An old person home safety monitoring and early warning system and method based on internet of things

Abstract

The application provides an old people home safety monitoring and early warning system and method based on Internet of Things. It comprises a sensor node for collecting home environment data, an Internet of Things communication module for transmitting data, a central processing module for processing data and an early warning module for issuing warning information. The sensor node comprises a voiceprint collection sub-node and an electromagnetic wave perception sub-node. The central processing module is integrated with a voiceprint modal decomposition unit, an electromagnetic wave distortion analysis unit and a multi-modal decision unit. The voiceprint modal decomposition unit decomposes the environmental voiceprint signals obtained by the voiceprint collection sub-node by intrinsic modal function. The electromagnetic wave distortion analysis unit establishes a second-order correlation model between human displacement and electromagnetic field distortion based on the electromagnetic field signals collected by the electromagnetic wave perception sub-node. The application can solve the false alarm problem and effectively identify the stationary dangerous state after falling.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of home safety monitoring, and in particular to an old person home safety monitoring and early warning system and method based on Internet of Things. BACKGROUND

[0002] In the field of old person home safety monitoring, fall detection is a key technology to ensure the safety of the elderly living alone. Existing fall detection schemes mostly rely on a single sensing technology, such as a pressure-sensitive mat, a visual camera or a single modality sensor. The pressure-sensitive mat can determine a fall by monitoring pressure changes, but it is easily triggered by non-fall events such as pet stepping, objects falling, etc., and the false alarm rate is high. Frequent invalid alarms not only interfere with the normal life of the elderly, but also may lead to a delayed response of the guardian to the real alarm. Although the visual camera can capture motion features, its reliability is greatly reduced in night or complex home environment; more importantly, when the elderly falls into a coma after falling, the body is still, and neither the pressure mat nor the camera can continuously identify the dangerous state, making it difficult to issue an effective warning and delaying the rescue opportunity.

[0003] The above problems result in that the existing technology cannot balance monitoring accuracy and full coverage of dangerous states in a complex home environment, neither effectively filtering interference signals nor responding to the high-risk scenario of falling into a coma after falling, and there is an urgent need for a monitoring technology that can solve the false alarm problem and effectively identify the stationary dangerous state after falling. SUMMARY

[0004] The application aims to solve the shortcomings in the prior art and provides an old person home safety monitoring and early warning system and method based on Internet of Things, which comprises a sensor node for collecting home environment data, an Internet of Things communication module for transmitting data, a central processing module for processing data, and an early warning module for issuing warning information:

[0005] The sensor node comprises a voiceprint acquisition sub-node and an electromagnetic wave perception sub-node, and the central processing module is integrated with a voiceprint modal decomposition unit, an electromagnetic wave distortion analysis unit and a multi-modal decision unit.

[0006] The voiceprint modal decomposition unit performs intrinsic modal function decomposition on the environmental voiceprint signals obtained by the voiceprint acquisition sub-node, the electromagnetic wave distortion analysis unit establishes a second-order correlation model of human body displacement and electromagnetic field distortion based on the electromagnetic field signals collected by the electromagnetic wave perception sub-node, and the multi-modal decision unit fuses the voiceprint entropy value obtained by voiceprint modal decomposition and the distortion rate obtained by electromagnetic wave distortion analysis to calculate the fall probability.

[0007] When the fall probability exceeds a preset threshold and the state lasts for a preset length of time, the multi-modal decision unit sends a trigger signal to a pre-warning module, and the pre-warning module issues a pre-warning information based on the trigger signal.

[0008] Preferably, the voiceprint collection sub-node comprises: an omnidirectional microphone array, a signal conditioning circuit and an analog-to-digital converter, the output end of the omnidirectional microphone array is connected with the input end of the signal conditioning circuit, the output end of the signal conditioning circuit is connected with the input end of the analog-to-digital converter, and the output end of the analog-to-digital converter is connected with the input end of the central processing module through the Internet of Things communication module; the omnidirectional microphone array collects environmental voiceprint signals, the signal conditioning circuit performs gain adjustment and filtering processing on the collected voiceprint signals, and the analog-to-digital converter converts the conditioned analog voiceprint signals into digital signals.

[0009] Further preferably, the electromagnetic wave perception sub-node comprises a 5.8GHz microwave sensor, a radio frequency front-end circuit and a data buffer, the output end of the 5.8GHz microwave sensor is connected with the input end of the radio frequency front-end circuit, the output end of the radio frequency front-end circuit is connected with the input end of the data buffer, and the output end of the data buffer is connected with the input end of the central processing module through the Internet of Things communication module; the 5.8GHz microwave sensor is respectively deployed on the central ceiling of the living room, above the bedside of the bedroom, the top of the bathroom and above the entrance of the kitchen, the radio frequency front-end circuit performs mixing, filtering and amplification processing on the radio frequency signals output by the sensor, and the data buffer temporarily stores the processed electromagnetic field signals.

[0010] Further preferably, the multi-modal decision unit comprises a feature normalization sub-unit, a weight dynamic adjustment sub-unit and a time window verification sub-unit, the input end of the feature normalization sub-unit is connected with the output end of the voiceprint modal decomposition unit and the output end of the electromagnetic wave distortion analysis unit respectively, the output end of the feature normalization sub-unit is connected with the input end of the weight dynamic adjustment sub-unit, and the output end of the weight dynamic adjustment sub-unit is connected with the input end of the time window verification sub-unit; the feature normalization sub-unit respectively maps the voiceprint entropy value and the electromagnetic wave distortion rate to a preset numerical interval, the weight dynamic adjustment sub-unit adjusts the fusion weight of the voiceprint entropy value and the electromagnetic wave distortion rate according to the environmental noise intensity, increases the weight proportion of the electromagnetic wave distortion rate when the environmental noise is high, and increases the weight proportion of the voiceprint entropy value when the environmental noise is low, and the time window verification sub-unit continuously monitors whether the fall probability meets the condition of exceeding the preset threshold and records the duration of the state.

[0011] Further preferably, the decomposition operation performed by the voiceprint modal decomposition unit satisfies:

[0012] ;

[0013] wherein, represents the ambient voiceprint signal, in volts; represents the i-th order intrinsic modal function, used to represent the oscillation components of different frequencies in the voiceprint signal, in volts; represents the residual term, used to represent the trend component in the voiceprint signal, in volts; n represents the decomposition order, the value of which is dynamically determined according to the complexity of the voiceprint signal, the more complex the frequency components of the signal are, the larger the value of n is.

[0014] Further preferably, the human body displacement-electromagnetic field distortion model established by the electromagnetic wave distortion analysis unit satisfies:

[0015] ;

[0016] wherein, represents the electromagnetic field distortion value, in volts per meter; represents the medium coupling coefficient, used to reflect the coupling degree of the human body and the air medium to the electromagnetic wave; represents the three-dimensional displacement vector of the human body, in meters, wherein x, y, and z respectively represent the displacement components in the x-axis, y-axis, and z-axis directions of the space rectangular coordinate system; represents the attenuation coefficient, used to represent the initial attenuation amplitude of the electromagnetic field distortion; represents the time attenuation rate, in per second, used to represent the attenuation speed of the electromagnetic field distortion with time; t represents time, in seconds.

[0017] Further preferably, the formula for calculating the fall probability by the multi-modal decision unit satisfies:

[0018] ;

[0019] wherein, represents the fall probability; represents the weight coefficient of the voiceprint entropy value; represents the weight coefficient of the electromagnetic wave distortion rate; represents the voiceprint entropy value, used to quantify the complexity of the voiceprint signal; represents the threshold value of the voiceprint entropy value; represents the electromagnetic wave distortion rate, used to quantify the distortion degree of the electromagnetic field signal; represents the threshold value of the electromagnetic wave distortion rate.

[0020] An old-age home safety monitoring and early warning method based on the Internet of Things, applied to an old-age home safety monitoring and early warning system based on the Internet of Things, comprises: collecting old-age home environment data through an Internet of Things device and transmitting the data to a processing center; further comprising:

[0021] Collecting an ambient voiceprint signal, decomposing the ambient voiceprint signal by using an improved empirical mode decomposition algorithm to obtain a plurality of intrinsic mode functions and a residual term, and calculating a voiceprint entropy value based on the intrinsic mode functions;

[0022] Deploying a 5.8 GHz microwave sensor to collect electromagnetic field signals of a home environment, establishing a second-order model of human body displacement and electromagnetic field distortion based on the electromagnetic field signals, and calculating an electromagnetic wave distortion rate through the second-order model;

[0023] Inputting the voiceprint entropy value and the electromagnetic wave distortion rate into a multi-modal decision model to calculate a fall probability;

[0024] When the fall probability exceeds a preset threshold and the state lasts for a preset duration, triggering a warning mechanism and issuing a warning information containing a fall location and a duration.

[0025] Further preferably, the step of collecting an ambient voiceprint signal and calculating a voiceprint entropy value comprises:

[0026] Collecting an ambient voiceprint signal by using an omnidirectional microphone array, and performing band-pass filtering on the ambient voiceprint signal to remove low-frequency vibration noise and high-frequency electromagnetic interference;

[0027] Performing a direct current component removal process on the filtered voiceprint signal to obtain an alternating current voiceprint signal;

[0028] Decomposing the alternating current voiceprint signal by using an improved empirical mode decomposition algorithm, wherein each intrinsic mode function in the decomposition process needs to satisfy that the mean value of the upper and lower envelope lines is 0, and the number of zero-crossing points is equal to or differs from the number of extreme points by at most one;

[0029] Calculating a Shannon entropy based on the intrinsic mode functions obtained by decomposition, and taking the Shannon entropy as a voiceprint entropy value, wherein the amplitude of each intrinsic mode function is normalized during the calculation process.

[0030] Further preferably, the step of triggering a warning mechanism comprises:

[0031] When the calculated fall probability exceeds a preset threshold, starting a secondary verification process, wherein the secondary verification process comprises calculating a change rate of the voiceprint entropy value and a change rate of the electromagnetic wave distortion rate within a preset period, and when both the change rate of the voiceprint entropy value and the change rate of the electromagnetic wave distortion rate satisfy a preset condition, determining that the secondary verification is passed;

[0032] After the secondary verification is passed, starting a time window monitoring of a preset duration, and continuously recording the fall probability within the time window;

[0033] If the fall probability within the time window of the preset duration all exceeds the preset threshold, it is determined that the warning mechanism is triggered;

[0034] The early warning information is sent to the preset guardian terminal and community service platform through the Internet of Things communication module, and the early warning information includes the specific position of the fall, the fall duration, and the characteristic parameters of the voiceprint and electromagnetic wave signal, and the specific position is determined based on the signal strength triangulation of each sensor node.

[0035] Technical effects: Through the creative design of cross-modal fusion of voiceprint and electromagnetic wave, combined with the second-order electromagnetic distortion model and multi-modal decision mechanism, the two major problems in the background technology are effectively solved: cross-modal fusion filters interference signals such as pet activity and object falling, greatly reducing false positives; the second-order electromagnetic distortion model can capture the continuous distortion of the electromagnetic field in the stationary state after falling, realizing the recognition of high-risk scenes such as coma; the multi-modal decision unit ensures accurate and effective early warning through probability calculation and continuous verification, and takes into account monitoring accuracy and full coverage of dangerous states. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 The connection block diagram of the home safety monitoring and early warning system for the elderly based on the Internet of Things of the present application;

[0037] Figure 2 The detailed connection diagram of the voiceprint collection sub-node of the present application;

[0038] Figure 3 The detailed connection diagram of the electromagnetic wave sensing sub-node of the present application;

[0039] Figure 4 The flow chart of the home safety monitoring and early warning method for the elderly based on the Internet of Things of the present application;

[0040] Figure 5 The flow chart of collecting environmental voiceprint signals and calculating voiceprint entropy values of the present application;

[0041] Figure 6 The step flow chart of triggering the early warning mechanism of the present application. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical scheme and advantages of the present application clearer and more apparent, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0043] The traditional home safety monitoring system for the elderly has the defect of single monitoring dimension, and relies on single sensors such as pressure pads and cameras, which are easily disturbed by pet activities, falling objects and other factors, leading to false positives, and cannot identify the coma state after falling. The pressure pad can only sense pressure changes, and the camera is invalid when blocked or insufficient light, and neither can continuously monitor the stationary state after falling, resulting in missed safety hazards. At the same time, the existing system lacks a multi-source data fusion mechanism, making it difficult to accurately extract falling features from complex environments, affecting the reliability of monitoring.

[0044] Based on this, please refer to Figure 1 The embodiment provides a kind of based on Internet of Things's home safety monitoring and early warning system for the elderly, including the sensor node for collecting home environment data, the Internet of Things communication module for transmitting data, the central processing module for processing data and the early warning module for sending warning information;Sensor node includes voiceprint acquisition subnode and electromagnetic wave perception subnode, central processing module is integrated with voiceprint modal decomposition unit, electromagnetic wave distortion analysis unit and multi-modal decision unit;

[0045] Voiceprint modal decomposition unit carries out intrinsic mode function decomposition to the environmental voiceprint signal obtained by voiceprint acquisition subnode, electromagnetic wave distortion analysis unit establishes the second-order correlation model of human displacement and electromagnetic field distortion based on the electromagnetic field signal collected by electromagnetic wave perception subnode, and multi-modal decision unit fuses the voiceprint entropy value obtained by voiceprint modal decomposition and the distortion rate obtained by electromagnetic wave distortion analysis to calculate the falling probability;

[0046] When the falling probability exceeds the preset threshold and the state lasts for a preset duration, the multi-modal decision unit sends a trigger signal to the early warning module, and the early warning module sends early warning information based on the trigger signal;

[0047] Through the cross-modal fusion of voiceprint and electromagnetic wave, the false positive problem caused by pet activities, falling objects and other factors in the prior art is solved, and through the second-order electromagnetic distortion model, the coma state after falling is identified, overcoming the failure defect of pressure pad or camera scheme in coma scene.

[0048] The technical scheme constructs a multi-dimensional perception system through voiceprint and electromagnetic wave dual-modal monitoring: voiceprint acquisition captures acoustic characteristics such as impact sound and distress sound when falling, electromagnetic wave perception monitors the dynamic and static changes of human displacement through electromagnetic field distortion, and the two form a complement.

[0049] The three units of the central processing module have clear division of labor: the voiceprint modal decomposition unit disassembles complex voiceprint signals into analyzable intrinsic mode functions, the electromagnetic wave distortion analysis unit captures the displacement changes after human falling through the second-order model, including the stationary state, and the multi-modal decision unit calculates the falling probability through feature fusion, avoiding the limitations of single modal.

[0050] The technical effects are embodied in two aspects: first, the cross-modal fusion filters out interference signals such as pet activities and item drops in a single modal, reducing invalid early warnings; second, the second-order electromagnetic distortion model can continuously monitor the stationary state after falling, solving the problem of recognition failure of the traditional scheme for the coma state, and realizing full-scene and high-reliability monitoring of the safety of the elderly at home.

[0051] Traditional voiceprint collection devices mostly use a single microphone, which has the problems of limited collection range and being easily disturbed by environmental noise, resulting in low signal-to-noise ratio of the voiceprint signal and difficulty in subsequent feature extraction. At the same time, if the analog voiceprint signal is not properly conditioned during the conversion to a digital signal, distortion will be introduced, affecting the integrity of the signal and further reducing the recognition accuracy of the fall voiceprint features, which cannot provide reliable data support for subsequent analysis.

[0052] Based on this, please refer to Figure 2 , the voiceprint collection sub-node includes an omnidirectional microphone array, a signal conditioning circuit, and an analog-to-digital converter, the output end of the omnidirectional microphone array is connected with the input end of the signal conditioning circuit, the output end of the signal conditioning circuit is connected with the input end of the analog-to-digital converter, and the output end of the analog-to-digital converter is connected with the input end of the central processing module through an Internet of Things communication module; the omnidirectional microphone array collects environmental voiceprint signals, the signal conditioning circuit performs gain adjustment and filtering processing on the collected voiceprint signals, and the analog-to-digital converter converts the conditioned analog voiceprint signals into digital signals.

[0053] This technical solution optimizes the voiceprint collection link from the hardware structure: the omnidirectional microphone array expands the collection range through multiple microphone cooperation to ensure coverage of key areas at home, such as the living room and bedroom, avoiding the monitoring blind area of a single microphone; the signal conditioning circuit undertakes the preprocessing function, gain adjustment can amplify weak voiceprint signals such as low-amplitude impact sound when the elderly fall, and filtering processing filters out steady-state noise in the environment such as appliance operation sound, improving signal purity; the analog-to-digital converter converts analog signals into digital signals to provide adaptive digital input for the voiceprint modal decomposition of the central processing module. The complete signal flow is formed through the clear connection relationship of the components, and the optimization of the whole link from collection to conversion ensures the integrity and reliability of the voiceprint signal.

[0054] The technical effects are embodied in two aspects: first, the omnidirectional microphone array expands the monitoring coverage range and reduces the dead space; second, the signal conditioning circuit reduces noise interference, making the effective voiceprint features more prominent; and the analog-to-digital conversion provides a high-quality data basis for subsequent digital signal processing, laying a foundation for the voiceprint modal decomposition unit to accurately extract the fall-related eigenmodal function, and improving the accuracy of voiceprint feature analysis.

[0055] The traditional electromagnetic wave sensor is not reasonably deployed, and is mostly deployed in single point, resulting in blind area in monitoring range, and failing to comprehensively capture the displacement change of human body in each area of home environment; meanwhile, the radio frequency signal processing lacks professional front-end circuit, the original signal is vulnerable to interference and has weak amplitude, and it is difficult to extract effective electromagnetic field distortion features, affecting the accuracy of the human displacement-electromagnetic field correlation model, and failing to reliably support the fall judgment.

[0056] Based on this, refer to Figure 3 , the electromagnetic wave perception sub-node includes a 5.8 GHz microwave sensor, a radio frequency front-end circuit and a data buffer, the output end of the 5.8 GHz microwave sensor is connected with the input end of the radio frequency front-end circuit, the output end of the radio frequency front-end circuit is connected with the input end of the data buffer, and the output end of the data buffer is connected with the input end of the central processing module through an Internet of Things communication module; the 5.8 GHz microwave sensor is respectively deployed on the central ceiling of the living room, above the bedside of the bedroom, the top of the bathroom and above the entrance of the kitchen, the radio frequency front-end circuit performs mixing, filtering and amplification processing on the radio frequency signal output by the sensor, and the data buffer temporarily stores the processed electromagnetic field signal.

[0057] The technical scheme optimizes the electromagnetic wave perception ability from the aspects of deployment and signal processing: the multi-point deployment of the 5.8 GHz microwave sensor covers the core areas of the activities of the elderly, such as the living room, bedroom, bathroom and kitchen, and forms a dead angle-free monitoring network through spatial distribution, ensuring that the displacement of human body in any key area can be captured; the radio frequency front-end circuit processes the microwave signal characteristics professionally, the mixing converts the high-frequency signal into intermediate frequency signal which is easy to process, the filtering removes the stray interference, and the amplification enhances the signal amplitude, so that the electromagnetic field distortion features are more significant; the data buffer temporarily stores the processed data, avoiding signal loss caused by communication delay and ensuring data continuity. The functions and connection relationships of the components are clear, forming a complete link from signal collection, processing to temporary storage.

[0058] The technical effects are embodied in: the multi-point deployment eliminates the monitoring blind area, ensuring electromagnetic wave signal coverage in the whole area of home; the radio frequency front-end circuit improves the signal quality, making the electromagnetic field distortion features more easily extracted; the data buffer guarantees the continuity of the signal, providing complete data for the electromagnetic wave distortion analysis unit to establish an accurate human displacement-electromagnetic field distortion second-order model, and improving the recognition accuracy of human motion state including fall and stillness.

[0059] In the traditional multi-modal decision mechanism, the fusion mode of different modal features is fixed, and the dynamic changes of the environment, such as noise intensity fluctuation, are not considered, resulting in that the feature weight does not match the actual scene, affecting the decision accuracy; meanwhile, the judgment of fall probability lacks time dimension verification, and is easy to trigger false alarm due to instantaneous interference signal, reducing the system reliability, and failing to effectively distinguish real fall from instantaneous interference.

[0060] Based on this, the multi-modal decision unit includes a feature normalization subunit, a weight dynamic adjustment subunit, and a time window verification subunit. The input ends of the feature normalization subunit are connected with the output ends of the voiceprint modal decomposition unit and the electromagnetic wave distortion analysis unit, respectively. The output end of the feature normalization subunit is connected with the input end of the weight dynamic adjustment subunit. The output end of the weight dynamic adjustment subunit is connected with the input end of the time window verification subunit. The feature normalization subunit maps the voiceprint entropy value and the electromagnetic wave distortion rate to a preset numerical interval, respectively. The weight dynamic adjustment subunit adjusts the fusion weight of the voiceprint entropy value and the electromagnetic wave distortion rate according to the environmental noise intensity. When the environmental noise is high, the weight proportion of the electromagnetic wave distortion rate is increased. When the environmental noise is low, the weight proportion of the voiceprint entropy value is increased. The time window verification subunit continuously monitors whether the fall probability meets the condition of exceeding the preset threshold and records the duration of the state.

[0061] The technical solution constructs an adaptive decision mechanism through three subunits. The feature normalization subunit maps the voiceprint entropy value and the electromagnetic wave distortion rate to a unified numerical interval, eliminating the dimensional difference of different modal features and providing a basis for fusion calculation. The weight dynamic adjustment subunit adjusts the weight in real time according to the environmental noise. When the noise is high, the reliability of the voiceprint signal decreases, and the electromagnetic wave modal is relied on. When the noise is low, the voiceprint modal is emphasized, so that the decision adapts to environmental changes. The time window verification subunit verifies the fall probability from the time dimension to avoid false alarms caused by instantaneous signals and ensure the stability of the judgment. The subunits form a synergy through clear signal flow to achieve precise control from feature preprocessing to final decision.

[0062] The technical effect is that feature normalization ensures the fusibility of multi-modal features, dynamic weight adjustment improves the adaptability of the system to complex environments, and time window verification reduces false alarms caused by instantaneous interference, enhances the reliability of fall judgment, and ultimately improves the accuracy and stability of the multi-modal decision unit in calculating the fall probability.

[0063] Traditional voiceprint signal decomposition methods mostly use fixed order decomposition, which cannot adapt to the complexity changes of voiceprint signals in home environments, such as complex signals mixed with fall sounds and environmental noise. This leads to incomplete decomposition, effective features being covered by residual terms or confused with noise terms, and the inability to accurately extract fall-related voiceprint features, affecting the accuracy of subsequent voiceprint entropy calculation and reducing the reliability of fall judgment.

[0064] Based on this, the decomposition operation performed by the voiceprint modal decomposition unit satisfies:

[0065] ;

[0066] wherein, represents the ambient acoustic signal, in volts, quantifying the voltage amplitude of the acoustic signal; represents the i-th intrinsic modal function, characterizing the oscillatory components of different frequencies in the acoustic signal, in volts. Each order corresponds to the acoustic features of a specific frequency band, such as the low-frequency components of the fall impact and the mid-high frequency components of the help-seeking sound. represents the residual term, characterizing the trend components in the acoustic signal, in volts, reflecting the slowly changing part of the signal, such as the continuous background noise; n represents the decomposition order, whose value is dynamically determined according to the complexity of the acoustic signal. The more complex the frequency components of the signal, the larger the value of n.

[0067] This formula is the core operation of the acoustic modal decomposition unit for processing the ambient acoustic signal. Its core function is to decompose the complex original acoustic signal into independent analysis components, laying the foundation for subsequent extraction of fall-related acoustic features.

[0068] From a technical implementation perspective, represents the original ambient acoustic signal, whose physical meaning is a function of time , covering all acoustic information in the home environment, including the impact sound and help-seeking sound when the elderly fall, as well as interference noise such as pet activity and object falling.

[0069] The right side of the formula decomposes into two parts: one is , the sum of n-order intrinsic modal functions (IMF), each order corresponding to a specific frequency oscillatory component in the acoustic signal, for example, the low-frequency oscillation produced by the fall impact will be concentrated in the low-order IMF, while the mid-high frequency component of the help-seeking sound will be distributed in the high-order IMF; the second is , the residual term, used to characterize the slowly changing trend in the acoustic signal, such as continuous background noise, such as air conditioner running sound.

[0070] The dynamic adjustment of the decomposition order is the key improvement point of this formula: when the ambient acoustic signal is simple, such as only a single background noise, a smaller value can be used for sufficient decomposition; when the signal is complex, such as the superposition of fall sound and multiple interference noise, automatically increases to ensure that each IMF contains only a single frequency component, avoiding modal aliasing.

[0071] This dynamic decomposition mechanism solves the problem of insufficient adaptation of traditional fixed-order decomposition to complex scenarios, enabling the subsequent acoustic entropy value calculation to accurately capture the changes in fall-related acoustic features.

[0072] From the perspective of patent law requirements, the disclosure of the formula fully embodies the implementability of the technical solution: the skilled person in the art can determine the value of the frequency complexity of the voiceprint signal, such as the preliminary analysis of the spectral distribution by the fast Fourier transform , effectively distinguish between fall sounds and interference noises by analyzing the amplitude and frequency characteristics of each IMF, and provide clear support for the anti-interference technical features of the voiceprint modal decomposition in the embodiments.

[0073] The technical solution realizes the adaptive decomposition of the voiceprint signal through an improved intrinsic modal decomposition algorithm: the formula decomposes the complex voiceprint signal into n-order intrinsic modal functions and residual terms Each parameter is clear and dimensionless, and is in volts, ensuring clear physical meaning; the decomposition order n is dynamically adjusted according to the signal complexity, solving the problem of insufficient adaptation of fixed-order decomposition to complex signals - when the signal contains multiple frequency components, such as fall sounds plus environmental noise, the value of n is increased to fully decompose, and when the signal is simple, such as a single background sound, the value of n is reduced to avoid over-decomposition.

[0074] The technical effect is that dynamic order decomposition ensures the full decomposition of voiceprint signals of different complexities, so that fall-related features such as the specific frequency oscillation of impact sound are accurately extracted into the corresponding , avoiding confusion with noise or trend components; the clear dimension of each parameter provides a quantitative basis for subsequent voiceprint entropy calculation, improving the accuracy of entropy in describing the complexity of voiceprint, and providing reliable voiceprint feature input for the multi-modal decision unit.

[0075] Traditional human displacement and electromagnetic field correlation models mostly use first-order signal models, which can only capture the instantaneous changes of human dynamic motion and cannot reflect the continuous distortion characteristics of electromagnetic fields in the stationary state after a fall, resulting in failure to recognize the dangerous state of no motion after a fall. At the same time, existing models do not consider the influence of medium coupling and time attenuation on electromagnetic field signals, making the distortion value calculation deviate greatly from the actual scene, and unable to provide accurate electromagnetic feature basis for fall judgment.

[0076] Therefore, the human displacement-electromagnetic field distortion model established by the electromagnetic wave distortion analysis unit satisfies: ;

[0077] Wherein, represents the electromagnetic field distortion value, in volts per meter, used to quantify the intensity change of the electromagnetic field due to human displacement; represents the medium coupling coefficient, used to reflect the coupling degree of the human body and air medium to the electromagnetic wave, without unit, and its value is determined by the difference in dielectric constant between human tissues and air; represents the three-dimensional displacement vector of the human body, with units of meters, where x, y, and z represent the displacement components in the x-axis, y-axis, and z-axis directions of the spatial rectangular coordinate system, respectively, and fully describes the position change of the human body in three-dimensional space; represents the attenuation coefficient, which is used to represent the initial attenuation amplitude of the electromagnetic field distortion, with units of volts per meter, and reflects the initial intensity of the distorted signal; represents the time attenuation rate, with units of per second, which is used to represent the attenuation speed of the electromagnetic field distortion over time, and the larger the value, the faster the signal attenuation; t represents time, with units of seconds, and is used to mark the time node of signal change.

[0078] This formula is the core model constructed by the electromagnetic wave distortion analysis unit, which is used to quantify the correlation between human displacement and electromagnetic field distortion, breaking through the limitation of traditional first-order models that can only monitor dynamic motion, and achieving continuous monitoring of the static state after a fall, such as coma.

[0079] The left side of the formula represents the electromagnetic field distortion value, which has the physical meaning of the change in electromagnetic field intensity after the human body enters the microwave sensor monitoring range, and directly reflects the disturbance degree of the human body to the electromagnetic field.

[0080] The right side of the formula contains two key terms: the first term is the second derivative term, where is the three-dimensional displacement vector of the human body, with units of meters, and its second time derivative reflects the acceleration change of the human body motion, specifically capturing the rapid displacement at the moment of falling, such as the accelerated falling when the human body loses balance; the second term is the exponential decay term, where is the initial attenuation amplitude, with units of volts per meter, is the time attenuation rate, with units of per second, which is used to describe the continuous distortion of the electromagnetic field when the human body is static after a fall - since the human body is still in the electromagnetic field, the distortion does not disappear instantaneously, but slowly decays over time. This characteristic enables the model to identify the coma state after a fall, while traditional first-order models fail due to their reflection of only instantaneous changes.

[0081] The introduction of the medium coupling coefficient is an important improvement of the model, and its value is determined by the difference in dielectric constant between human tissue and air, ensuring that the mapping relationship between displacement and distortion remains accurate under different body types and different clothing conditions. This second-order model design covers both dynamic falling processes and static continuous states, providing specific implementation basis for the technical features of the second-order electromagnetic distortion model in the embodiments to identify the coma state,

[0082] This technical solution achieves a comprehensive characterization of human displacement through a second-order electromagnetic distortion model: the second-order partial derivative term Capture the acceleration change of human motion, such as the rapid displacement at the moment of falling, exponential decay term Then reflect the continuous distortion of the electromagnetic field in the static state after falling, because the human body is still in the electromagnetic wave field, and the combination of the two covers the whole scene of dynamic motion and static existence. The dimensions of each parameter are clear and the physical meaning is clear, ensuring the interpretability and accuracy of the model calculation.

[0083] The technical effect is embodied in: the second-order model breaks through the limitation of the traditional first-order model that can only monitor dynamic motion, can effectively identify the static state after falling, and solves the problem of missing judgment of coma state; The introduction of medium coupling coefficient And time decay rate Make the calculation of electromagnetic field distortion value more in line with the actual environment, considering the medium difference between human body and air and the natural attenuation of signal, and improve the adaptability of the model to real scene, providing a reliable foundation for accurate calculation of electromagnetic wave distortion rate.

[0084] Traditional fall probability calculation mostly uses simple weighted summation to fuse multi-modal features, without considering the non-linear relationship between features, resulting in large deviation between probability value and actual fall risk mapping; At the same time, the weight coefficient is fixed and cannot be dynamically adjusted according to the reliability of the feature, so when a certain modality signal is disturbed, such as unreliable voiceprint signal under strong noise, it is still fused according to the fixed weight, which is easy to cause distortion of probability calculation and affect the accuracy of fall judgment.

[0085] Therefore, the formula for calculating the fall probability of the multi-modal decision unit satisfies:

[0086] ;

[0087] Wherein, represents the fall probability, unitless, the value range is 0 to 1, and the value closer to 1 indicates a higher possibility of falling; represents the weight coefficient of voiceprint entropy value, unitless, used to adjust the contribution of voiceprint feature in fusion; represents the weight coefficient of electromagnetic wave distortion rate, unitless, used to adjust the contribution of electromagnetic feature; represents the voiceprint entropy value, unitless, used to quantify the complexity of voiceprint signal, and the higher the entropy value, the more chaotic the signal, such as impact sound when falling; represents the threshold value of voiceprint entropy value, unitless, which is the critical value for judging falling of voiceprint feature; represents the electromagnetic wave distortion rate, unitless, used to quantify the distortion degree of electromagnetic field signal, and the higher the distortion rate, the more intense the human displacement; represents the threshold value of electromagnetic wave distortion rate, unitless, which is the critical value for judging falling of electromagnetic feature.

[0088] The formula is the core algorithm of the multimodal decision unit to fuse the voiceprint and electromagnetic wave features. The double-mode features are converted into a fall probability through nonlinear mapping, solving the problem that the traditional linear weighted fusion cannot accurately reflect the nonlinear relationship between the features and the fall risk.

[0089] The left side of the formula represents the fall probability, with a value of 0 to 1. The closer the value is to 1, the higher the possibility of falling. The right side of the formula uses a sigmoid function to achieve nonlinear fusion: where is the voiceprint entropy value, quantifying the complexity of the voiceprint, is the electromagnetic wave distortion rate, quantifying the degree of electromagnetic field disturbance; and are the threshold values of the voiceprint entropy value and the electromagnetic wave distortion rate, respectively, used to define the critical values of the normal state and the abnormal state; and are the weight coefficients, used to dynamically adjust the contribution of the two modal features. When the environmental noise is high and the voiceprint signal reliability decreases, the weight of the electromagnetic wave feature is increased to improve the weight of the electromagnetic wave feature, and vice versa. .

[0090] The advantage of this fusion mechanism is that the nonlinear characteristics of the sigmoid function make the probability value sensitive to feature mutations, such as sudden increases in voiceprint entropy and electromagnetic distortion rate during a fall, while remaining stable to small fluctuations, such as small disturbances caused by pet activity. Dynamic weight adjustment allows the model to adapt to different environmental scenarios, avoiding false positives caused by the failure of a single modality.

[0091] The specific parameters of the formula, such as the threshold and the weight, can be determined through a large number of sample training, providing a clear implementation path for those skilled in the art, and fully supporting the technical effect of the dual-mode collaborative decision-making in the embodiment to reduce false positives.

[0092] This technical solution achieves nonlinear fusion of multimodal features through a sigmoid function: the formula converts the deviation between the voiceprint entropy value and the electromagnetic wave distortion rate into a fall probability through an exponential function, allowing the probability value to respond nonlinearly to changes in the feature, such as when the feature is far beyond the threshold, the probability quickly approaches 1, which is more consistent with the nonlinear distribution of actual fall risk. The weight coefficients , can be dynamically adjusted to adapt to the reliability of different modalities.

[0093] The technical effect is that the nonlinear fusion method can better capture the true relationship between the feature and the fall risk compared to the traditional linear weighted fusion, making the probability calculation more accurate. The dynamic weight adjustment mechanism allows the weight of a certain modality to be reduced when the signal is disturbed, such as reducing and increasing , the robustness of fusion is improved; the normalized range of probability value facilitates the setting of a unified judgment threshold, providing an intuitive and comparable quantitative index for trigger logic that exceeds the preset threshold and lasts for a preset duration, ultimately improving the accuracy and reliability of fall judgment.

[0094] The traditional home safety monitoring method for the elderly has a simple process, relies on single sensor data to directly determine falls, does not go through multi-modal feature extraction and fusion, and cannot filter environmental interference; at the same time, it lacks time dimension verification of the fall state, is easy to trigger false alarms due to transient interference signals, and cannot accurately capture key information such as fall location and duration, affecting the effectiveness of early warning and rescue efficiency.

[0095] Based on this, please refer to Figure 4 The embodiment provides a home safety monitoring and early warning method for the elderly based on the Internet of Things, comprising:

[0096] S1: Collecting home environment data of the elderly through Internet of Things devices and transmitting to the processing center;

[0097] S2: Collecting environmental voiceprint signals, decomposing the environmental voiceprint signals using an improved empirical mode decomposition algorithm to obtain multiple intrinsic mode functions and residual terms, and calculating voiceprint entropy values based on the intrinsic mode functions;

[0098] S3: Deploying a 5.8 GHz microwave sensor to collect electromagnetic field signals of the home environment, establishing a second-order model of human body displacement and electromagnetic field distortion based on the electromagnetic field signals, and calculating the electromagnetic wave distortion rate through the second-order model;

[0099] S4: Inputting the voiceprint entropy value and the electromagnetic wave distortion rate into a multi-modal decision model to calculate the fall probability;

[0100] S5: When the fall probability exceeds the preset threshold and the state lasts for a preset duration, triggering the early warning mechanism and issuing early warning information containing the fall location and duration.

[0101] This technical solution constructs a complete method process of collection, processing, fusion, decision-making, and early warning: first, through voiceprint collection and decomposition to extract voiceprint entropy values, and through electromagnetic wave sensing and second-order model calculation to calculate distortion rates, realizing parallel extraction of multi-modal features; then inputting the features into a multi-modal decision model to calculate the fall probability, introducing time dimension verification to ensure the stability of the judgment; finally, triggering the early warning and containing key information, location, and duration. The logic of each step is coherent, forming a closed loop from data collection to early warning output.

[0102] The technical effects are embodied in that: the multi-modal feature extraction solves the problem of single mode being easily disturbed, the voiceprint entropy value and the electromagnetic wave distortion rate are complementary, and the reliability of the features is improved; the time dimension verification reduces false positives caused by instantaneous interference, and enhances the robustness of decision-making; the early warning information contains the position and duration, provides accurate guidance for rescue, and solves the problem of traditional early warning information being ambiguous. The overall process covers all links from signal processing to early warning execution, ensuring the accuracy of monitoring and the effectiveness of early warning.

[0103] The traditional voiceprint signal processing method lacks targeted preprocessing steps, and the mixed low frequency such as ground resonance vibration noise and high frequency electromagnetic interference in the original signal are not effectively removed, resulting in the intrinsic mode function decomposed subsequently containing a large amount of noise components; at the same time, the constraint condition of the intrinsic mode function is not standardized in the empirical mode decomposition process, and the decomposition result is prone to modal aliasing, which cannot accurately extract the voiceprint features related to falling, affecting the accuracy of voiceprint entropy value calculation.

[0104] Based on this, refer to Figure 5 , the steps of collecting the environmental voiceprint signal and calculating the voiceprint entropy value include:

[0105] S11: Collecting the environmental voiceprint signal by using an omnidirectional microphone array, and performing band-pass filtering processing on the environmental voiceprint signal to remove low-frequency vibration noise and high-frequency electromagnetic interference;

[0106] S12: Removing the direct current component of the filtered voiceprint signal to obtain an alternating voiceprint signal;

[0107] S13: Decomposing the alternating voiceprint signal by using an improved empirical mode decomposition algorithm, wherein each intrinsic mode function in the decomposition process needs to satisfy that the mean value of the upper and lower envelope lines is 0, and the number of zero-crossing points is equal to or differs from the number of extreme points by at most one;

[0108] S14: Calculating the Shannon entropy based on the intrinsic mode functions obtained by decomposition, and taking the Shannon entropy as the voiceprint entropy value, wherein the amplitude of each intrinsic mode function needs to be normalized in the calculation process.

[0109] The technical scheme improves the quality of voiceprint features through multi-step preprocessing and standardized decomposition: band-pass filtering removes low-frequency and high-frequency interference in a targeted manner, ensuring that the effective frequency components of the fall voiceprint are retained; removing the direct current component eliminates the static offset in the signal, making the alternating voiceprint signal purer; the constraint condition of the empirical mode decomposition, i.e., the mean value of the envelope lines is 0 and the number of zero-crossing points matches the number of extreme points, avoids modal aliasing, ensuring that each intrinsic mode function corresponds to a single frequency component; amplitude normalization provides a uniform scale for Shannon entropy calculation. Each step is closely linked to form a high-quality conversion link from the original signal to the feature entropy value.

[0110] The technical effects are that: the preprocessing step effectively filters out noise, significantly improving the signal-to-noise ratio of the voiceprint signal; the normalized decomposition ensures that the intrinsic mode function accurately reflects the frequency characteristics of the fall sound, such as the low-frequency component of the impact sound and the medium-high frequency component of the help sound; the Shannon entropy value is calculated based on pure features, which can more accurately quantify the complexity of the voiceprint signal, provide reliable voiceprint input for multi-modal decision-making, and reduce false positives caused by feature distortion.

[0111] The traditional early warning trigger mechanism is simple, only triggers early warning when the single fall probability exceeds the threshold, does not consider the probability fluctuation caused by instantaneous interference, and is prone to false positives; at the same time, the early warning information lacks key details such as specific location and duration, and is not synchronized to multiple terminals, affecting the timeliness and accuracy of rescue, and cannot meet the emergency response needs of the elderly at home.

[0112] Based on this, please refer to Figure 6 , the steps of triggering the early warning mechanism include:

[0113] S41: When the calculated fall probability exceeds the preset threshold, start the secondary verification process, which includes calculating the change rate of the voiceprint entropy value and the change rate of the electromagnetic wave distortion rate within a preset period, and when the change rate of the voiceprint entropy value and the change rate of the electromagnetic wave distortion rate both meet the preset conditions, determine that the secondary verification is passed;

[0114] S42: After the secondary verification is passed, start the time window monitoring for a preset duration, and continuously record the fall probability within the time window;

[0115] S43: If the fall probability within the preset time window exceeds the preset threshold, determine that the early warning mechanism is triggered;

[0116] S45: The early warning information is sent to the preset guardian terminal and community service platform through the Internet of Things communication module, and the early warning information includes the specific location of the fall, the fall duration, and the characteristic parameters of the voiceprint and electromagnetic wave signals, and the specific location is determined based on the signal strength triangulation of each sensor node.

[0117] This technical solution combines secondary verification and time window monitoring to build a multi-level early warning trigger mechanism: the secondary verification passes the feature change rate, the voiceprint entropy value change rate, and the electromagnetic wave distortion rate change rate to determine whether the probability surge is caused by a real fall or instantaneous interference; Time window monitoring ensures the persistence of the fall state and avoids accidental triggering; the early warning information contains multi-dimensional details such as location, duration, and characteristic parameters and is synchronized to multiple terminals, improving response efficiency. Each step forms a progressive verification logic to ensure the accuracy and comprehensiveness of the early warning.

[0118] Technical effects are embodied in: the secondary verification and time window monitoring significantly reduce the false positive rate, making the early warning more reliable; the specific location is determined by triangulation, solving the problem of traditional early warning location ambiguity; multi-terminal synchronization and detailed information provide accurate guidance for rescue, shortening the response time; the inclusion of characteristic parameters provides data support for subsequent review and system optimization, and the overall practicality of the early warning mechanism and the effectiveness of emergency rescue are improved.

[0119] The above is only a preferred embodiment of the present application, not other forms of limitations on the present application, any skilled in the art can use the above disclosed technical content to change or modify as equivalent embodiments applied to other fields, but any simple modification, equivalent change and modification made to the above embodiments without departing from the technical solution content of the present application, according to the technical essence of the present application, still belongs to the protection scope of the technical solution of the present application.

Claims

1. An Internet of Things-based home safety monitoring and early warning system for the elderly, comprising a sensor node for collecting home environment data, an Internet of Things communication module for transmitting data, a central processing module for processing data, and an early warning module for issuing warning information; characterized in that: The sensor node comprises a voiceprint acquisition sub-node and an electromagnetic wave sensing sub-node, and the central processing module is integrated with a voiceprint modal decomposition unit, an electromagnetic wave distortion analysis unit and a multi-modal decision unit; the voiceprint modal decomposition unit performs intrinsic modal function decomposition on the environmental voiceprint signal acquired by the voiceprint acquisition sub-node, the electromagnetic wave distortion analysis unit establishes a second-order correlation model of human body displacement and electromagnetic field distortion based on the electromagnetic field signal collected by the electromagnetic wave sensing sub-node, the second-order correlation model comprises a time decay rate, and the time decay rate is used to describe the continuous distortion of the electromagnetic field when the human body is still after falling; the multi-modal decision unit fuses the voiceprint entropy value obtained through voiceprint modal decomposition and the distortion rate obtained through electromagnetic wave distortion analysis to calculate a fall probability; when the fall probability exceeds a preset threshold and the state lasts for a preset length of time, the multi-modal decision unit sends a trigger signal to a warning module, and the warning module sends a warning information based on the trigger signal; the electromagnetic wave sensing sub-node comprises a 5.8GHz microwave sensor, a radio frequency front-end circuit and a data buffer, the 5.8GHz microwave sensor is respectively arranged on the central ceiling of the living room, above the bedside of the bedroom, the top of the bathroom and above the entrance of the kitchen, and the multi-modal decision unit comprises a weight dynamic adjustment sub-unit, the weight dynamic adjustment sub-unit adjusts the fusion weight of the voiceprint entropy value and the electromagnetic wave distortion rate according to the environmental noise intensity, the weight proportion of the electromagnetic wave distortion rate is increased when the environmental noise is high, and the weight proportion of the voiceprint entropy value is increased when the environmental noise is low. 2.The home safety monitoring and early warning system for the elderly based on the Internet of Things according to claim 1, characterized in that, The voiceprint acquisition sub-node comprises an omnidirectional microphone array, a signal conditioning circuit and an analog-to-digital converter, the output end of the omnidirectional microphone array is connected with the input end of the signal conditioning circuit, the output end of the signal conditioning circuit is connected with the input end of the analog-to-digital converter, and the output end of the analog-to-digital converter is connected with the input end of the central processing module through an Internet of Things communication module; the omnidirectional microphone array collects environmental voiceprint signals, the signal conditioning circuit performs gain adjustment and filtering processing on the collected voiceprint signals, and the analog-to-digital converter converts the conditioned analog voiceprint signals into digital signals. 3.The home safety monitoring and early warning system for the elderly based on the Internet of Things according to claim 1, characterized in that, The electromagnetic wave sensing sub-node comprises a 5.8GHz microwave sensor, a radio frequency front-end circuit and a data buffer, the output end of the 5.8GHz microwave sensor is connected with the input end of the radio frequency front-end circuit, the output end of the radio frequency front-end circuit is connected with the input end of the data buffer, and the output end of the data buffer is connected with the input end of the central processing module through an Internet of Things communication module; the 5.8GHz microwave sensor is respectively arranged on the central ceiling of the living room, above the bedside of the bedroom, the top of the bathroom and above the entrance of the kitchen, the radio frequency front-end circuit performs mixing, filtering and amplification processing on the radio frequency signals output by the sensor, and the data buffer temporarily stores the processed electromagnetic field signals.

4. The home security monitoring and early warning system for the elderly based on the Internet of Things according to claim 1, characterized in that, The multi-modal decision unit comprises a feature normalization sub-unit, a weight dynamic adjustment sub-unit and a time window verification sub-unit, the input ends of the feature normalization sub-unit are connected with the output ends of the voiceprint modal decomposition unit and the output end of the electromagnetic wave distortion analysis unit respectively, the output end of the feature normalization sub-unit is connected with the input end of the weight dynamic adjustment sub-unit, and the output end of the weight dynamic adjustment sub-unit is connected with the input end of the time window verification sub-unit; The feature normalization sub-unit maps the voiceprint entropy value and the electromagnetic wave distortion rate to a preset numerical interval respectively, the weight dynamic adjustment sub-unit adjusts the fusion weight of the voiceprint entropy value and the electromagnetic wave distortion rate according to the environmental noise intensity, the weight proportion of the electromagnetic wave distortion rate is increased when the environmental noise is higher, and the weight proportion of the voiceprint entropy value is increased when the environmental noise is lower, and the time window verification sub-unit continuously monitors whether the fall probability meets the condition of exceeding the preset threshold value and records the duration of the state. 5.The home safety monitoring and early warning system for the elderly based on the Internet of Things according to claim 1, characterized in that, The decomposition operation performed by the voiceprint modal decomposition unit satisfies: ； wherein, represents an ambient voiceprint signal, in volts; represents an i-th order eigenmode function, used to represent different frequency oscillation components in the voiceprint signal, in volts; represents a residual term, used to represent a trend component in the voiceprint signal, in volts; n represents a decomposition order, whose value is dynamically determined according to the complexity of the voiceprint signal, the more complex the frequency components of the signal are, the larger the value of n is. 6.The home safety monitoring and early warning system for the elderly based on the Internet of Things according to claim 1, characterized in that, The human body displacement-electromagnetic field distortion model established by the electromagnetic wave distortion analysis unit satisfies: ； wherein, represents the electromagnetic field distortion value, unit: volt per meter; represents the medium coupling coefficient, used to reflect the coupling degree of human body and air medium to electromagnetic wave; represents the three-dimensional displacement vector of human body, unit: meter, wherein x, y, z respectively represent the displacement components in the x-axis, y-axis, z-axis directions of the space rectangular coordinate system; represents the attenuation coefficient, used to represent the initial attenuation amplitude of electromagnetic field distortion; represents the time attenuation rate, unit: per second, used to represent the attenuation speed of electromagnetic field distortion with time; t represents time, unit: second. 7.The home safety monitoring and early warning system for the elderly based on the Internet of Things according to claim 1, characterized in that, The formula for calculating the fall probability by the multi-modal decision unit satisfies: ； wherein, represents a probability of falling; represents a weight coefficient of the voiceprint entropy value; represents a weight coefficient of the electromagnetic wave distortion rate; represents a voiceprint entropy value, used to quantify the complexity of the voiceprint signal; represents a threshold value of the voiceprint entropy value; represents an electromagnetic wave distortion rate, used to quantify the distortion degree of the electromagnetic field signal; represents a threshold value of the electromagnetic wave distortion rate.

8. An Internet of Things-based home safety monitoring and early warning method for the elderly, applied to an Internet of Things-based home safety monitoring and early warning system for the elderly according to any one of claims 1-7, comprising: The method comprises the following steps: Collecting environmental voiceprint signals and calculating voiceprint entropy values comprise: Collecting environmental voiceprint signals and calculating voiceprint entropy values comprise: Deploying a 5.8 GHz microwave sensor to collect electromagnetic field signals of the home environment, establishing a second-order model of human body displacement and electromagnetic field distortion based on the electromagnetic field signals, and calculating an electromagnetic wave distortion rate through the second-order model; Inputting the voiceprint entropy value and the electromagnetic wave distortion rate into a multi-modal decision model to calculate a fall probability; 9.The home safety monitoring and early warning method for the elderly based on the Internet of Things according to claim 8, characterized in that, When the fall probability exceeds a preset threshold value and the state lasts for a preset duration, triggering a warning mechanism and issuing a warning information containing a fall position and a duration. The step of collecting environmental voiceprint signals and calculating voiceprint entropy values comprises: Collecting environmental voiceprint signals and calculating voiceprint entropy values comprise: Collecting environmental voiceprint signals and calculating voiceprint entropy values comprise: The step of triggering the warning mechanism comprises: 10.The home safety monitoring and early warning method for the elderly based on the Internet of Things according to claim 8, characterized in that, When the calculated fall probability exceeds a preset threshold value, starting a secondary verification process, the secondary verification process comprises calculating the change rate of the voiceprint entropy value and the change rate of the electromagnetic wave distortion rate within a preset period, and when the change rate of the voiceprint entropy value and the change rate of the electromagnetic wave distortion rate both meet the preset conditions, determining that the secondary verification is passed. ​ After the secondary verification, a preset time window monitoring is started, and the fall probability in the time window is recorded continuously; If the fall probability in the preset time window exceeds a preset threshold, it is determined that the early warning mechanism is triggered; The early warning information is sent to the preset guardian terminal and community service platform through the Internet of Things communication module, and the early warning information includes the specific location of the fall, the fall duration, and the characteristic parameters of the voiceprint and electromagnetic wave signal, and the specific location is determined based on the signal strength triangulation of each sensor node.

Images