corresponding to all windows that match the current window time period identifier within the historical normal time period, and construct a training sample set of normal behavior by time period; The sample points in the time-segmented normal behavior training sample set are mapped to a high-dimensional feature space by a kernel function preset in the single classification model, and the minimum hypersphere containing the vast majority of sample points is obtained by solving the problem. The normal behavioral boundaries of elderly people living alone for a given period of time are obtained by fitting a closed surface in a high-dimensional space defined by the coordinates of the center of the minimum hypersphere and the radius parameters.

5. The method as described in claim 1, characterized in that, During real-time monitoring, if the behavior feature vector of the current window is outside the normal behavior boundary, or the activity intensity index of the current window is lower than the lower limit of the activity intensity benchmark, then the corresponding window is determined to be an abnormal window. The determination of a high-risk rhythmic abnormal window when both conditions are met simultaneously includes: Obtain the behavior feature vector and activity intensity index corresponding to the current sliding window; The behavioral feature vector is input into the decision function corresponding to the normal behavior boundary of the corresponding time period to obtain a first judgment result that characterizes the relative positional relationship between the behavioral feature vector and the normal behavior boundary. The activity intensity index is compared with the lower limit of the baseline for the corresponding time period to obtain a second judgment result indicating whether the activity intensity index is lower than the statistical baseline level; When the first judgment result indicates that the behavior feature vector is outside the normal behavior boundary, or when the second judgment result indicates that the activity intensity index is lower than the activity intensity benchmark lower limit, the corresponding window is determined to be an abnormal window; When the first judgment result indicates that the behavioral feature vector is outside the normal behavioral boundary, and the second judgment result indicates that the activity intensity index is lower than the lower limit of the activity intensity benchmark, the corresponding window is determined to be a high-risk rhythm abnormality window.

6. The method as described in claim 1, characterized in that, Based on the number of consecutive abnormal windows and the weight of abnormal types, the activity monitoring and early warning system for elderly people living alone is tiered and includes the following: The number of consecutive windows marked as anomalous before the current time is counted, and the anomaly type weight of each window is accumulated to obtain a weighted consecutive anomaly count value; The weighted continuous anomaly count value is compared with the preset multi-level early warning threshold to determine the current corresponding early warning level; Based on the aforementioned warning level, a corresponding warning response strategy is selected to conduct tiered warning monitoring of the activities of elderly people living alone.

7. The method as described in claim 1, characterized in that, The amplitude data refers to fine-grained physical layer measurements that reflect the degree of amplitude attenuation of a wireless signal after multipath propagation during its transmission between the transmitter and receiver.

8. A non-contact health monitoring device, comprising an activity anomaly monitoring unit, characterized in that, The activity anomaly monitoring unit includes: The acquisition module is used to collect amplitude data of Wi-Fi channel status information in the living area, preprocess it to extract transient amplitude components, and calculate Shannon entropy based on the distribution of the transient amplitude components of all subcarriers within a short time window to form a transient vitality index sequence. The processing module is used to process the instantaneous vitality index sequence through a sliding time window, determine the behavioral feature vector containing numerical dispersion, distribution symmetry and its deviation from the time series reference, and determine the proportion of vitality index frames exceeding the adaptive transient noise base within the window as the activity intensity index; The processing module is also used to train a single-classification model using behavioral feature vectors from historical normal periods to define the boundaries of normal behavior, and to determine the lower limit of the activity intensity benchmark based on the time-segmented statistical distribution of the activity intensity index during historical normal periods. The processing module is also used to determine that if the behavior feature vector of the current window is outside the normal behavior boundary or the activity intensity index of the current window is lower than the activity intensity benchmark lower limit during real-time monitoring, the corresponding window is determined to be an abnormal window. When both conditions are met at the same time, the corresponding window is determined to be a high-risk rhythm abnormal window. The execution module is used to perform hierarchical early warning monitoring of the activities of elderly people living alone based on the number of consecutive abnormal windows and the weight of the abnormal type.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the non-contact health monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the non-contact health monitoring method as described in any one of claims 1 to 7.

Images