A home abnormal behavior monitoring method and system

Abstract

The application provides a kind of home abnormal behavior monitoring method and system, it is related to Internet of Things and home security monitoring technical field, comprising the following steps: by multi-source sensor acquisition home monitoring multi-source data, and the pre-processing of home monitoring multi-source data, obtain standardized monitoring data;Standardized monitoring data is input into the preset fusion decision model, and the weighted fusion and risk quantization of standardized monitoring data are carried out using fusion decision model, and the abnormal risk coefficient is output;Time series statistics are carried out to abnormal risk coefficient, and time series statistical result is obtained;Abnormal risk coefficient threshold is defined, and time series anomaly identification is carried out based on time series statistical result and standardized monitoring data by threshold method, and abnormal behavior state is obtained;Based on abnormal behavior state, corresponding abnormal early warning information is generated, and abnormal early warning information is pushed to the care terminal.The application effectively improves the accuracy of home abnormal behavior identification and the reliability of real-time early warning.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things and home security monitoring technology, and in particular to a method and system for monitoring abnormal home behavior. Background Technology

[0002] With the deepening of population aging and the continuous increase in the number of elderly people living alone, abnormal behavior monitoring technology for home scenarios has gradually become an important research direction in the fields of smart elderly care, home security monitoring, and smart homes. Existing home monitoring technologies typically use cameras, infrared sensors, door magnetic sensors, millimeter-wave radar sensors, or wearable devices to detect the activity status of people, and combine wireless communication and edge computing technologies to achieve early warning of abnormal events. Among them, non-contact behavior monitoring technology based on millimeter-wave radar is gradually replacing traditional video surveillance solutions and is widely used in fall detection, loitering monitoring, and elderly care scenarios due to its advantages such as strong privacy protection, high environmental adaptability, and ability to sense subtle human movements. Meanwhile, the development of multi-source sensor fusion, behavioral temporal analysis, and intelligent risk assessment algorithms has enabled the recognition of abnormal home behavior to gradually evolve from single threshold judgment to multi-dimensional behavioral correlation analysis, thereby improving the ability to recognize behavior and provide real-time early warnings in complex scenarios.

[0003] However, existing technologies for monitoring abnormal home behavior still have significant limitations. First, most current technologies rely on a single sensor for behavior detection, such as using only infrared sensors to detect human activity or millimeter-wave radar to determine a fall. These technologies are easily affected by environmental obstructions, multipath reflections, pet activity, and furniture interference, leading to false alarms or missed alarms and hindering stable anomaly identification in complex home environments. Second, existing technologies typically make anomaly judgments based on instantaneous detection results, lacking statistical analysis of the continuity of behavior over time. This makes the system prone to false alarms triggered by short-term movements, occasional noise, or instantaneous posture changes, failing to effectively improve the stability and reliability of abnormal behavior identification. Finally, some existing solutions depend on video image acquisition or cloud-based behavior analysis, which not only poses privacy risks but also suffers from significant network transmission latency, making it difficult to meet the needs for real-time monitoring and rapid early warning of abnormal home behavior.

[0004] How to solve the above-mentioned technical problems is the challenge facing this invention. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for monitoring abnormal home behavior that effectively improves the accuracy of identifying such behavior and the reliability of real-time early warning.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows: This invention provides a method for monitoring abnormal home behavior, comprising the following steps: S1. Collect multi-source home monitoring data through multi-source sensors, and preprocess the multi-source home monitoring data to obtain standardized monitoring data; the multi-source sensors include a first sensing channel, a second sensing channel, and a third sensing channel; the standardized monitoring data includes human motion data, human presence status data, and door status data; S2. Input the standardized monitoring data into the preset fusion decision model, use the fusion decision model to perform weighted fusion and risk quantification on the standardized monitoring data, and output the abnormal risk coefficient; S3. Perform time-series statistics on the abnormal risk coefficients to obtain the time-series statistical results; S4. Define the threshold for the abnormal risk coefficient. Based on the time series statistical results and standardized monitoring data, use the threshold method to identify time series anomalies and obtain the abnormal behavior status. S5. Based on the abnormal behavior status, generate corresponding abnormal warning information and push the abnormal warning information to the care terminal.

[0007] Preferably, the first sensing channel uses a millimeter-wave radar sensor to collect human motion data including the amplitude of human micro-movements, the rate of change of posture angle, spatial three-dimensional coordinates, moving speed, and duration of stillness. The second sensing channel uses an infrared dual-mode sensor to collect human presence status data, including human presence confidence and movement behavior trigger markers. The third sensing channel uses a high-precision magnetoresistive door magnetic sensor, and the collected door status data includes the door opening and closing status and the door action timestamp.

[0008] Preferably, the preprocessing in step S1 includes: The movement speed, micro-motion amplitude, and attitude angle change rate in human motion data, as well as the human presence confidence in human presence state data, are mapped to a unified numerical range using the maximum-minimum normalization method. A three-point median filtering algorithm is then used to denoise the normalized data and remove isolated interference signals. The door state data retains its original dimensions.

[0009] Preferably, the fusion decision model in step S2 includes a radar behavior feature mapping subfunction, an infrared space unmanned probability subfunction, and a door magnetic state time-series distributed subfunction; The radar behavior feature mapping sub-function is configured to output motion risk feature values ​​based on the normalized human movement speed, micro-motion amplitude, and attitude angle change rate through the product coupling calculation of S-shaped growth curve, Gaussian membership function, and sine function. The infrared space unmanned probability sub-function is configured to calculate the indoor space vacancy probability through a negative exponential nonlinear transfer function based on the normalized human presence confidence and movement behavior trigger marker. The door magnetic state time-series distributed sub-function is configured to calculate the time-series risk value based on the door's open / closed state and the duration of the door being closed, using a hyperbolic tangent nonlinear time-series growth operator.

[0010] Preferably, step S2 involves using a fusion decision model to perform weighted fusion and risk quantification of standardized monitoring data, specifically including: The radar behavior feature mapping subfunction, the infrared space unmanned probability subfunction, and the gate magnetic state time-series distributed subfunction are assigned first weights, second weights, and third weights, respectively. Construct an infrared spatial collaborative existence factor based on the probability of indoor space vacancy; The abnormal risk coefficient is obtained by summing the product of the motion risk characteristic value with the first weight and the infrared space co-existence factor, the complementary value of the indoor space vacancy probability with the second weight and the infrared space co-existence factor, and the product of the time series risk value with the third weight.

[0011] Preferably, the sliding window method is used to perform time-series statistics on the abnormal risk coefficient in step S3.

[0012] Preferably, the abnormal behavior states in step S4 include risk of falling, prolonged stillness, prolonged stay in an area, and not leaving home for a long time; The risk of falling is determined based on whether the human body's movement speed, rate of change of posture angle, amplitude of micro-movement, and duration of stillness simultaneously reach a first preset abnormal threshold. The determination of prolonged stillness is based on the confidence level of the human body and whether the duration of stillness reaches a second preset threshold. The determination of timeout in the area is based on whether the change in spatial coordinates and the duration of time spent in the area reach a third preset threshold. The determination of whether someone has not left home for a long time is based on the door's open / closed status, the duration of the door being closed, and whether the confidence level of the presence of a human body meets the fourth preset threshold.

[0013] The present invention also provides a home abnormal behavior monitoring system, including The data acquisition and preprocessing module is used to collect multi-source data for home monitoring through multi-source sensors, and to preprocess the multi-source data for home monitoring to obtain standardized monitoring data. The risk coefficient calculation module is used to input standardized monitoring data into a preset fusion decision model, and use the fusion decision model to perform weighted fusion and risk quantification on the standardized monitoring data, and output the abnormal risk coefficient. The time series statistics module is used to perform time series statistics on the abnormal risk coefficients and obtain the time series statistics results. The anomaly detection module is used to define the anomaly risk coefficient threshold. Based on time-series statistical results and standardized monitoring data, it uses the threshold method to identify time-series anomalies and obtain the anomaly behavior status. The anomaly push module is used to generate corresponding anomaly warning information based on the abnormal behavior status and push the anomaly warning information to the care terminal.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method for monitoring abnormal home behavior.

[0015] The present invention also provides a computer storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method for monitoring abnormal home behavior.

[0016] The beneficial effects of this invention are: effectively improving the accuracy of identifying abnormal home behavior and the reliability of real-time early warning. Multi-dimensional data is collected through multiple sensing channels, and a fusion decision model is used to perform weighted fusion and nonlinear risk quantification of heterogeneous data, achieving collaborative verification of motion characteristics, existence status, and door timing information. When a channel is interfered with and generates a false signal, other channels can automatically filter out the anomaly through cross-verification, thereby significantly reducing the false alarm rate and improving the stability of anomaly identification in complex home environments. Sliding window timing statistics are applied to the anomaly risk coefficient, effectively eliminating the impact of instantaneous disturbances on the anomaly identification results. Simultaneously, by defining anomaly risk coefficient thresholds and combining them with a continuous periodic confirmation mechanism, anomaly judgment is triggered only when the risk continuously exceeds the threshold, thus greatly improving the reliability and robustness of abnormal behavior identification. Using millimeter-wave radar, infrared, and door magnetic sensors, etc., no image information is collected; the raw sensing data is only cached and processed in the local embedded main control unit and is never transmitted externally. All data preprocessing, fusion calculation, timing statistics, and anomaly judgment are completed on the on-chip edge computing unit, without relying on cloud computing power, achieving real-time monitoring and rapid early warning while ensuring user privacy and security. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating the method steps of the present invention.

[0018] Figure 2 This is a system module diagram of the present invention.

[0019] Figure 3 This is a diagram of the internal structure of a computer device according to Embodiment 3 of the present invention. Detailed Implementation

[0020] To clearly illustrate the technical features of this solution, the following detailed implementation method will be used to explain the solution.

[0021] Example 1: See Figure 1 As shown, this embodiment is a method for monitoring abnormal home behavior, including the following steps: S1. Collect multi-source home monitoring data through multi-source sensors, and preprocess the multi-source home monitoring data to obtain standardized monitoring data; the multi-source sensors include a first sensing channel, a second sensing channel, and a third sensing channel; the standardized monitoring data includes human motion data, human presence status data, and door status data; It should be noted that in this embodiment, the multi-source sensor includes a first sensing channel, a second sensing channel, and a third sensing channel, which physically correspond to a millimeter-wave radar sensor, an infrared dual-mode sensor, and a high-precision magnetoresistive door sensor, respectively. All of these sensors are deployed in key locations within the home monitoring area and are non-visual sensors, meaning they do not collect any image information throughout the process.

[0022] Furthermore, in this embodiment, the first sensing channel employs a millimeter-wave radar sensor, deployed in the center of the ceiling in a bedroom or bathroom, using frequency-modulated continuous wave detection technology. Its detection distance is set to 0.5m to 5m, and the detection angle is 120°. This sensing channel collects human motion data in real time, specifically including: human micro-motion amplitude (unit: cm / s), attitude angle change rate (unit: ° / s), spatial three-dimensional coordinates (unit: m), movement speed (unit: m / s), and static duration (unit: s). The static duration is calculated as follows: when the movement speed is no greater than 0.05m / s and the micro-motion amplitude is no greater than 0.05cm / s for three consecutive acquisition cycles, timing begins and ends when the movement speed or micro-motion amplitude exceeds this range.

[0023] It should also be noted that in this embodiment, the second sensing channel uses an infrared dual-mode sensor, deployed on the side wall of the bedroom headboard or the bathroom wall, employing a dual-mode detection method of pyroelectric infrared and passive infrared. The detection sensitivity is set to a medium level. This sensing channel outputs real-time human presence status data, specifically including: human presence confidence (value range [0,1]) and movement behavior trigger flags (values ​​of 0 or 1, where 1 indicates someone is moving and 0 indicates no one is moving). The human presence confidence is calculated by weighted fusion of the presence probability output by the pyroelectric infrared sensor and the presence probability output by the passive infrared sensor, with weighting coefficients of 0.6 and 0.4 respectively.

[0024] Furthermore, in this embodiment, the third sensing channel employs a high-precision magnetoresistive door magnetic sensor, installed at the junction of the door frame and the door leaf, utilizing the magnetoresistive effect detection principle. The threshold for determining the door's open / closed state is set to 50μT. This sensing channel collects door status data in real time, specifically including: the door's open / closed state (values ​​of 0 or 1, where 0 indicates the door is closed and 1 indicates it is open) and the door's action timestamp (unit: seconds, accurate to milliseconds). The door's action timestamp is directly synchronized to the embedded main control unit's real-time clock. The moment the door state switches from closed to open or from open to closed, the current real-time clock time is recorded as the action timestamp.

[0025] It should also be noted that in this embodiment, the data acquisition of all sensing channels is uniformly scheduled by the embedded main control unit (in this embodiment, an STM32 series microcontroller is used). The main control unit starts a timer, sets the timing period to 100ms, and uses the built-in real-time clock as the time base to add timestamps to the data acquired by each sensing channel and perform alignment processing to ensure that the data acquisition timestamp error does not exceed 5ms when the three sensing channels work in parallel, thereby providing time consistency guarantee for subsequent multi-source data fusion and time series analysis.

[0026] Furthermore, preprocessing includes normalization and median filtering for noise reduction. Specifically: The normalization process employs a max-min normalization method to map the original data of different physical dimensions to the [0,1] standard interval. Normalization is performed on the movement speed, micro-motion amplitude, and attitude angle change rate output from the first sensing channel, as well as the confidence level of human presence output from the second sensing channel. The door opening / closing status and door action timestamp output from the third sensing channel, being binary state or time reference data, are not included in the normalization process, preserving their original dimensions to ensure the accuracy of timing determination.

[0027] Furthermore, the median filtering denoising process employs a three-point median filtering algorithm, filtering all normalized data point by point. This algorithm takes normalized data from three consecutive acquisition cycles and calculates the median value as the effective output value for the current cycle. This algorithm effectively eliminates isolated interference signals caused by environmental temperature fluctuations, object obstruction, or small pet activity, while also removing outliers caused by sensor malfunctions, retaining effective feature data reflecting the true behavioral state of the target object.

[0028] It should also be noted that the standardized monitoring data obtained after the above preprocessing includes: normalized human movement speed. Micro-motion amplitude Rate of change of attitude angle Confidence level of human existence And the door opening / closing state that maintains its original dimensions. Door body action timestamp All standardized monitoring data is cached only in the main control unit's built-in Flash memory, with a 7-day cache period and automatic cyclic overwriting. No raw sensing data is transmitted externally, eliminating the risk of privacy leaks throughout the entire process from data collection and processing to storage.

[0029] Through the above step S1, this embodiment realizes the synchronous collection, standardized preprocessing and time alignment of multi-source data for home monitoring, providing high-quality, highly consistent and highly privacy-secure input data for the subsequent steps S2 to S5 of weighted fusion, risk quantification, time series statistics and anomaly identification.

[0030] S2. Input the standardized monitoring data into the preset fusion decision model, use the fusion decision model to perform weighted fusion and risk quantification on the standardized monitoring data, and output the abnormal risk coefficient; In step S2, the main control unit receives the standardized monitoring data output in step S1 in real time and inputs it into the preset fusion decision model in parallel. The fusion decision model is used to perform multi-dimensional cross-linking verification and nonlinear risk quantification mapping on the multi-source heterogeneous standardized monitoring data, thereby calculating and outputting the abnormal risk coefficient to characterize the degree of risk of the current instantaneous state deviating from the normal behavior pattern.

[0031] It should be noted that the fusion decision model in this embodiment relies directly on the on-chip computing resources of the embedded main control unit (the STM32 series microcontroller used in this embodiment) for efficient operation at the physical level. Its software core includes three specific decision sub-functions that are computed in parallel and decoupled within the current acquisition cycle, corresponding to radar behavior feature mapping sub-functions. Infrared Space Unmanned Probability Subfunction and gate magnet state time-distributed subfunction The risk and state scalar values ​​for different observation dimensions are extracted through the three sub-functions mentioned above.

[0032] Furthermore, in order to improve the sensitivity of detecting sudden falls and other severe instability behaviors, and to eliminate the risk of missed detections and artifacts caused by the traditional crude hard shear threshold at critical boundaries, the radar behavior feature mapping subfunction... Instead of simple binary determination, a continuous nonlinear composite operator based on the fusion of the sigmoid growth curve and the Gaussian membership function is adopted. Its explicit mathematical analytical expression is shown below:

[0033] It should also be noted that the above-mentioned radar behavior feature mapping sub-function This achieves the product coupling of velocity characteristics, micro-motion characteristics, and attitude characteristics at the instantaneous dynamics level. Among these, For speed risk activation, a preset gain coefficient is used. center threshold When the normalized human movement speed is output in step S1 Once the value exceeds 0.4, the risk value increases exponentially and steeply. The probability gain term for unconscious stillness after falling to the ground is preset with a central value. Standard deviation When the amplitude of the micro-motion The closer the value is to 0.01 (meaning that after falling, the body loses the ability to struggle or resist, and only the slight fluctuations caused by breathing remain), the closer the calculation result is to the upper limit of 1.0, thus positively amplifying the overall risk of falling. Then the rate of change of attitude angle A sinusoidal nonlinear space mapped to the first quadrant is used to dynamically and nonlinearly weight the sudden body fall process. This continuous composite operator greatly enhances the algorithm's depth of low-level extraction of weak, gradual, and sudden fall features.

[0034] Furthermore, the unmanned probability subfunction in infrared space It primarily provides dynamic cross-validation from the perspective of environmental conditions. It employs a negative exponential nonlinear transfer function with a flexible edge penalty mechanism to automatically calculate the vacancy probability of indoor spaces. This corrects and compensates for measurement deviations caused by false micro-motion signals generated by single-channel millimeter-wave radar sensors due to multipath reflections, occasional environmental clutter, or obstructions from bedding and furniture. The calculation formula is defined as follows:

[0035] It should be noted that sub-functions The input is entirely derived from the normalized confidence level of the human body in the standardized monitoring data output from step S1. and movement behavior trigger flags Among them, the preset environmental attenuation coefficient This function mathematically constructs a flexible isolation band using the negative exponential square operator: when the infrared channel outputs a confidence level of the human body... Approaching 0 and movement trigger flag When the value is 0, it indicates that there is a very high probability that there are no living organisms in the indoor physical space. The output value of this function quickly approaches the absolute upper limit of 1.0, that is, the current space vacancy index is determined to be 1.0; while when... When fluctuating in the medium range and triggered by activity, the function provides a smooth verification transition through a negative exponential curve, outputting an extremely low idle probability, effectively avoiding sudden misjudgments caused by sensor mutation noise, and perfectly realizing the active coordination and complementarity between radar and infrared channels in terms of physical mechanism.

[0036] Furthermore, the gate magnet state time-distributed subfunction It is mainly used to extract the temporal distribution trend of the door opening and closing status on a long-period time axis. It utilizes the current system real-time clock time minus the timestamp of the most recent door action automatically recorded in step S1. Calculate the current duration of the closed door. To better suit the individualized routines of elderly people living alone and to deeply explore the risks of long-term regional confinement and disability, sub-functions... The hyperbolic tangent (Tanh) nonlinear time-series growth operator is adopted, and its mapping rule is configured as follows:

[0037] It should also be noted that sub-functions Preset timing scaling factor Basic security duration boundary (That is, corresponding to 8 hours). When the main control unit determines the door's open / closed status... When the input variable is 0 (i.e., the door is closed) and the current closed duration does not exceed 8 hours, the hyperbolic tangent function calculation result is negative because the input variable is less than 0. The system automatically rectifies and truncates this result to 0.0 during local execution, thus not triggering a timing risk. However, when the closed duration exceeds 8 hours, as the elderly person remains indoors in a closed environment, the hyperbolic tangent function begins to enter a non-linear rapid growth phase, approaching its upper limit of 1.0 near 12 hours (43200 seconds). This design highly simulates the non-linear positive correlation between the passage of daily routines of elderly people living alone and the potential probability of encountering danger while the door is closed, providing time-deep technical support for subsequent anomaly detection.

[0038] Furthermore, after the main control unit calculates and obtains the radar behavior characteristic function value, the infrared spatial unmanned probability function value, and the door magnetic state time-series distributed function value corresponding to the current acquisition cycle, the fusion decision model calls the preset personalized weight coefficient matrix to perform spatial weighted fusion calculation. Even further, to reflect the linkage and constraint relationships between different heterogeneous sensors, this embodiment not only introduces basic channel weights... (Corresponding radar channel) (Corresponding infrared channel) (Corresponding to the magnetic gate channel), a dynamic intermodulation penalty mechanism for the spatial presence factor is creatively constructed by utilizing the probability of no one in the infrared space. The final composite fusion calculation formula is shown below:

[0039] It should be noted that the main control unit performs cycle-by-cycle compound multiplication and superposition by running the aforementioned fusion calculation formula that includes a dynamic collaborative penalty factor. Among these, Defined as an infrared spatial co-existence factor. When the infrared channel determines that there is indeed no one indoors (i.e., ... When it approaches 1.0, due to the existence of a co-existence factor... Forcing the value close to 0 will exponentially eliminate radar channels at the multiplicative level. The high-risk indicators caused by multipath reflection environmental clutter also reduce the inherent channel risk of infrared to zero, thereby lowering the single-channel false alarm rate to near the theoretical limit of zero; while when the infrared and radar channels generate high-risk resonance in the same direction (i.e. Approaching 0, the co-existence factor becomes 1, and Approaching 1), the spatial linkage term releases its maximum gain, and the anomaly risk coefficient rapidly approaches the absolute high-risk value of 1.0, achieving a highly robust environmental adaptive defense. The anomaly risk coefficient ultimately output by the main control unit is a continuously varying scalar value within the range [0,1].

[0040] It should also be noted that, to ensure extremely high algorithm execution efficiency and extremely low local decision latency, the complex nonlinear composite model described above, when running in the STM32 microcontroller, does not employ complex online Taylor series expansions or floating-point series approximations for the natural exponent, sine function, and hyperbolic tangent function. Instead, a one-dimensional static mathematical look-up table (LUT) for the corresponding function is pre-built in the on-chip Flash memory of the main control chip during the system initialization phase. In each sampling cycle, the microcontroller only needs to perform low-overhead addressing and bilinear interpolation using standardized monitoring data to obtain the function results. This keeps the on-chip computation time for a single weighted fusion and risk quantification operation within 2.8 microseconds. This not only provides ample local computing power margin for the overall decision but also ensures that the abnormal risk coefficient output in step S2 has extremely high real-time responsiveness.

[0041] Through step S2 described above, this embodiment achieves cross-channel correlation risk quantification of the multi-channel standardized monitoring data output in step S1 across multiple spatial and temporal dimensions. It utilizes highly interpretable continuous mathematical operators to replace the traditional, brute-force threshold-based judgment, fundamentally eliminating artifact false alarms and boundary false alarms. This provides a fully standardized, highly consistent, and highly deterministic core input source for the subsequent time-series analysis and final anomaly state determination in steps S3 and S4.

[0042] S3. Perform time-series statistics on the abnormal risk coefficients to obtain the time-series statistical results; It should be noted that the abnormal risk coefficient output in step S2 is the instantaneous comprehensive risk value corresponding to each acquisition cycle. This instantaneous value is susceptible to isolated fluctuations caused by instantaneous environmental disturbances, short-term sensor noise, or accidental movements of the target object. In order to eliminate the above-mentioned instantaneous interference and improve the stability and reliability of abnormal behavior judgment, this step performs time-series statistical processing on the abnormal risk coefficients of multiple consecutive acquisition cycles.

[0043] Furthermore, this step employs a sliding window timing statistics method. Specifically, the duration of the sliding window is set to 10 minutes, and the window sliding step size is 1 minute. The embedded main control unit maintains a first-in-first-out queue in its local memory to cache the abnormal risk coefficients corresponding to all acquisition cycles within the current sliding window.

[0044] It should also be noted that, since the acquisition period is 100 milliseconds, each sliding window contains a total of 6000 acquisition periods. After each acquisition period, the main control unit pushes the newly calculated anomaly risk coefficient to the tail of the queue and pops the oldest data from the head of the queue, ensuring that the queue always stores 6000 consecutive anomaly risk coefficient values ​​within the last 10 minutes.

[0045] It should be noted that the 10-minute sliding window duration and 1-minute step size are both optimized based on the time scale of the home behavior of elderly people living alone. The 10-minute window duration is sufficient to cover most daily short-term actions without excessively delaying the recognition and response to continuous abnormalities. The 1-minute step size ensures that abnormal state changes can be captured at the minute-level granularity, meeting the need for timely early warning.

[0046] Through the above step S3, this embodiment converts the high-frequency instantaneous abnormal risk coefficient output in step S2 into a low-frequency stable time-series statistical result, effectively filtering out false risk fluctuations caused by environmental noise and accidental actions, and providing highly stable and reliable input data for abnormal behavior identification based on time-series statistical results and personalized behavior benchmark thresholds in the subsequent step S4.

[0047] S4. Define the threshold for the abnormal risk coefficient. Based on the time series statistical results and standardized monitoring data, use the threshold method to identify time series anomalies and obtain the abnormal behavior status. It should be noted that step S4 in this embodiment is used to identify time-series anomalies in the time series of abnormal risk coefficients output in step S3, and, combined with the standardized monitoring data in step S1, to determine the type of abnormal behavior in the current behavioral state of the target object. By establishing a synergistic relationship between the comprehensive risk assessment mechanism and the behavioral feature classification mechanism, the hierarchical identification and type differentiation of abnormal home behaviors can be achieved.

[0048] Furthermore, in this embodiment, the timing anomaly identification is performed locally by the embedded main control unit, and all abnormal behavior identification processes are completed on the edge side without relying on cloud servers for behavior analysis, thereby improving the real-time performance of anomaly identification and avoiding the risk of privacy leakage caused by the external transmission of behavior data.

[0049] It should also be noted that the abnormal risk coefficient threshold in this embodiment is pre-generated by statistical analysis of historical normal home behavior data, and can be dynamically updated according to the long-term behavioral change patterns of the target object.

[0050] Furthermore, the main control unit first performs a continuous time-series scan based on the time series of abnormal risk coefficients output in step S3, and then compares the thresholds of the abnormal risk coefficients corresponding to each acquisition cycle in chronological order.

[0051] Specifically, when the abnormal risk coefficient exceeds the abnormal risk coefficient threshold in multiple consecutive collection cycles, the main control unit determines that the target object's current behavior state has an abnormal behavior risk and initiates the abnormal behavior classification and identification process.

[0052] It should also be noted that, in this embodiment, to avoid misjudgment due to instantaneous noise fluctuations, environmental interference, or short-term random actions, the main control unit adopts a continuous periodic confirmation mechanism to verify the validity of anomalies. Specifically, an abnormal behavior is determined to be established only when the anomaly risk coefficient exceeds the corresponding risk threshold for a consecutive preset number of collection periods; if the anomaly risk coefficient falls back below the corresponding risk threshold within the continuous confirmation period, the current anomaly determination is cancelled.

[0053] Furthermore, after determining that there is a risk of abnormal behavior, the main control unit further combines the standardized monitoring data in step S1 to classify and identify the types of abnormal behavior.

[0054] Furthermore, the abnormal behavior types in this embodiment include abnormal fall behavior, abnormal lingering behavior, abnormal departure behavior, and abnormal nighttime wandering behavior.

[0055] The rules for determining abnormal fall behavior include: The human body's movement speed exceeds a preset speed threshold; the rate of change of posture angle exceeds a preset posture change threshold; the amplitude of human body micro-movement is lower than a preset micro-movement threshold; and the above states continue for a preset confirmation time.

[0056] When the above conditions are met simultaneously, the main control unit determines that the target object has an abnormal falling behavior state.

[0057] Furthermore, the rules for determining abnormal loitering behavior include: The confidence level of the presence of a human body is consistently higher than the preset presence threshold; the door remains closed; and the duration of the door remaining stationary exceeds the preset dwell time threshold.

[0058] When the above conditions are met, the main control unit determines that the target object has an abnormal lingering behavior.

[0059] Furthermore, the rules for determining abnormal departure behavior include: The door is opened; the confidence level of the human presence is lower than the preset presence threshold; the movement behavior trigger flag remains in an unattended state; and the duration exceeds the preset departure confirmation time.

[0060] When the above conditions are met, the main control unit determines that the target object has an abnormal departure behavior.

[0061] Furthermore, the rules for determining abnormal nocturnal wandering behavior include: The current system time is within the preset nighttime monitoring period; the human movement speed frequently exceeds the movement threshold within the preset time window; the attitude angle change rate fluctuates continuously within multiple acquisition cycles; and the door status changes frequently during the nighttime monitoring period.

[0062] When the above conditions are met, the main control unit determines that the target object is exhibiting abnormal wandering behavior at night.

[0063] It should also be noted that each preset threshold in this embodiment was obtained by cluster analysis of historical normal home behavior data.

[0064] Furthermore, the main control unit performs behavioral clustering on the human movement speed, human micro-movement amplitude, posture angle change rate, human presence confidence, and door action interval time in the historical normal behavior data, and generates the corresponding normal behavior distribution interval based on the clustering results.

[0065] Furthermore, when the real-time monitoring data deviates from the corresponding normal behavior distribution range to a preset degree, the main control unit determines that the corresponding behavioral characteristics have an abnormal behavioral tendency.

[0066] It should also be noted that the behavioral thresholds in this embodiment can be automatically adjusted according to the long-term behavioral patterns of the target object.

[0067] For example, when a target object exhibits nighttime wake-up behavior over a long period, the system automatically increases the action frequency threshold corresponding to abnormal nighttime wandering behavior; when a target object exhibits prolonged sitting behavior over a long period, the system automatically extends the duration threshold corresponding to abnormal loitering behavior, thereby improving the ability of anomaly identification to adapt to individual behavioral differences.

[0068] Through the above step S4, this embodiment realizes multi-dimensional time-series anomaly identification based on anomaly risk coefficient time series and standardized monitoring data. It can accurately classify and identify different types of abnormal home behaviors while ensuring a low false alarm rate, thereby improving the system's environmental adaptability, behavior differentiation ability and anomaly identification reliability in complex home behavior scenarios.

[0069] S5. Based on the abnormal behavior status, generate corresponding abnormal warning information and push the abnormal warning information to the care terminal.

[0070] It should be noted that, in this embodiment, step S5 is used to convert the determined abnormal behavior status into early warning information that can be understood by caregivers or monitoring systems after the abnormal behavior identification is completed in step S4, and push it to the remote care terminal in real time through a secure communication channel.

[0071] Furthermore, the generation of abnormal early warning information includes the following: Abnormal behavior type identification: Based on the abnormal behavior type identified in step S4, such as falling, abnormal staying, abnormal leaving, or abnormal wandering at night, the abnormal type is clearly identified in the warning information to facilitate the receiving end to quickly determine the nature of the event. Abnormal Behavior Time Recording: At the moment an abnormal event occurs, the corresponding system time is recorded, including the start and end times, to provide evidence of the timing of the behavior and a reference for its duration. Monitoring area identifier: The early warning information includes the specific monitoring area or room number where the abnormal behavior occurred, so that caregivers can quickly locate the target object. Optional additional information: Optionally, the warning information may also include a summary of sensor channel data that triggered the anomaly, such as key parameters of the human motion channel, human presence channel, and door status channel, to assist in remote analysis and assessment of the severity of the abnormal behavior. It should also be noted that the generation of early warning information is completed locally by the embedded main control unit, without relying on external servers for data processing, thereby ensuring that abnormal event information can be generated in a timely manner and guaranteeing high timeliness.

[0072] Furthermore, the generated anomaly warning information is transmitted in real time to the monitoring terminal via a pre-configured wireless communication link, such as Wi-Fi, Bluetooth Low Energy, or cellular network. The monitoring terminal can be a mobile terminal, tablet, or monitoring system interface, which receives and displays the anomaly information and can trigger sound, vibration, or visual cues to remind caregivers to respond promptly.

[0073] It should also be noted that, in order to prevent the loss of early warning information due to network anomalies or communication delays, this embodiment retains an abnormal early warning information cache locally in the main control unit. When the communication channel is restored, the early warning data that was not successfully transmitted can be resent to ensure complete recording and reliable notification of abnormal events.

[0074] Through the above step S5, this embodiment realizes the real-time conversion and remote notification function of the abnormal behavior status identified in step S4, which not only ensures the immediate response capability of the home monitoring system, but also improves the safety protection level of people living alone or at high risk.

[0075] Example 2: See Figure 2 As shown, this embodiment is a home abnormal behavior monitoring system, including: The data acquisition and preprocessing module is used to collect multi-source data for home monitoring through multi-source sensors, and to preprocess the multi-source data for home monitoring to obtain standardized monitoring data. The risk coefficient calculation module is used to input standardized monitoring data into a preset fusion decision model, and use the fusion decision model to perform weighted fusion and risk quantification on the standardized monitoring data, and output the abnormal risk coefficient. The time series statistics module is used to perform time series statistics on the abnormal risk coefficients and obtain the time series statistics results. The anomaly detection module is used to define the anomaly risk coefficient threshold. Based on time-series statistical results and standardized monitoring data, it uses the threshold method to identify time-series anomalies and obtain the anomaly behavior status. The anomaly push module is used to generate corresponding anomaly warning information based on the abnormal behavior status and push the anomaly warning information to the care terminal.

[0076] Example 3: This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0077] This computer device can be a server, and its internal structure diagram can be as follows: Figure 3As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores server data. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a method for monitoring abnormal home behavior.

[0078] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0079] Example 4: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0080] If the functions implemented by the method are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art or the current technical solution, can be embodied in the form of a software product. This current computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0081] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0082] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0083] The technical features of this invention not described can be implemented by or using existing technology, and will not be repeated here. Of course, the above description is not a limitation of this invention, and this invention is not limited to the examples above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of this invention should also be within the protection scope of this invention.

Claims

1. A method for monitoring abnormal behavior at home, characterized in that, Includes the following steps: S1. Collect multi-source home monitoring data through multi-source sensors, and preprocess the multi-source home monitoring data to obtain standardized monitoring data; the multi-source sensors include a first sensing channel, a second sensing channel, and a third sensing channel; the standardized monitoring data includes human motion data, human presence status data, and door status data; S2. Input the standardized monitoring data into the preset fusion decision model, use the fusion decision model to perform weighted fusion and risk quantification on the standardized monitoring data, and output the abnormal risk coefficient; S3. Perform time-series statistics on the abnormal risk coefficients to obtain the time-series statistical results; S4. Define the threshold for the abnormal risk coefficient. Based on the time series statistical results and standardized monitoring data, use the threshold method to identify time series anomalies and obtain the abnormal behavior status. S5. Based on the abnormal behavior status, generate corresponding abnormal warning information and push the abnormal warning information to the care terminal.

2. The method for monitoring abnormal home behavior according to claim 1, characterized in that, The first sensing channel uses a millimeter-wave radar sensor to collect human motion data including the amplitude of human micro-movements, the rate of change of posture angle, spatial three-dimensional coordinates, moving speed, and duration of stillness. The second sensing channel uses an infrared dual-mode sensor to collect human presence status data, including human presence confidence and movement behavior trigger markers. The third sensing channel uses a high-precision magnetoresistive door magnetic sensor, and the collected door status data includes the door opening and closing status and the door action timestamp.

3. The method for monitoring abnormal home behavior according to claim 2, characterized in that, The preprocessing in step S1 includes: The movement speed, micro-motion amplitude, and attitude angle change rate in human motion data, as well as the human presence confidence in human presence state data, are mapped to a unified numerical range using the maximum-minimum normalization method. A three-point median filtering algorithm is then used to denoise the normalized data and remove isolated interference signals. The door state data retains its original dimensions.

4. The method for monitoring abnormal home behavior according to claim 3, characterized in that, The fusion decision model in step S2 includes a radar behavior feature mapping sub-function, an infrared space unmanned probability sub-function, and a gate magnetic state time-distributed sub-function. The radar behavior feature mapping sub-function is configured to output motion risk feature values ​​based on the normalized human movement speed, micro-motion amplitude, and attitude angle change rate through the product coupling calculation of S-shaped growth curve, Gaussian membership function, and sine function. The infrared space unmanned probability sub-function is configured to calculate the indoor space vacancy probability through a negative exponential nonlinear transfer function based on the normalized human presence confidence and movement behavior trigger marker. The door magnetic state time-series distributed sub-function is configured to calculate the time-series risk value based on the door's open / closed state and the duration of the door being closed, using a hyperbolic tangent nonlinear time-series growth operator.

5. The method for monitoring abnormal home behavior according to claim 4, characterized in that, Step S2 involves using a fusion decision model to perform weighted fusion and risk quantification of standardized monitoring data, specifically including: The radar behavior feature mapping subfunction, the infrared space unmanned probability subfunction, and the gate magnetic state time-series distributed subfunction are assigned first weights, second weights, and third weights, respectively. Construct an infrared spatial collaborative existence factor based on the probability of indoor space vacancy; The abnormal risk coefficient is obtained by summing the product of the motion risk characteristic value with the first weight and the infrared space co-existence factor, the complementary value of the indoor space vacancy probability with the second weight and the infrared space co-existence factor, and the product of the time series risk value with the third weight.

6. The method for monitoring abnormal home behavior according to claim 5, characterized in that, In step S3, the time-series statistics of the abnormal risk coefficient are performed using the sliding window method.

7. The method for monitoring abnormal home behavior according to claim 6, characterized in that, The abnormal behavior states in step S4 include risk of falling, prolonged stillness, prolonged stay in an area, and not leaving the house for a long time. The risk of falling is determined based on whether the human body's movement speed, rate of change of posture angle, amplitude of micro-movement, and duration of stillness simultaneously reach a first preset abnormal threshold. The determination of prolonged stillness is based on the confidence level of the human body and whether the duration of stillness reaches a second preset threshold. The determination of timeout in the area is based on whether the change in spatial coordinates and the duration of time spent in the area reach a third preset threshold. The determination of whether someone has not left home for a long time is based on the door's open / closed status, the duration of the door being closed, and whether the confidence level of the presence of a human body meets the fourth preset threshold.

8. A home abnormal behavior monitoring system, characterized in that, The steps for performing the home abnormal behavior monitoring method according to any one of claims 1 to 7 include: The data acquisition and preprocessing module is used to collect multi-source data for home monitoring through multi-source sensors, and to preprocess the multi-source data for home monitoring to obtain standardized monitoring data. The risk coefficient calculation module is used to input standardized monitoring data into a preset fusion decision model, and use the fusion decision model to perform weighted fusion and risk quantification on the standardized monitoring data, and output the abnormal risk coefficient. The time series statistics module is used to perform time series statistics on the abnormal risk coefficients and obtain the time series statistics results. The anomaly detection module is used to define the anomaly risk coefficient threshold. Based on time-series statistical results and standardized monitoring data, it uses the threshold method to identify time-series anomalies and obtain the anomaly behavior status. The anomaly push module is used to generate corresponding anomaly warning information based on the abnormal behavior status and push the anomaly warning information to the care terminal.

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 home abnormal behavior monitoring method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the home abnormal behavior monitoring method according to any one of claims 1 to 7.

Images