A monitoring system and method for solitary living based on multi-source data fusion and non-contact sensing

Abstract

The present application belongs to the technical field of wisdom pension, and particularly relates to a lonely monitoring system and method based on multi-source data fusion and non-contact sensing, which solves the problems of high false alarm and missed alarm rate, poor compliance and invasion of privacy. The lonely monitoring method comprises the following steps: collecting water, electricity and gas consumption data and non-contact life sensing signals such as millimeter wave radar, infrared and door magnetic in real time; a controller performs space-time alignment on different frequency data based on the minimum transmission rate as the benchmark period, and constructs a multi-feature correlation matrix; an individual behavior baseline model is established based on historical data; the real-time matrix is calculated based on the baseline deviation, and the alarm is triggered after the abnormality is confirmed by the auxiliary verification camera with a physical baffle directly or when the deviation is continuous. The all-weather monitoring is realized without feeling and zero compliance, the false alarm rate is reduced through the multi-modal logic mutual verification of water, electricity, gas and space sensing, and the privacy is protected by the auxiliary verification camera with normal power failure and physical baffle.

Description

Technical Field

[0001] This invention belongs to the field of smart elderly care technology, and specifically relates to a monitoring system and method for living alone based on multi-source data fusion and non-contact sensing. Background Technology

[0002] With the aging process accelerating, the number of elderly people living alone or in empty nests is growing explosively. A considerable proportion of the elderly live alone or are unattended during the day. Due to natural aging and age-related diseases, the elderly are prone to sudden cardiovascular and cerebrovascular diseases, falls, coma and other accidents. Tragic cases caused by failure to receive timely help are also common.

[0003] Currently, in order to enhance the monitoring of elderly people living alone, some existing technologies use wearable devices to collect vital signs through accelerometers, heart rate sensors, etc., while others use video surveillance technology to collect images using cameras and use computer vision algorithms to identify falls and determine postures.

[0004] However, these existing technologies suffer from inconvenience and privacy issues. Wearable devices are prone to being forgotten by the elderly, they resist charging, and they often remove them in high-risk situations such as bathing or sleeping, causing monitoring to fail at crucial moments and impacting comfort. Video surveillance technology raises privacy concerns, particularly the deployment of cameras in highly private areas like bedrooms and bathrooms, which severely infringes on personal privacy and can easily trigger strong psychological resistance from the elderly. Summary of the Invention

[0005] The purpose of this invention is to address the aforementioned problems in the existing technology by proposing a method for monitoring isolated dwellings based on multi-source data fusion and non-contact sensing.

[0006] The objectives of this invention can be achieved through the following technical solutions:

[0007] A method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing includes the following steps:

[0008] S1. Collect real-time data on the usage of practical living facilities, and obtain the usage status of at least one of water, electricity, and gas.

[0009] S2. Real-time acquisition of non-contact life sensing signals, including at least one of the following: pyroelectric characteristic data representing body micro-movement and spatial movement, millimeter-wave radar data, and door magnetic status data;

[0010] S3. The controller dynamically extracts the data collected in steps S1 and S2 using different original sampling frequencies, and establishes a multi-feature correlation matrix aligned on a unified time axis with the lowest transmission rate as the basic period.

[0011] S4. Based on the rolling data of the multi-feature correlation matrix stored in history, construct a baseline model representing the daily habits of the monitored object;

[0012] S5. Perform deviation calculation on the multi-feature correlation matrix of step S3 and the benchmark model of step S4. If the deviation from the benchmark model persists for a certain period of time, directly or enable the auxiliary verification camera to acquire thermal imaging or real-time image data to assist in verification. If an anomaly is confirmed, trigger the alarm mechanism.

[0013] This method for monitoring the safety of elderly people living alone is specifically designed for home security monitoring. Steps S1 and S2 collect data from two completely different physical sources: one from household utilities and the other from spatial non-contact sensors. However, the original sampling frequencies of these two types of data differ significantly. Step S3 balances these two sampling frequencies by dynamically extracting data through a controller and using a matching mechanism with the lowest transmission rate as the basic cycle. This achieves spatiotemporal alignment of heterogeneous data under low-computing-power conditions at the edge, constructing a multi-feature correlation matrix on a unified time axis, reducing computational costs and improving response speed. Step S4 uses rolling historical data to construct a personalized baseline model that learns each elderly person's unique lifestyle patterns and uses this as a judgment benchmark. Step S5 calculates the deviation between the real-time multi-feature correlation matrix and the baseline model. If the deviation persists, as a preferred method, an auxiliary verification camera can be used for further confirmation of the anomaly, or an alarm mechanism can be directly triggered. Furthermore, this auxiliary verification camera is kept disconnected under normal conditions through power outages or physical obstruction to prevent privacy issues. Overall, this method of monitoring individuals living alone is largely imperceptible to their daily lives. Furthermore, the use of an auxiliary verification camera reduces the false alarm rate. This auxiliary verification camera is only activated during the verification phase, ensuring the privacy of the monitored individuals and eliminating any resistance to cameras.

[0014] In the above-mentioned method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing, step S1 specifically includes:

[0015] Instantaneous current and voltage waveforms are obtained through a smart AMI architecture power meter, and home appliances are classified using NILM monitoring technology;

[0016] The water usage status is obtained through a smart water meter on the main water inlet pipe, which includes the water usage rate and duration.

[0017] Gas usage status is obtained through smart gas meters on natural gas pipelines, including the gas usage rate and duration.

[0018] Step S1 of this method for monitoring the use of water, electricity, and gas is mainly used to detect the usage of these three components. The AMI architecture power meter can acquire high-frequency sampling instantaneous current and voltage waveforms, the smart water meter can detect the water flow rate and duration, and the smart gas meter can detect the gas consumption rate and duration. Based on this data, the appliances are classified and identified using NILM non-invasive load monitoring technology. This technology is common knowledge and can obtain indirect evidence of the elderly's daily activities, such as boiling water, cooking, bathing, and heating, without the need to install additional sensors indoors.

[0019] In the above-mentioned method for monitoring solitary living based on multi-source data fusion and non-contact sensing, step S2 includes at least one of the following: detecting the rise and fall of the human chest using millimeter-wave radar; acquiring pyroelectric characteristic data representing human activity using an infrared sensor array; acquiring door magnetic status data using a door magnetic sensor component; detecting the heartbeat and turning frequency of the person on the bed using a vibration sensor under the bed; and detecting the gripping and gripping force using a grip force sensor integrated on the bathroom wall and the auxiliary handrail next to the toilet.

[0020] Non-contact life sensing signals encompass five types of sensors: millimeter-wave radar, infrared sensor arrays, door magnetic sensing components, vibration sensors, and grip force sensors. In actual deployment, different sensor combinations can be selected based on the apartment type, cost, and specific needs of the elderly. The complementary coverage of multiple sensor types provides physical redundancy for subsequent multimodal logic mutual verification, reducing the risk of missed detections caused by the failure of a single sensor.

[0021] In the above-mentioned method for monitoring isolated dwellings based on multi-source data fusion and non-contact sensing, step S3 specifically includes: the controller dynamically polls the real-time transmission rate of all data interfaces and determines the minimum transmission rate as the reference period; using the reference period as a window, the high-frequency sensing data is subjected to time-domain dimensionality reduction and aggregation; the low-frequency discrete data is subjected to interpolation and extrapolation to complete the data, thereby generating the multi-feature correlation matrix; the time-domain dimensionality reduction and aggregation includes calculating at least one statistical feature among the mean, variance, or energy spectrum.

[0022] Step S3 is a spatiotemporal alignment mechanism with the lowest transmission rate as the basic period. This mechanism transforms the physical limit of the sensor with the slowest transmission rate in the system into a globally synchronized heartbeat reference, avoiding excessive consumption of edge gateway computing power and achieving the effect of reducing gateway CPU utilization and alarm response latency.

[0023] In the above-mentioned method for monitoring solitary living based on multi-source data fusion and non-contact sensing, step S4 specifically includes: continuously collecting and recording the feature data of the monitored object in steps S1 and S2 of the previous A days; dividing the day into (1440 / B) decision intervals with B minutes as the minimum step size; using a feature adaptive weighted clustering algorithm and support vector clustering to train the water, electricity, and gas usage and spatial transfer patterns under normal living conditions; and constructing a baseline of behavioral patterns for each time period as a benchmark model.

[0024] The baseline model is adaptively corrected based on seasonal and weather changes: using a weekly or monthly sliding window, it adaptively tracks the patterns of changes in sleep time, bathing frequency, and air conditioning power load caused by seasonal changes and weather changes, and dynamically corrects the baseline boundary; using a minute or hour sliding window, it filters the baseline boundary for noise based on micro-leakage in water pipes and changes in power consumption during periodic self-checks of equipment, and corrects the baseline model in real time.

[0025] The baseline model learns from stored data over a period of time, establishing a baseline model representing the behavioral habits of the monitored subject. Based on this baseline model, lifestyle habits may change due to seasonal weather and other factors, such as reduced bathing frequency in winter. This method adaptively corrects these changes based on the baseline model. Macroscopically, it tracks seasonal changes in behavioral habits using a weekly / monthly sliding window, automatically adapting to regular changes such as increased sleep in winter and increased bathing frequency in summer, avoiding misjudging normal behavior as abnormal. Microscopically, it uses a minute / hourly sliding window to filter noise, eliminating interference from occasional events such as minor leaks in water pipes and refrigerator compressor self-checks, preventing false alarms. This solves the problem of high false alarm rates caused by seasonal changes and changes in bodily functions in traditional fixed-threshold alarms, effectively reducing false alarm rates while maintaining high sensitivity and improving the reliability of long-term deployment.

[0026] In the above-mentioned method for monitoring solitary living based on multi-source data fusion and non-contact sensing, step S5 includes determining the state of being out of the house. Specifically, this includes: when the door magnetic state of the life sensing signal indicates the door opening and closing sequence, and there are no other life sensing signals, if the household utility usage data shows that the real-time total power of the household is within the power range of static appliances in standby mode, then it is determined that the monitored person has left the house, and the alarm mechanism is activated; if the real-time total power of the household is higher than the power range of static appliances in standby mode, then it is determined that there is a blind spot anomaly and the alarm is triggered.

[0027] Step S5 is a specific method for logically verifying the absence status of an elderly person by combining NILM non-invasive load monitoring with spatial sensing. The determination conditions are: door magnet generating an open-close sequence + no vital signs from all spatial sensors + total household power within the static standby noise range. In this state, the alarm mechanism is suspended to avoid false alarms. If the power exceeds the noise range and there is no feedback from the spatial sensors, it is determined to be an abnormal blind spot, triggering the corresponding alarm. This solves the fatal problem of the system misjudging absence of an elderly person due to a fall and stillness in blind spots such as under the bed or in a corner, resulting in missed alarms. In other words, this method of detecting elderly people living alone uses the characteristics of the power grid load as a redundant check for spatial sensing, helping to distinguish between true absence (when all appliances are off) and false stillness (when appliances are still running but the person is not moving), thus improving monitoring safety.

[0028] In the above-mentioned method for monitoring solitary living based on multi-source data fusion and non-contact sensing, step S5 includes the judgment of abnormal bathroom conditions, specifically including: when the pyroelectric characteristic data of the bathroom space indicates no macroscopic displacement, the water use characteristic data indicates that the flow rate is zero or stable for a long time, the millimeter-wave radar echo characteristic data continuously captures the micro-movement of chest breathing, and the grip force sensing component integrated in the bathroom safety handrail indicates no effective grip, the system records the composite state.

[0029] If the millimeter-wave radar data determines whether the behavior matches the characteristics of a fall, and the duration of the composite state exceeds the maximum dwell time threshold of the baseline model for the bathroom scene, an auxiliary verification camera is activated, and an alarm mechanism is immediately triggered if the abnormality is confirmed.

[0030] The multimodal logic mutual verification method for determining falls and abnormal stays in bathroom scenarios targets high-risk and privacy-sensitive scenarios. It utilizes the simultaneous fulfillment of four independent physical dimensions—no water flow, radar breathing, no gripping, and no infrared displacement—to eliminate interference caused by shower steam, normal toilet use, etc. It then calculates the approximate entropy of the millimeter-wave echo phase sequence over a previous period. When the approximate entropy exceeds a preset fall action complexity threshold, it is determined to meet the characteristics of a fall action. When all the above conditions are met, an auxiliary verification camera is introduced to capture images to further ensure accuracy. The auxiliary verification camera is normally turned off or even physically obstructed to meet the requirements of privacy scenarios.

[0031] In the above-mentioned method for monitoring isolated dwellings based on multi-source data fusion and non-contact sensing, the auxiliary verification camera is equipped with an electronically controlled switch and a baffle with a closing tendency. When the multi-feature correlation matrix deviates from the benchmark model for a certain period of time, the baffle is controlled to release the obstruction of the lens. At the same time, the auxiliary verification camera is powered on, captures thermal imaging or real-scene data and sends it to the controller. After the transmission is completed, the auxiliary verification camera is powered off and the baffle blocks the lens. The controller further confirms whether the anomaly judgment is accurate. If the judgment is accurate, an alarm mechanism is triggered; otherwise, step S5 is re-executed.

[0032] The auxiliary verification camera features an electrically controlled baffle in front of its lens. This baffle, driven by a spring or gravity, tends to close. Under normal conditions, the camera is in a dual-closed state—power off and the baffle physically obstructs the lens—ensuring a physical blockage and eliminating any possibility of image leakage, thus alleviating lens-related anxiety. However, in truly life-or-death extreme situations, it can be quickly activated, preventing false alarms and wasting rescue resources. The activation steps for the auxiliary verification camera are: camera power supply - baffle opening - thermal imaging / real-time scene capture - data upload - immediate power off - baffle automatically resetting.

[0033] Another objective of this invention is to address the aforementioned problems in the prior art by proposing a solitary monitoring system based on multi-source data fusion and non-contact sensing.

[0034] As an optimization, the alarm mechanism can be directly linked to medical units, or it can be linked to guardians such as children through apps, text messages, etc.

[0035] To achieve this objective of the present invention, the following technical solutions can be used:

[0036] A system for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing, applicable to the aforementioned method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing, includes:

[0037] A data collection component for practical household facilities, used to collect real-time data on water, electricity, and gas usage.

[0038] Non-contact data acquisition components installed on walls or ceilings are used to collect vital signs data and spatial movement data of the monitored units in real time.

[0039] The auxiliary verification cameras installed in each room are used to obtain thermal imaging or real-time image data when an anomaly is initially identified, and are turned off under normal conditions.

[0040] An alarm component is used to send early warning information to guardians or medical units when an anomaly is confirmed;

[0041] The controller that connects the above components;

[0042] The controller is configured to: receive data from each component and construct a multi-feature correlation matrix; construct a benchmark model representing the behavioral habits of the monitored unit based on feature data from the past N days; compare the multi-feature correlation matrix with the benchmark model; preliminarily determine an anomaly when the multi-feature correlation matrix deviates from the benchmark model for a certain period of time; activate an auxiliary verification camera to capture image data; perform verification based on the image data; and trigger an alarm mechanism through an alarm component when it is determined that an anomaly has indeed occurred.

[0043] Corresponding to the aforementioned method for monitoring solitary living, the solitary living monitoring system consists of five main parts: a component for collecting data on the usage of living facilities, a non-contact data acquisition component, an auxiliary verification camera, an alarm component, and a controller. The controller is configured to execute the core logic of the solitary living monitoring method, namely, data reception, multi-feature correlation matrix construction, baseline modeling, deviation judgment, camera-assisted verification, and alarm triggering. This achieves a complete closed loop from physical perception to edge computing and then to hierarchical alarms. In addition, compared with single sensor products, this system effectively reduces the false alarm and missed alarm rates by using a multimodal perception component in conjunction with the controller's built-in multimodal logic mutual verification algorithm.

[0044] In the aforementioned monitoring system for single-occupancy living based on multi-source data fusion and non-contact sensing, the components for collecting data on the usage of practical living facilities include a smart water meter installed on the main water inlet pipe, a smart AMI architecture power meter, and a smart gas meter on the natural gas pipeline; the components for collecting non-contact data include frequency-modulated continuous wave millimeter-wave radar installed on the ceiling of the bedroom and bathroom, an infrared array sensor consisting of 4×4 pyroelectric detection units deployed on the room walls, a door magnetic sensor installed on the entrance door, a vibration sensor installed under the bed, and safety handrails with integrated grip force sensors installed on the shower room wall and next to the toilet; the auxiliary verification camera has an electromagnetically controlled baffle at the lens, which normally blocks the lens through elastic components or gravity, and is opened by the controller when verification is required.

[0045] The data collection components for practical household facilities are implemented through the built-in acquisition modules of smart water meters, smart AMI-based main electricity meters, and smart gas meters on natural gas pipelines. The preferred hardware for the non-contact data collection components includes 24GHz or 60GHz millimeter-wave radar, 4×4 or 8×8 pyroelectric infrared arrays, door magnetic sensors on the entrance door, vibration sensors under the bed, safety handrails with integrated grip force sensors, and auxiliary verification cameras with electromagnetically controlled baffles, which are kept closed by elastic components or gravity. This hardware selection achieves comprehensive, blind-spot-free coverage of the home environment. The millimeter-wave radar captures still breathing, the infrared array senses movement, the vibration sensor extracts nighttime heartbeats and body movements, the grip force sensor determines whether the user is using force to dislodge objects in the bathroom, and the door magnetic sensor records entry and exit times. Moreover, these hardware components work non-contactly with the monitored objects, enabling the construction of a highly redundant sensing network without relying on wearable devices or regular cameras.

[0046] Compared with the prior art, the present invention has the following main advantages:

[0047] 1. This method for monitoring individuals living alone employs non-imaging millimeter-wave radar, infrared arrays, and water, electricity, and gas data during the routine monitoring phase, without collecting any image information; auxiliary verification cameras are only invoked during the extremely low-probability anomaly verification phase, thus blocking privacy leaks at the physical source and resolving the deployment contradictions of video surveillance solutions in private areas such as bedrooms and bathrooms.

[0048] 2. This independent monitoring method abandons the threshold judgment of a single sensor and constructs a multi-dimensional logical mutual verification architecture of water, electricity, gas + pyroelectric displacement + chest respiration + door magnetic state + grip force load. It also integrates approximate entropy to judge the fall action, reducing the false alarm rate and improving the care efficiency.

[0049] 3. This monitoring system for people living alone requires no bracelets, pendants or patches from the elderly, and does not require them to actively charge or operate it, achieving zero intervention in their daily lives and reducing the impact on their comfort.

[0050] 4. This standalone monitoring method uses the lowest transmission rate as a benchmark to uniformly match various detection components with different rates, thereby reducing computing power costs and improving response speed.

[0051] 5. This method for monitoring people living alone adopts a dual-scale sliding window dynamic correction benchmark model, which automatically adapts to seasonal changes in the elderly’s sleep patterns, bathing frequency, and energy usage habits. At the same time, it eliminates occasional interferences such as minor leaks in water pipes and equipment self-checks, thus solving the problem of increased false alarm rates caused by changes in physical function or environment in fixed threshold alarms.

[0052] 6. This method of monitoring individuals living alone forms a redundant interlock between the non-invasive load monitoring system (NILM) and the spatial sensing data. When the spatial sensor does not detect vital signs, if the NILM still detects the operation of high-power electrical appliances, the system will determine that the blind zone is abnormal, thus solving the problem of misjudging that someone is away from home due to unconsciousness or stillness, and improving the reliability of monitoring. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the solitary monitoring method provided by the present invention;

[0054] Figure 2 This is a flowchart illustrating step S3 of the solitary monitoring method provided by the present invention;

[0055] Figure 3 This is a flowchart illustrating step S4 of the solitary monitoring method provided by the present invention;

[0056] Figure 4 This is a schematic diagram of the composition of the solitary monitoring system provided by the present invention. Detailed Implementation

[0057] The following are specific embodiments of the present invention, which are described in conjunction with the accompanying drawings. However, the present invention is not limited to these embodiments.

[0058] Specific implementation examples Figures 1-3 As shown, this method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing includes the following steps:

[0059] S1. Collect real-time data on the usage of practical living facilities to obtain information on the use of water, electricity, and gas;

[0060] S2. Real-time acquisition of non-contact life sensing signals, obtaining pyroelectric characteristic data, millimeter-wave radar data, and door magnetic status data that characterize body micro-movement and spatial movement;

[0061] S3. The controller dynamically extracts the data collected in steps S1 and S2 using different original sampling frequencies, and establishes a multi-feature correlation matrix aligned on a unified time axis with the lowest transmission rate as the basic period.

[0062] S4. Based on the rolling data of the multi-feature correlation matrix stored in history, construct a baseline model representing the daily habits of the monitored object;

[0063] S5. Perform deviation calculation on the multi-feature correlation matrix of step S3 and the benchmark model of step S4. When the deviation from the benchmark model persists for a certain period of time, activate the auxiliary verification camera to acquire thermal imaging or real-time image data to assist in verification. When an anomaly is confirmed, trigger the alarm mechanism.

[0064] Specifically, this method for monitoring the safety of elderly people living alone is designed for home security. Steps S1 and S2 collect data from two completely different physical sources: one from household utilities and the other from spatial non-contact sensors. However, the original sampling frequencies of these two types of data differ significantly. Step S3 balances these two sampling frequencies by dynamically extracting data through a controller and using a matching mechanism with the lowest transmission rate as the basic cycle. This achieves spatiotemporal alignment of heterogeneous data under low-computing-power conditions at the edge, constructing a multi-feature correlation matrix on a unified time axis, reducing computational costs, and improving response speed. Step S4 uses rolling historical data to construct a personalized baseline model that learns each elderly person's unique lifestyle patterns and uses this as a judgment benchmark. Step S5 calculates the deviation between the real-time multi-feature correlation matrix and the baseline model. If the deviation duration exceeds a threshold, an anomaly is further confirmed by an auxiliary verification camera. This auxiliary verification camera is normally disconnected through power outages or physical obstruction to prevent privacy issues. Overall, this method of monitoring individuals living alone is largely imperceptible to their daily lives. Furthermore, the use of an auxiliary verification camera reduces the false alarm rate. This auxiliary verification camera is only activated during the verification phase, ensuring the privacy of the monitored individuals and eliminating any resistance to cameras.

[0065] In this embodiment, step S1 specifically includes: acquiring instantaneous current and voltage waveforms through the smart AMI architecture power meter and classifying home appliances through NILM monitoring technology; acquiring water usage status through the smart water meter on the main water inlet pipe, which includes water usage rate and duration; and acquiring gas usage status through the smart gas meter on the natural gas pipeline, which includes gas usage rate and duration.

[0066] Specifically, step S1 of this method for monitoring the use of water, electricity, and gas is mainly used to detect the usage of these three types of appliances. The AMI architecture power meter can acquire high-frequency sampling instantaneous current and voltage waveforms, the smart water meter can detect the water flow rate and duration, and the smart gas meter can detect the gas consumption rate and duration. Based on this data, the appliances are classified and identified using NILM non-invasive load monitoring technology. This technology is common knowledge and allows for the acquisition of indirect evidence of the elderly's daily activities without the need to install additional sensors indoors, such as determining behaviors like boiling water, cooking, bathing, and heating.

[0067] In this embodiment, step S2 includes: detecting the rise and fall of the human chest using millimeter-wave radar, acquiring pyroelectric characteristic data representing human activity using an infrared sensor array, acquiring door magnetic status data using a door magnetic sensor component, detecting the heartbeat and turning frequency of the person on the bed using a vibration sensor under the bed, and detecting the gripping and gripping force using grip force sensors integrated on the bathroom wall and the auxiliary handrail next to the toilet.

[0068] Specifically, the non-contact life sensing signal covers five types of sensors: millimeter-wave radar, infrared sensor array, door magnetic sensing component, vibration sensor, and grip force sensor. In actual deployment, different sensor combinations can be selected according to the house type, cost, and specific needs of the elderly. The complementary coverage of multiple types of sensors provides physical redundancy for subsequent multimodal logic mutual verification and reduces the risk of missed detection caused by the failure of a single sensor.

[0069] In this embodiment, step S3 specifically includes: the controller dynamically polls the real-time transmission rate of all data interfaces and determines the minimum transmission rate as the reference period; using the reference period as a window, it performs time-domain dimensionality reduction and aggregation on high-frequency sensor data, performs interpolation and extrapolation to complete low-frequency discrete data, and generates the multi-feature correlation matrix. The time-domain dimensionality reduction and aggregation includes calculating various statistical features such as mean, variance, or energy spectrum.

[0070] Specifically, step S3 is a spatiotemporal alignment mechanism with the lowest transmission rate as the basic period. This mechanism transforms the physical limit of the sensor with the slowest transmission rate in the system into a globally synchronized heartbeat reference, avoiding excessive consumption of the edge gateway's computing power and achieving the effect of reducing the gateway's CPU utilization and alarm response latency.

[0071] In this embodiment, step S4 specifically includes: continuously collecting and recording the feature data of the monitored object in steps S1 and S2 for the first 30 days; dividing the day into 96 decision intervals with a minimum step size of 15 minutes; using a feature adaptive weighted clustering algorithm and support vector clustering to train the water, electricity, and gas usage and spatial transfer patterns under normal living conditions; and constructing a baseline of behavioral patterns for each time period as a benchmark model.

[0072] The baseline model is adaptively corrected based on seasonal and weather changes: using a weekly or monthly sliding window, it adaptively tracks the patterns of changes in sleep time, bathing frequency, and air conditioning power load caused by seasonal changes and weather changes, and dynamically corrects the baseline boundary; using a minute or hour sliding window, it filters the baseline boundary for noise based on micro-leakage in water pipes and changes in power consumption during periodic self-checks of equipment, and corrects the baseline model in real time.

[0073] Specifically, the baseline model learns from stored data over a period of time and establishes a baseline model representing the behavioral habits of the monitored object. Based on this baseline model, lifestyle habits may change due to seasonal weather and other factors, such as a decrease in bathing frequency in winter. This method can adaptively correct these changes based on the baseline model. On a macro level, it tracks seasonal changes in behavioral habits using a weekly / monthly sliding window, automatically adapting to regular changes such as increased sleep in winter and increased bathing frequency in summer, avoiding misjudging normal behavior as abnormal. On a micro level, it uses a minute / hourly sliding window to filter out noise, eliminating interference from occasional events such as minor leaks in water pipes and refrigerator compressor self-checks, preventing false alarms. This solves the problem of high false alarm rates caused by seasonal changes and changes in physical function in traditional fixed threshold alarms, effectively reducing false alarm rates while maintaining high sensitivity and improving the reliability of long-term deployment.

[0074] As an optimization of this embodiment, step S5 includes determining the state of being out of home, specifically including: when the door magnetic state of the life sensing signal indicates the door opening and closing sequence, and there are no other life sensing signals, if the household utility usage data shows that the real-time total power of the household is within the power range of static appliances in standby mode, then it is determined that the monitored object has left home, and the alarm mechanism is activated; if the real-time total power of the household is higher than the power range of static appliances in standby mode, then it is determined that there is a blind spot abnormality and an alarm is triggered.

[0075] Specifically, NILM non-intrusive load monitoring and spatial sensing are logically cross-validated to determine the absence status. When the door sensor generates an open-close sequence, all spatial sensors show no signs of life, and the total household power is within the static standby noise range, the alarm mechanism is suspended to avoid false alarms. If the power exceeds the noise range and there is no feedback from the spatial sensors, it is considered an abnormal blind spot, triggering the corresponding alarm. This solves the critical problem of the system misinterpreting absence of alarms when an elderly person falls and remains still in a blind spot such as under a bed or in a corner, leading to a false alarm. In other words, this method of detecting homelessness uses the characteristics of the power grid load as a redundant check for spatial sensing, helping to distinguish between a true absence when all appliances are off and a false absence when appliances are still running but the person is not moving, thus improving monitoring safety.

[0076] As an optimization of this embodiment, step S5 further includes the judgment of abnormal bathroom conditions, specifically including: when the pyroelectric characteristic data of the bathroom space indicates no macroscopic displacement, the water usage characteristic data indicates that the flow rate is zero or stable for a long time, the millimeter-wave radar echo characteristic data continuously captures micro-movements of chest breathing, and the grip force sensing component integrated in the bathroom safety handrail indicates no effective grip, the system records this composite state; the millimeter-wave radar data is used to determine whether it conforms to the characteristics of a fall, and when the duration of the composite state exceeds the maximum dwell time threshold of the baseline model for the bathroom scene, the auxiliary verification camera is activated, and an alarm mechanism is immediately triggered when it is determined that there is indeed an abnormality.

[0077] Specifically, the multimodal logic mutual verification method for determining falls and abnormal stays in bathroom scenarios targets high-risk and privacy-sensitive scenarios. It utilizes the simultaneous fulfillment of four independent physical dimensions—no water flow, radar breathing, no gripping, and no infrared displacement—to eliminate interference caused by shower steam, normal toilet use, etc. It then calculates the approximate entropy of the millimeter-wave echo phase sequence over a previous period. When the approximate entropy exceeds a preset fall action complexity threshold, it is determined to meet the characteristics of a fall action. When all the above conditions are met, an auxiliary verification camera is introduced to capture images to further ensure accuracy. The auxiliary verification camera is normally turned off or even physically obstructed to meet the requirements of privacy scenarios.

[0078] As an optimization of this embodiment, the auxiliary verification camera is equipped with an electronically controlled switch and a baffle with a closing tendency. When the multi-feature correlation matrix deviates from the baseline model for a certain period of time, the baffle is controlled to release the obstruction of the lens. At the same time, the auxiliary verification camera is powered on, captures thermal imaging or real-scene data and sends it to the controller. After the transmission is completed, the auxiliary verification camera is powered off and the baffle blocks the lens. The controller further confirms whether the anomaly judgment is accurate. If the judgment is accurate, an alarm mechanism is triggered; otherwise, step S5 is re-executed.

[0079] Specifically, the auxiliary verification camera features an electrically controlled baffle in front of its lens. This baffle, driven by a spring or gravity, tends to close. Under normal conditions, the camera is in a dual-closed state—power off and the baffle physically obstructs the lens—ensuring physical blockage from a hardware perspective and eliminating any possibility of image leakage, thus alleviating lens anxiety. However, in truly life-or-death extreme situations, it can be quickly activated, avoiding the waste of rescue resources caused by false alarms. The activation steps for the auxiliary verification camera are: camera power supply - drive the baffle to open - capture thermal imaging / real-time scene - data upload - immediate power off - baffle automatically reset.

[0080] like Figure 4 As shown, the solitary dwelling monitoring system based on multi-source data fusion and non-contact sensing of the present invention is applicable to the above-mentioned solitary dwelling monitoring method based on multi-source data fusion and non-contact sensing, and includes:

[0081] The household utility usage data collection components are installed on the main water pipe, the smart AMI architecture power meter, and the natural gas pipeline to collect real-time data on water, electricity, and gas usage.

[0082] Non-contact data acquisition components installed on walls or ceilings are used to collect vital signs data and spatial movement data of the monitored units in real time.

[0083] The auxiliary verification cameras installed in each room are used to obtain thermal imaging or real-time image data when an anomaly is initially identified, and are turned off under normal conditions.

[0084] An alarm component is used to send early warning information to guardians or medical units when an anomaly is confirmed;

[0085] The controller that connects the above components;

[0086] The controller is configured to: receive data from each component and construct a multi-feature correlation matrix; construct a benchmark model representing the behavioral habits of the monitored unit based on feature data from the past N days; compare the multi-feature correlation matrix with the benchmark model; preliminarily determine an anomaly when the multi-feature correlation matrix deviates from the benchmark model for a certain period of time; activate an auxiliary verification camera to capture image data; perform verification based on the image data; and trigger an alarm mechanism through an alarm component when it is determined that an anomaly has indeed occurred.

[0087] Specifically, corresponding to the aforementioned method for monitoring solitary living, the solitary living monitoring system consists of five main parts: a component for collecting data on the usage of living facilities, a non-contact data acquisition component, an auxiliary verification camera, an alarm component, and a controller. The controller is configured to execute the core logic of the solitary living monitoring method, namely, data reception, multi-feature correlation matrix construction, baseline modeling, deviation judgment, camera-assisted verification, and alarm triggering. This achieves a complete closed loop from physical perception to edge computing and then to hierarchical alarms. In addition, compared with single sensor products, this system effectively reduces the false alarm and missed alarm rates by using a multimodal perception component in conjunction with the controller's built-in multimodal logic mutual verification algorithm.

[0088] In this embodiment, the non-contact data acquisition components include frequency-modulated continuous wave millimeter-wave radar installed on the ceiling of the bedroom and bathroom, an infrared array sensor consisting of 4×4 pyroelectric detection units deployed on the room wall, a door magnetic sensor installed on the entrance door, a vibration sensor installed under the bed, and safety handrails with integrated grip force sensors installed on the shower room wall and next to the toilet; the auxiliary verification camera has an electromagnetically controlled baffle at the lens, which normally blocks the lens through an elastic component or gravity, and is controlled by the controller to open when verification is required.

[0089] Specifically, the preferred hardware selection for the non-contact data acquisition components includes 24GHz or 60GHz millimeter-wave radar, an infrared array with 4×4 or 8×8 pyroelectric units, a door magnetic sensor installed on the entrance door, a vibration sensor under the bed, a safety handrail with an integrated grip force sensor, and an auxiliary verification camera with an electromagnetically controlled baffle, which is kept closed by elastic components or gravity. This hardware selection achieves comprehensive, blind-spot-free coverage of the home environment. The millimeter-wave radar captures still breathing, the infrared array senses movement, the vibration sensor extracts nighttime heartbeats and body movements, the grip force sensor determines whether the user is using force to dislodge objects in the bathroom, and the door magnetic sensor records entry and exit times. Moreover, these hardware components work non-contactly with the monitored objects, enabling the construction of a highly redundant sensing network without relying on wearable devices or regular cameras.

[0090] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

Claims

1. A method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing, characterized in that, Includes the following steps: S1. Collect real-time data on the usage of practical living facilities, and obtain the usage status of at least one of water, electricity, and gas. S2. Real-time acquisition of non-contact life sensing signals, including at least one of the following: pyroelectric characteristic data representing body micro-movement and spatial movement, millimeter-wave radar data, and door magnetic status data; S3. The controller dynamically extracts the data collected in steps S1 and S2 using different original sampling frequencies, and establishes a multi-feature correlation matrix aligned on a unified time axis with the lowest transmission rate as the basic period. S4. Based on the rolling data of the multi-feature correlation matrix stored in history, construct a baseline model representing the daily habits of the monitored object; S5. Perform deviation calculation between the multi-feature correlation matrix of step S3 and the benchmark model of step S4. If the deviation from the benchmark model persists for a certain period of time, directly or enable the auxiliary verification camera to acquire thermal imaging or real-time image data to assist in verification. If an anomaly is confirmed, trigger the alarm mechanism.

2. The method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing according to claim 1, characterized in that, Step S1 specifically includes: Instantaneous current and voltage waveforms are obtained through the smart AMI architecture power meter, and home appliances are classified through NILM monitoring technology; The water usage status is obtained through a smart water meter on the main water pipe in the house. This water usage status includes the water usage rate and duration. Gas usage status is obtained through smart gas meters on natural gas pipelines, including the gas usage rate and duration.

3. The method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing according to claim 1, characterized in that, Step S2 includes at least one of the following: detecting the rise and fall of the human chest using millimeter-wave radar; acquiring pyroelectric characteristic data representing human activity using an infrared sensor array; acquiring door magnetic status data using a door magnetic sensor component; detecting the heartbeat and turning frequency of the person on the bed using a vibration sensor under the bed; and detecting the gripping and gripping force using grip force sensors integrated on the bathroom wall and the auxiliary handrail next to the toilet.

4. The method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing according to claim 1, characterized in that, Step S3 specifically includes: the controller dynamically polls the real-time transmission rate of all data interfaces and determines the minimum transmission rate as the reference period. Using the reference period as a window, the controller performs time-domain dimensionality reduction and aggregation on the high-frequency sensor data and performs interpolation and extrapolation to complete the low-frequency discrete data, generating the multi-feature correlation matrix. The time-domain dimensionality reduction and aggregation includes calculating at least one statistical feature among the mean, variance, or energy spectrum.

5. The method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing according to claim 1, characterized in that, The specific steps S4 include: continuously collecting and recording the feature data of the monitored object in steps S1 and S2 of the previous A days; dividing the day into (1440 / B) decision intervals with B minutes as the minimum step size; using a feature adaptive weighted clustering algorithm and support vector clustering to train the water, electricity and gas usage and spatial transfer patterns under normal living conditions; constructing a baseline of behavioral patterns for each time period as a benchmark model. The baseline model is adaptively corrected based on seasonal and weather changes: using a weekly or monthly sliding window, it adaptively tracks the patterns of changes in sleep time, bathing frequency, and air conditioning power load caused by seasonal changes and weather changes, and dynamically corrects the baseline boundary; using a minute or hour sliding window, it filters the baseline boundary for noise based on micro-leakage in water pipes and changes in power consumption during periodic self-checks of equipment, and corrects the baseline model in real time.

6. The method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing according to claim 1, characterized in that, Step S5 includes determining the state of being out of the house, specifically: when the door magnetic sensor status indicates the door opening and closing sequence and there are no other life sensing signals, if the household utility usage data shows that the real-time total power of the household is within the power range of static appliances in standby mode, then it is determined that the monitored person has left the house and the alarm mechanism is activated; if the real-time total power of the household is higher than the power range of static appliances in standby mode, then it is determined that there is a blind spot abnormality and the alarm is triggered.

7. The method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing according to claim 1, characterized in that, Step S5 includes the judgment of abnormal bathroom conditions, specifically including: when the pyroelectric characteristic data of the bathroom space indicates no macroscopic displacement, the water use characteristic data indicates that the flow rate is zero or stable for a long time, the millimeter-wave radar echo characteristic data continuously captures the micro-movement of chest breathing, and the grip force sensing component integrated in the bathroom safety handrail indicates no effective grip, the system records the composite state. If the millimeter-wave radar data determines whether the behavior matches the characteristics of a fall, and the duration of the composite state exceeds the maximum dwell time threshold of the baseline model for the bathroom scene, an auxiliary verification camera is activated, and an alarm mechanism is immediately triggered if the abnormality is confirmed.

8. The method for monitoring solitary dwellings based on multi-source data fusion and non-contact sensing according to claim 1, characterized in that, The auxiliary verification camera is equipped with an electrically controlled switch and a baffle with a closing tendency. When the multi-feature correlation matrix deviates from the baseline model for a certain period of time, the baffle is controlled to release the obstruction of the lens. At the same time, the auxiliary verification camera is powered on, captures thermal imaging or real-scene data and sends it to the controller. After the data is sent, the auxiliary verification camera is powered off and the baffle blocks the lens. The controller further confirms whether the anomaly judgment is accurate. If the judgment is accurate, an alarm mechanism is triggered; otherwise, step S5 is re-executed.

9. A solitary monitoring system based on multi-source data fusion and non-contact sensing, applicable to the solitary monitoring method based on multi-source data fusion and non-contact sensing as described in any one of claims 1-8, characterized in that, include: A data collection component for practical household facilities, used to collect real-time data on water, electricity, and gas usage; Non-contact data acquisition components installed on walls or ceilings are used to collect vital signs data and spatial movement data of the monitored units in real time. The auxiliary verification cameras installed in each room are used to obtain thermal imaging or real-time image data when an anomaly is initially identified, and are turned off under normal conditions. An alarm component is used to send early warning information to guardians or medical units when an anomaly is confirmed; The controller that connects the above components; The controller is configured to: receive data from each component and construct a multi-feature correlation matrix; construct a benchmark model representing the behavioral habits of the monitored unit based on feature data from the past N days; compare the multi-feature correlation matrix with the benchmark model; preliminarily determine an anomaly when the multi-feature correlation matrix deviates from the benchmark model for a certain period of time; activate an auxiliary verification camera to capture image data; perform verification based on the image data; and trigger an alarm mechanism through an alarm component when it is determined that an anomaly has indeed occurred.

10. The solitary monitoring system based on multi-source data fusion and non-contact sensing according to claim 9, characterized in that, The aforementioned household utility usage data collection components include a smart water meter installed on the main water inlet pipe, a smart AMI architecture main power meter, and a smart gas meter on the natural gas pipeline. The non-contact data acquisition components include frequency-modulated continuous wave millimeter-wave radar installed on the top of the bedroom and bathroom, an infrared array sensor consisting of 4×4 pyroelectric detection units deployed on the wall of the room, a door magnetic sensor installed on the entrance door, a vibration sensor installed under the bed, and a safety handrail with integrated grip force sensor installed on the shower room wall and next to the toilet. The auxiliary verification camera is equipped with an electromagnetically controlled baffle at the lens. Under normal conditions, the baffle is kept in place by an elastic component or gravity to block the lens, and is opened by the controller when verification is required.

Images