Method and device for detecting abnormal breathing state of elevator passengers based on millimeter waves

Abstract

The invention relates to a millimeter wave-based elevator passenger breathing state abnormity detection method and device. The elevator passenger breathing state anomaly detection method based on the millimeter waves comprises the steps that body micro-motion data of elevator passengers are obtained; adopting a millimeter wave micro-motion feature algorithm to perform respiration feature extraction on the body micro-motion data, and adopting a respiration fitting model to perform correlation matching on the respiration data subjected to the respiration feature extraction to obtain a respiration frequency estimation value; calculating the respiratory frequency estimation value by adopting a respiratory abnormal frequency weighting algorithm to obtain a respiratory frequency abnormal value; judging whether the respiratory rate abnormal value meets a behavior abnormal condition or not; if yes, sending alarm information; and if not, continuing to execute detection. The elevator passenger breathing state anomaly detection method based on the millimeter waves has the advantages that the privacy data leakage risk is low, and the environment robustness of the system is high.

Description

Technical Field

[0001] The present invention relates to the field of abnormal detection of physiological signals, and particularly to a method for detecting abnormal breathing states of elevator passengers based on millimeter waves, a device for detecting abnormal breathing states of elevator passengers, and a system for monitoring abnormal behaviors of elevator passengers. Background Art

[0002] With the acceleration of the urbanization process and the popularization of high-rise buildings, elevators have become an indispensable vertical transportation facility in modern life. However, sudden failures may occur during elevator operation, such as sudden stops, outages, long-term accelerations, continuous descents, etc., which may cause elevator passengers to be trapped in the enclosed space of the elevator and lead to physiological and psychological abnormalities of elevator passengers, such as panic, hypoxia, and even syncope. This acute stress response will hinder the elevator passengers from calling for help independently, increase the difficulty of rescue and the rescue response time, and pose serious safety hazards.

[0003] The existing technologies for monitoring elevator passengers include contact and non-contact methods. The contact method identifies the identity of elevator passengers and associates sensors of elevator passengers, such as smart watches or other physiological data collection sensors, to collect the physiological information of elevator passengers and construct state thresholds for corresponding abnormal and non-abnormal states, so as to achieve abnormal early warning. However, the existing contact technologies have the risk of privacy data leakage and require elevator passengers to actively cooperate with the detection device.

[0004] The non-contact method generally uses video image analysis. Generally, deep learning technologies, such as convolutional neural networks, LSTM, etc., are used to process the collected video data to identify whether the corresponding actions conform to the abnormal labels. However, when sudden situations occur in the elevator resulting in the absence or occlusion of light sources, it is easy to cause the failure of the video image analysis method. At the same time, using video images also has the risk of privacy data leakage. Accordingly, the existing technologies for abnormal monitoring of elevator passengers have the problems of high risk of privacy data leakage and poor environmental robustness. Summary of the Invention

[0005] Based on this, the purpose of the present invention is to provide a method for detecting abnormal breathing states of elevator passengers based on millimeter waves.

[0006] A method for detecting abnormal breathing states of elevator passengers based on millimeter waves includes the following steps:

[0007] S1. Obtain the body micro-motion data of the elevator passenger; wherein, the body micro-motion data of the elevator passenger is collected by a millimeter wave radar;

[0008] S2. Use a millimeter wave micro-motion feature algorithm to extract breathing features from the body micro-motion data of the elevator passenger to obtain the breathing data of the current elevator passenger;

[0009] S3. Use a breathing fitting model to perform correlation matching on the breathing data of the current elevator rider to obtain an estimated breathing frequency value of the current elevator rider;

[0010] S4. Use a breathing anomaly frequency weighting algorithm to calculate the estimated breathing frequency value of the current elevator rider to obtain a breathing frequency anomaly value of the current elevator rider;

[0011] S5. Determine whether the breathing frequency anomaly value of the current elevator rider meets a behavior anomaly condition: if so, send an alarm message; if not, continue to execute step S1.

[0012] The method for detecting abnormal breathing status of elevator riders based on millimeter waves according to the present invention, compared with the prior art, uses millimeter wave radar to non-contact sense the body micro-motion data of elevator riders, and combines multi-stage algorithms for collaborative optimization to extract breathing signals in an anonymized manner, effectively solving the problem of high risk of privacy data leakage in traditional monitoring technologies; at the same time, the present invention effectively suppresses environmental interference through the millimeter wave micro-motion feature algorithm and the breathing fitting model, that is, a predetermined physiological frequency template, and significantly improves the environmental robustness of the system.

[0013] Further, the millimeter wave micro-motion feature algorithm includes the following steps:

[0014] S21. Construct feature points based on the body micro-motion data of the current elevator rider, and calculate the offset according to the feature points to obtain the displacement data of the elevator rider; among them, for the displacement data of the i-th feature point at time t The specific calculation is expressed as follows:

[0015]

[0016] In the formula, λ represents the wavelength of the millimeter wave emitted by the millimeter wave radar; Δp(t,r i ) represents the signal intensity of the body micro-motion data of the elevator rider of the echo at the t-th moment and a distance of r i ; n represents the total number of feature points, and its specific calculation is expressed as follows:

[0017] n = 4u + 1

[0018] In the formula, u represents the distribution quantity of feature points in different directions;

[0019] S22. Normalize the displacement data of the elevator rider, and use the non-negative matrix factorization algorithm to decompose and reconstruct the normalized data to obtain a set of potential breathing data to be extracted; among them, for the normalized displacement data z i (t) of the i-th feature point at the t-th moment, the specific calculation is expressed as follows:

[0020]

[0021] Wherein, μ(t) is the mean value; σ represents the standard deviation; C represents the translation data;

[0022] S23. Use the analysis of variance method to extract respiratory data from the set of potential respiratory data to be extracted, and obtain the respiratory data of the current elevator rider.

[0023] The present invention constructs feature points and calculates displacements of elevator riders to improve the resolution of respiratory signals in millimeter waves, and suppresses the interference of non-respiratory signals and noise through normalization, NMF decomposition and reconstruction, significantly improving the reliability of respiratory abnormality detection in complex elevator environments.

[0024] Further, the non-negative matrix factorization algorithm is used for decomposition and reconstruction to obtain the set of potential respiratory data to be extracted, which is specifically expressed as follows:

[0025]

[0026] Wherein, Z represents the displacement data of the elevator rider after normalization, and its dimension is S * and T * respectively represent the sets of all optimal spatial pattern matrices and optimal temporal pattern matrices after Q times of decomposition and reconstruction by the non-negative matrix factorization algorithm, and are used to represent the set of potential respiratory data to be extracted, which is specifically expressed as follows:

[0027]

[0028] Wherein, Q represents the total number of reconstructions; represents the weight coefficient of the k-th mode of the q-th optimal spatial pattern matrix contributing to the respiratory signal at all feature points, and its dimension is represents the time-domain waveform of the k-th mode of the q-th optimal temporal pattern matrix, that is, the time series, and its dimension is r represents the rank of decomposition using the non-negative matrix factorization algorithm;

[0029] Among them, for the set of potential respiratory data S * and T * in the q-th reconstruction of the optimal spatial matrix S (q) and the optimal temporal matrix T (q) , the specific steps of their decomposition and reconstruction are expressed as follows:

[0030] S221. Randomly initialize the spatial pattern matrix and the temporal pattern matrix to generate non-negative matrices S (0) and T (0) ;

[0031] S222. Update the non - negative spatial pattern matrix and temporal pattern matrix to obtain the spatial pattern matrix and temporal pattern matrix for the current iteration, which are specifically expressed as follows:

[0032]

[0033] where S (q,m) and T (q,m) respectively represent the m - th iteration spatial pattern matrix and temporal pattern matrix of the q - th reconstruction; ⊙ represents element - by - element multiplication; ∈ represents removing zero terms;

[0034] S223. Determine whether the spatial pattern matrix and temporal pattern matrix for the current iteration meet a convergence condition; if so, obtain the spatial pattern matrix and temporal pattern matrix with the minimum reconstruction error, and use them as the optimal spatial pattern matrix and optimal temporal pattern matrix for the q - th reconstruction; if not, continue to execute step S222;

[0035] Among them, the specific judgment of the convergence condition is expressed as follows:

[0036]

[0037] where ε represents the threshold of the minimum reconstruction error change rate, which is used as the convergence condition, and its value range is [0, 1]; m max represents the maximum number of iterations, which is used as the convergence condition; is used to represent the reconstruction error change rate, and the reconstruction error for its m - th iteration is specifically calculated as follows:

[0038]

[0039] The specific representation of obtaining the spatial pattern matrix and temporal pattern matrix with the minimum reconstruction error is as follows:

[0040]

[0041] where represents obtaining the spatial pattern matrix S and temporal pattern matrix T corresponding to the minimum reconstruction error during all current iteration processes and using them as the optimal spatial pattern matrix and optimal temporal pattern matrix for the q - th reconstruction.

[0042] The present invention performs low-rank decomposition on the normalized displacement data through the non-negative matrix factorization (NMF) algorithm, extracts the potential respiratory signals from the time patterns, and then uses analysis of variance to directly extract the time patterns with significant differences as the most significant respiratory signals, thereby improving the accuracy of respiratory signal extraction; in addition, the potential respiratory signals extracted by NMF can strip the privacy-related information such as position and body shape from the original millimeter-wave data, further avoiding the risk of privacy leakage; accordingly, the present invention further improves the accuracy of respiratory signal extraction, environmental robustness, and privacy protection level.

[0043] Furthermore, the specific representation of the respiratory fitting model is as follows:

[0044]

[0045] In the formula, Template j (t) is the respiratory fitting model, which is used to represent the sine template of the j-th candidate frequency; F i represents the j-th discretized frequency parameter, and its specific representation is:

[0046] F j = f low + f interval × (j - 1)

[0047] In the formula, f low represents the preset lower limit of the normal breathing frequency; f interval represents the breathing frequency increment interval; N represents the total number of discretized frequency parameters, and its specific calculation is as follows:

[0048]

[0049] In the formula, f up represents the upper limit of the normal breathing frequency;

[0050] represents the offset term; P represents the number of time periods generated by the sine signal of each frequency; its specific calculation is as follows:

[0051]

[0052] In the formula, Time total is the signal duration; represents rounding up;

[0053] Among them, the specific calculation of the correlation matching is as follows:

[0054]

[0055] In the formula, f prerepresents the estimated respiratory rate of the current elevator rider; k represents the index number, where k ∈ [1, N]; l j represents the Euclidean distance between the jth respiratory fitting model and the respiratory signal, which is specifically expressed as follows:

[0056]

[0057] In the formula, represents the time series of the most significant pattern.

[0058] Accordingly, the present invention generates a discretized sine template, that is, a respiratory fitting model, through a preset physiological respiratory rate range, and combines periodic extension to exclude abnormal interference and improve the detection accuracy of the respiratory rate; at the same time, by minimizing the Euclidean distance between the respiratory fitting model and the respiratory signal extracted by NMF, the data belonging to the respiratory rate, that is, the estimated respiratory rate, is screened out in the time pattern with the most significant respiratory characteristics, so as to suppress the high-frequency noise remaining after signal decomposition, and further significantly improve the accuracy of the estimated respiratory rate.

[0059] Furthermore, the specific expression of the respiratory abnormality frequency weighting algorithm is as follows:

[0060] I(t) = w1·T low (t) + w2·T up (t) + w3·T sud (t)

[0061] In the formula, I(t) represents the abnormal value of the respiratory rate of the current elevator rider at the t-th moment; T low (t) represents the duration of too low respiratory rate at the t-th moment, and its specific calculation is expressed as follows:

[0062]

[0063] In the formula, Δt represents the step size of the time window; f pre (t) represents the estimated respiratory rate corresponding to the t-th moment; and T up represents the duration of too high respiratory rate at the t-th moment, and its specific calculation is expressed as follows:

[0064]

[0065] T sud (t) represents the duration of abnormal change in respiratory rate at the t-th moment, and its specific calculation is expressed as follows:

[0066]

[0067] In the formula, Δf pre (t) represents the frequency change amount between adjacent moments, and its specific calculation is:

[0068] Δf pre (t) = |f pre (t) - f pre (t - 1)|

[0069] θ th is the frequency mutation threshold; w1, w2, and w3 respectively represent the weight coefficients of the low frequency, high frequency, and mutation frequency changes.

[0070] The present invention extracts the abnormal respiratory rate value by fusing the persistent abnormalities of the respiratory rate, namely too low and too high, and combining the instantaneous mutation abnormalities for weighted summation, so as to balance the harm degrees of different abnormal types, thereby improving the clinical adaptability and scenario generalization of the respiratory state assessment.

[0071] Furthermore, the specific judgment representation of the behavior abnormality condition is as follows:

[0072]

[0073] In the formula, I(t) > th noraml is used to indicate that the abnormal respiratory rate value of the current elevator rider is greater than a normal threshold th normal ; is used to indicate that the respiratory signal in the current elevator is valid.

[0074] An abnormal respiratory state detection device for elevator riders includes a body micro - motion data acquisition unit, a respiratory feature extraction unit, a respiratory rate matching unit, a respiratory rate abnormal value extraction unit, and a behavior abnormality judgment unit;

[0075] The body micro - motion data acquisition unit is used to acquire the body micro - motion data of the elevator rider; among them, the body micro - motion data of the elevator rider is collected by a millimeter - wave radar;

[0076] The respiratory feature extraction unit is used to extract the respiratory features from the body micro - motion data of the elevator rider by using a millimeter - wave micro - motion feature algorithm to obtain the respiratory data of the current elevator rider;

[0077] The respiratory rate matching unit is used to perform correlation matching on the respiratory data of the current elevator rider by using a respiratory fitting model to obtain the estimated respiratory rate value of the current elevator rider;

[0078] The respiratory rate abnormal value extraction unit is used to calculate the abnormal respiratory rate value of the current elevator rider by using a respiratory abnormal frequency weighting algorithm;

[0079] The abnormal behavior judgment unit is used to judge whether the abnormal value of the breathing frequency of the current elevator rider meets an abnormal behavior condition: if so, an alarm message is sent; if not, the body micro-motion data acquisition unit is called continuously.

[0080] Further, the millimeter-wave micro-motion feature algorithm includes the following steps:

[0081] S21. Construct feature points based on the body micro-motion data of the current elevator rider, calculate the offset according to the feature points, and obtain the displacement data of the elevator rider; among them, for the displacement data of the i-th feature point at time t The specific calculation is expressed as follows:

[0082]

[0083] In the formula, λ represents the wavelength of the millimeter wave emitted by the millimeter-wave radar; Δp(t,r i ) represents the signal intensity of the body micro-motion data of the elevator rider in the echo at time t and distance r i ; n represents the total number of feature points, and its specific calculation is expressed as follows:

[0084] n = 4u + 1

[0085] In the formula, u represents the distribution quantity of feature points in different directions;

[0086] S22. Normalize the displacement data of the elevator rider, and use the non-negative matrix factorization algorithm to decompose and reconstruct the normalized data to obtain the set of potential breathing data to be extracted; among them, for the normalized displacement data z i (t) of the i-th feature point at time t, the specific calculation is expressed as follows:

[0087]

[0088] In the formula, μ(t) is the mean value; σ represents the standard deviation; C represents the translation data;

[0089] S23. Use the analysis of variance method to extract the breathing data from the set of potential breathing data to be extracted to obtain the breathing data of the current elevator rider;

[0090] Among them, using the non-negative matrix factorization algorithm for decomposition and reconstruction to obtain the set of potential breathing data to be extracted, which is specifically expressed as follows:

[0091]

[0092] In the formula, Z represents the normalized displacement data of the elevator rider, and its dimension is S * and T *respectively represent the sets of all optimal spatial pattern matrices and optimal temporal pattern matrices after decomposition and reconstruction by the non - negative matrix factorization algorithm for Q times, and are used to represent the set of potential respiratory data to be extracted, specifically as follows:

[0093]

[0094] In the formula, Q represents the total number of reconstructions; is used to represent the weight coefficient of the k - th pattern of the q - th optimal spatial pattern matrix for the contribution to the respiratory signal at all feature points, and its dimension is is used to represent the time - domain waveform of the k - th pattern of the q - th optimal temporal pattern matrix, that is, the time series, and its dimension is r represents the rank of decomposition using the non - negative matrix factorization algorithm;

[0095] Among them, for the set of potential respiratory data S * and T * in the q - th reconstruction, the optimal spatial matrix S (q) and the optimal temporal matrix T (q) The specific steps of their decomposition and reconstruction are as follows:

[0096] S221. Randomly initialize the spatial pattern matrix and the temporal pattern matrix to generate non - negative matrices S (0) and T (0) ;

[0097] S222. Update the non - negative spatial pattern matrix and the temporal pattern matrix to obtain the spatial pattern matrix and the temporal pattern matrix of the current iteration, which are specifically represented as follows:

[0098]

[0099] In the formula, S (q,m) and T (q,m) respectively represent the spatial pattern matrix and the temporal pattern matrix of the i - th iteration of the q - th reconstruction; ⊙ represents element - by - element multiplication; ∈ represents removing zero terms;

[0100] S223. Judge whether the spatial pattern matrix and the temporal pattern matrix of the current iteration meet a convergence condition; if so, obtain the spatial pattern matrix and the temporal pattern matrix with the minimum reconstruction error and use them as the optimal spatial pattern matrix and the optimal temporal pattern matrix of the q - th reconstruction; if not, continue to execute step S222;

[0101] Among them, the specific judgment of the convergence condition is as follows:

[0102]

[0103] In the formula, ε represents the threshold of the minimum reconstruction error change rate, which is used as the convergence condition, and its value range is [0, 1]; m max represents the maximum number of iterations, which is used as the convergence condition; is used to represent the reconstruction error change rate, and the reconstruction error at its m-th iteration is specifically calculated as follows:

[0104]

[0105] The specific representations of obtaining the spatial pattern matrix and the temporal pattern matrix of the minimum reconstruction error are as follows:

[0106]

[0107] In the formula, represents obtaining during all current iteration processes the spatial pattern matrix S and the temporal pattern matrix T corresponding to the minimum reconstruction error, and taking them as the optimal spatial pattern matrix and the optimal temporal pattern matrix for the q-th reconstruction.

[0108] Furthermore, the specific representation of the respiration fitting model is as follows:

[0109]

[0110] In the formula, Template j (t) is the respiration fitting model, which is used to represent the sine template of the j-th candidate frequency; F i represents the j-th discretized frequency parameter, and its specific representation is:

[0111] F j = f low + f interval × (j - 1)

[0112] In the formula, f low represents the preset lower limit of the normal respiration frequency; f interval represents the respiration frequency increment interval; N represents the total number of discretized frequency parameters, and its specific calculation is as follows:

[0113]

[0114] In the formula, f up represents the upper limit of the normal respiration frequency;

[0115] represents the offset term; P represents the number of time periods generated by the sine signal of each frequency; its specific calculation is as follows:

[0116]

[0117] In the formula, Time total is the signal duration; represents rounding up;

[0118] Among them, the specific calculation of the correlation matching is expressed as follows:

[0119]

[0120] In the formula, f pre represents the estimated respiratory rate of the current elevator rider; k represents the index number, where k ∈ [1, N]; l k represents the Euclidean distance between the jth respiratory fitting model and the respiratory signal, and its specific expression is as follows:

[0121]

[0122] In the formula, represents the time series of the most significant pattern, that is, the main respiratory signal;

[0123] Among them, the specific expression of the respiratory abnormal frequency weighting algorithm is as follows:

[0124] I(t) = w1·T low (t) + w2·T up (t) + w3·T sud (t)

[0125] In the formula, I(t) represents the abnormal respiratory rate value of the current elevator rider at the t-th moment; T low (t) represents the duration of too low respiratory rate at the t-th moment, and its specific calculation is expressed as follows:

[0126]

[0127] In the formula, Δt represents the step size of the time window; f pre (t) represents the estimated respiratory rate corresponding to the t-th moment; and T up represents the duration of too high respiratory rate at the t-th moment, and its specific calculation is expressed as follows:

[0128]

[0129] T sud (t) represents the duration of abnormal respiratory rate change at the moment t, and its specific calculation is expressed as follows:

[0130]

[0131] In the formula, Δf pre (t) represents the frequency change amount between adjacent moments, and its specific calculation is:

[0132] Δf pre (t) = |f pre (t) - f pre (t - 1)|

[0133] θ th is the frequency mutation threshold; w1, w2, and w3 respectively represent the weight coefficients of the low frequency, high frequency, and mutation frequency changes;

[0134] Among them, the specific judgment of the abnormal behavior condition is as follows:

[0135]

[0136] In the formula, I(t) > th noraml is used to indicate that the abnormal value of the breathing frequency of the current elevator rider is greater than a normal threshold th normal ; is used to indicate that the breathing signal in the current elevator is valid.

[0137] An elevator rider abnormal behavior monitoring system includes a millimeter-wave radar, an elevator rider breathing state abnormal detection device, and an alarm device;

[0138] The millimeter-wave radar is arranged inside the elevator car and is used to transmit a frequency-modulated continuous wave signal, that is, FMCW, inside the elevator car. After the FMCW encounters the body of the elevator rider, it is reflected to generate a millimeter-wave signal of the body micro-motion of the elevator rider, and the millimeter-wave signal is transmitted to the elevator rider breathing state abnormal detection device wirelessly or by wire;

[0139] The elevator rider breathing state abnormal detection device is used to receive the millimeter-wave signal of the body micro-motion of the elevator rider, analyze the breathing state, and judge whether the elevator rider is in an abnormal state according to the analysis result: if so, send an alarm signal to the alarm device; if not, continue to receive the millimeter-wave signal of the millimeter-wave radar and analyze the breathing state;

[0140] The alarm device is arranged inside the elevator car and is used to receive the alarm signal of the elevator rider breathing state abnormal detection device and issue an audible and visual alarm to remind the elevator rider that they are currently in an abnormal state, and send an alarm message to the property management center or the rescue department through a wired or wireless network;

[0141] Among them, the elevator rider breathing state abnormal detection device is the elevator rider breathing state abnormal detection device described above.

[0142] For better understanding and implementation, the present invention will be described in detail below with reference to the accompanying drawings. Description of the Drawings

[0143] Figure 1 It is a schematic diagram of the simple setting structure of the abnormal behavior monitoring system for elevator passengers;

[0144] Figure 2 It is a schematic diagram of the simple structure of the breathing state device for elevator passengers according to the present invention;

[0145] Figure 3 It is a schematic diagram of the simple process of the abnormal breathing state detection method for elevator passengers based on millimeter waves according to the present invention;

[0146] Figure 4 It is a schematic diagram of the simple structure for setting feature points in the millimeter wave micro - motion feature algorithm. Specific implementation manners

[0147] In order to solve the problems of high risk of privacy data leakage and poor environmental robustness in the prior art, the present invention obtains the body micro - motion data of elevator passengers, and uses a millimeter wave micro - motion feature algorithm to extract breathing features from the data to obtain the breathing data of the current elevator passengers; then, uses a breathing fitting model to perform correlation matching on the breathing data, extracts the estimated breathing frequency value, and statistically processes the estimated breathing frequency value according to a breathing abnormal frequency weighting algorithm to obtain the breathing frequency abnormal value of the elevator passengers; finally, determines whether the breathing frequency abnormal value meets the abnormal behavior condition: if so, the current elevator passenger is in an abnormal state and an alarm message is sent; if not, continues to obtain the body micro - motion data of the elevator passengers for detection. Accordingly, the present invention uses a millimeter wave radar to obtain the breathing state of elevator passengers to effectively avoid the risk of privacy leakage of elevator passengers and enhance environmental adaptability; at the same time, by combining the millimeter wave micro - motion feature algorithm with the breathing fitting model, the judgment accuracy of breathing frequency abnormality is improved, the possibility of false alarms and missed alarms is reduced, and thus more accurate detection of abnormal breathing states is achieved.

[0148] Based on the above design, the present invention proposes a method for detecting abnormal breathing states of elevator passengers based on millimeter waves, and proposes an abnormal breathing state detection device for elevator passengers based on this method.

[0149] Please refer to Figure 1 , Figure 1 It is a schematic diagram of the simple setting structure of the abnormal behavior monitoring system for elevator passengers.

[0150] An abnormal behavior monitoring system for elevator passengers includes a millimeter wave radar 100, an abnormal breathing state detection device 101 for elevator passengers, and an alarm device 102.

[0151] The millimeter-wave radar 100 is disposed inside the elevator car and is used to transmit a Frequency Modulated Continuous Wave (FMCW) signal inside the elevator car. After the FMCW signal is reflected by the body of the elevator rider, a millimeter-wave signal of the body micro-motion of the elevator rider is generated, and the millimeter-wave signal is transmitted to the elevator rider's abnormal breathing state detection device 101 wirelessly or through a wire; wherein, the inside of the elevator car can be centered at the top to ensure the maximum transmission range of the millimeter wave.

[0152] The elevator rider's abnormal breathing state detection device 101 is used to receive the millimeter-wave signal of the body micro-motion of the elevator rider, perform breathing state analysis, and judge whether the elevator rider is in an abnormal state according to the analysis result: if so, send an alarm signal to the alarm device; if not, continue to receive the millimeter-wave signal of the millimeter-wave radar and perform breathing state analysis.

[0153] The alarm device 102 is disposed inside the elevator car and is used to receive the alarm signal from the elevator rider's abnormal breathing state detection device 101, and issue an audible and visual alarm to remind the elevator rider that the current state is abnormal, and send an alarm message to the property management center or the rescue department through a wired or wireless network; wherein, the alarm message includes the position where the current elevator car is located and the abnormal state information of the elevator rider.

[0154] Please also refer to Figure 2 and Figure 3 , Figure 2 which is a schematic diagram of the simple structure of the elevator rider's breathing state device described in the present invention, Figure 3 and which is a schematic diagram of the simple process of the method for detecting abnormal breathing state of elevator riders based on millimeter waves described in the present invention.

[0155] The elevator rider's abnormal breathing state detection device 101 includes a body micro-motion data acquisition unit 1, a breathing feature extraction unit 2, a breathing frequency matching unit 3, a breathing frequency abnormal value extraction unit 4, and a behavior abnormal judgment unit 5.

[0156] The body micro-motion data acquisition unit 1 is used to execute step S1: acquire the body micro-motion data of the elevator rider.

[0157] When passengers are in the waiting state for the elevator, their behavior patterns inside the elevator car are relatively stable. Usually, they just stand or adjust their body postures within a short period. Therefore, millimeter-wave radar, that is, millimeter waves, can very effectively capture the micro-motion signals related to the breathing of passengers. In addition, inside the elevator car, the movement of passengers is relatively restricted. Therefore, it can be assumed that most of the changes come from the physiological micro-motions of breathing rather than large-scale movements or displacements. Thus, the micro-motion data belonging to breathing can be further stripped from the millimeter-wave signals.

[0158] Please refer to Figure 4 , Figure 4 is a schematic diagram of a simple structure for setting feature points in the millimeter-wave micro-motion feature algorithm.

[0159] The respiration feature extraction unit 2 is used to execute step S2: adopt a millimeter-wave micro-motion feature algorithm to extract the respiration features from the body micro-motion data of the passengers in the elevator, and obtain the respiration data of the current passengers in the elevator.

[0160] Specifically, the millimeter-wave micro-motion feature algorithm includes the following steps:

[0161] S21. Construct feature points based on the body micro-motion data of the current passengers in the elevator, and calculate their offsets according to the feature points to obtain the displacement data of the passengers in the elevator. The displacement data of the i-th feature point at time t is specifically calculated as follows:

[0162]

[0163] In the formula, λ represents the wavelength of the millimeter waves emitted by the millimeter-wave radar, and the specific calculation is: λ = c / f, where c is the speed of light and f is the center frequency of the millimeter-wave radar, which are known data; Δp(t, r i ) represents the signal intensity of the body micro-motion data of the passengers in the echo at the t-th moment and at a distance of r i , which is used to characterize the signal change caused by the body micro-motion of the passengers in the elevator; n represents the total number of feature points, and its specific calculation is as follows:

[0164] n = 4u + 1

[0165] In the formula, u represents the distribution quantity of feature points in different directions. In the present invention, it is defaulted to 2; among them, in order to better extract the micro-motion data belonging to breathing, the present invention selects the point with the maximum received signal intensity of the millimeter radar, that is, i = 1, as the central feature point to characterize the position of the body of the passengers in the elevator, and expands the feature points around this center to form a cross-shaped distribution, so as to accurately capture the micro-motion information of the passengers in the elevator through the small changes in the reflection signal intensity of the millimeter-wave radar.

[0166] S22. Normalize the displacement data of the elevator passengers, and use the Non - negative Matrix Factorization (NMF) algorithm to decompose and reconstruct the normalized data to obtain the set of potential respiration data to be extracted.

[0167] Specifically, the decomposition and reconstruction of the normalized data using the non - negative matrix factorization algorithm are specifically represented as follows:

[0168]

[0169] In the formula, Z represents the normalized displacement data of the elevator passengers, and its dimension is For the normalized displacement data z i (t) of the i - th feature point at the t - th moment, the specific calculation is represented as follows:

[0170]

[0171] In the formula, μ(t) is the mean value, which is used to represent the mean value of the displacement data of all feature points at the current moment t; σ represents the standard deviation; C represents the translation data, which is a constant and is used to ensure that the elements in the current normalized displacement data of the elevator passengers do not contain negative numbers. Its default value is C = max i,t (z i (t)+∈), ∈>0. According to different requirements, different values can be selected. The present invention does not specifically limit the selection of the translation data here;

[0172] S * and T * respectively represent the sets of all optimal spatial pattern matrices and optimal temporal pattern matrices after Q - times of NMF decomposition and reconstruction, and are used to represent the set of potential respiration data to be extracted. The specific representation is as follows:

[0173]

[0174] In the formula, Q represents the total number of reconstructions, and the default value is 20. In order to avoid local optimality, different total numbers of reconstructions can be adjusted to iteratively obtain the best spatial pattern matrix and temporal pattern matrix. The present invention does not specifically limit the selection of the total number of reconstructions;

[0175] is used to represent the weight coefficient of the k - th mode of the q - th optimal spatial pattern matrix in the contribution to the respiration signal at all feature points, and its dimension is is used to represent the time - domain waveform, that is, the time series, of the k - th mode of the q - th optimal temporal pattern matrix, and its dimension is r represents the rank of the NMF decomposition, and the default value is 1;

[0176] Among them, for the set S of potential respiratory data to be extracted * and T * the optimal spatial matrix S (q) and the optimal temporal matrix T (q) for the q-th reconstruction, the specific steps of their decomposition and reconstruction are as follows:

[0177] S221. Randomly initialize the spatial pattern matrix and the temporal pattern matrix to generate a non-negative spatial pattern matrix S (0) and the temporal pattern matrix T (0) ;

[0178] S222. Update the non-negative spatial pattern matrix and the temporal pattern matrix to obtain the spatial pattern matrix and the temporal pattern matrix for the current iteration, which are specifically expressed as follows:

[0179]

[0180] where S (q,m) and T (q,m) respectively represent the spatial pattern matrix and the temporal pattern matrix for the m-th iteration of the q-th reconstruction; ⊙ represents element-wise multiplication; ∈ represents removing zero terms, and is a constant used to prevent division by zero;

[0181] S223. Determine whether the spatial pattern matrix and the temporal pattern matrix for the current iteration meet a convergence condition; if so, obtain the spatial pattern matrix and the temporal pattern matrix with the minimum reconstruction error and use them as the optimal spatial pattern matrix and the optimal temporal pattern matrix for the q-th reconstruction; if not, continue to execute step S222;

[0182] Among them, the specific determination of the convergence condition is expressed as follows:

[0183]

[0184] where ε represents the threshold of the minimum reconstruction error change rate, which is used as the convergence condition, and its value range is [0, 1]; m max

[0185] represents the maximum number of iterations, which is used as the convergence condition; is used to represent the reconstruction error change rate, and the reconstruction error for its m-th iteration is specifically calculated as follows:

[0186]

[0187] The specific expression for obtaining the spatial pattern matrix and the temporal pattern matrix with the minimum reconstruction error is as follows:

[0188]

[0189] In the formula, represents obtaining, during all current iterative processes, the spatial pattern matrix S and the temporal pattern matrix T corresponding to the minimum reconstruction error, and taking them as the optimal spatial pattern matrix and the optimal temporal pattern matrix for the q-th reconstruction.

[0190] S23. Use the analysis of variance (ANOVA) method to extract respiratory data from the set of potential respiratory data to be extracted, and obtain the respiratory data of the current elevator rider.

[0191] Specifically, the specific representation of the respiratory data of the current elevator rider is as follows:

[0192] T (x) = ANOVA(S * , T * )

[0193] In the formula, T (x) represents the respiratory data of the current elevator rider, and is used to represent the optimal temporal pattern matrix T of the x-th reconstruction with the most significant difference in the set of potential respiratory data to be extracted. (x) Since the respiratory data is related to the temporal pattern, in this invention, the temporal pattern matrix is generally taken as the respiratory data of the current elevator rider;

[0194] ANOVA(·) represents the analysis of variance, and is used to perform joint statistical feature calculation on the input set of potential respiratory data S * , T * to calculate the ratio of the between-group variance to the within-group variance according to the joint statistical features, and screen out the set of potential respiratory data S (x) , T (x) that is significantly different from the critical value, and extract the corresponding temporal pattern matrix T (x) , and take it as the respiratory data of the current elevator rider;

[0195] Among them, since the analysis of variance is a conventional method, it can be specifically implemented using other tools such as python or matlab, and it is not the invention creation of this invention, so the implementation details of the analysis of variance are not specifically described in this invention.

[0196] Accordingly, this invention obtains the body micro-motion data of the elevator rider by using a millimeter-wave radar, and based on the analysis of the feature points and the signal strength, combines the non-negative matrix factorization (NMF) algorithm for multiple reconstructions, and in the set of potential respiratory data obtained from the multiple reconstructions, uses the analysis of variance (ANOVA) to extract the temporal pattern T that is potentially related to the respiratory frequency. (x), which avoids the influence of possible environmental data in millimeter waves and significantly improves the extraction accuracy of the breathing rate.

[0197] It should be noted that in the present invention, the breathing data T of the current elevator rider (x) corresponding to the first mode, that is, is used as the main breathing signal.

[0198] The breathing frequency matching unit 3 is used to execute step S3: perform correlation matching on the breathing data of the current elevator rider by using a breathing fitting model to obtain an estimated value of the breathing frequency of the current elevator rider.

[0199] Specifically, the breathing fitting model represents a normal breathing frequency fitting template signal, and its specific representation is as follows:

[0200]

[0201] In the formula, Template j (t) represents the j-th breathing frequency fitting template signal at the t-th moment, and is used to represent the sine template of the j-th candidate frequency; F i represents the j-th discretized frequency parameter, and is used to generate sine template signals of different frequencies; represents the offset term, and is used to represent the starting time offset of each time period; among them, F j = f low + f interval ×(j - 1), f low represents the preset lower limit of the normal breathing frequency, which is defaulted to 0.2 hz in the present invention; f interval represents the breathing frequency increment interval, which is preset to 0.01 in the present invention, and N represents the total number of discretized frequency parameters, and its specific calculation is as follows:

[0202]

[0203] In the formula, f up represents the upper limit of the normal breathing frequency, which is defaulted to 0.4 hz in the present invention;

[0204] P represents the number of time periods for generating the sine signal of each frequency, and is used to enhance the matching ability of the breathing fitting model to the periodicity of the breathing signal, and its specific calculation is as follows:

[0205]

[0206] In the formula, Time total is the signal duration, and is used to characterize the signal acquisition duration of the body micro-motion data of the elevator rider; represents rounding up.

[0207] The specific calculation of the correlation matching is as follows:

[0208]

[0209] In the formula, f pre represents the estimated value of the breathing frequency of the current elevator rider; k represents the index number, which is used to represent the index corresponding to the minimum Euclidean distance selected from all candidate frequency indices j, and k ∈ [1, N]; l j represents the Euclidean distance between the j-th breathing fitting model and the breathing signal, and its specific representation is as follows:

[0210]

[0211] In the formula, represents the time series with the most significant pattern, that is, the main breathing signal;

[0212] In the present invention, a candidate sine template, that is, a breathing fitting model, is constructed by presetting to conform to the physiological breathing frequency range. After being decomposed and reconstructed by NMF multiple times, the optimal time pattern obtained by variance analysis, that is, the main breathing signal is matched with the template signal in the time domain, and the data belonging to the breathing frequency characteristics in the signal is extracted as the breathing frequency estimated value. Accordingly, the present invention uses template matching and utilizes the global similarity characteristics of periodic signals to avoid the phenomenon of noise sensitivity caused by directly performing time-domain waveform analysis, and significantly improves the anti-interference ability and reliability of the breathing frequency under potential environmental interference in the elevator.

[0213] The breathing frequency outlier extraction unit 4 is used to execute step S4: calculate the breathing frequency estimated value of the current elevator rider by using a breathing abnormal frequency weighting algorithm to obtain the breathing frequency outlier of the current elevator rider.

[0214] Specifically, the breathing abnormal frequency weighting algorithm is used to compare the current breathing frequency estimated value with the preset normal breathing frequency range, so as to calculate the outlier and prevent misjudgment. The specific calculation is as follows:

[0215] I(t) = w1·T low (t) + w2·T up (t) + w3·T sud (t)

[0216] In the formula, I(t) represents the breathing frequency outlier of the current elevator rider at the t-th moment; T low (t) represents the duration of too low breathing frequency at the t-th moment, and its specific calculation is as follows:

[0217]

[0218] where Δt represents the step size of the time window, which is used to represent the sampling interval; f pre (t) represents the estimated value of the breathing rate corresponding to the t-th moment; and T up represents the duration of the excessive breathing rate at the t-th moment, and its specific calculation is as follows:

[0219]

[0220] T sud (t) represents the duration of the abnormal change in the breathing rate at the moment t, that is, a sharp decrease or increase, and its specific calculation is as follows:

[0221]

[0222] where Δf pre (t) represents the frequency change amount between adjacent moments, and its specific calculation is Δf pre (t) = |f pre (t) - f pre (t - 1)|, θ th is the frequency mutation threshold, which is defaulted to 0.1; w1, w2, and w3 respectively represent the weight coefficients of the low frequency, high frequency, and mutation frequency changes, which are defaulted to 0.5, 0.5, and 1, and are used to characterize the influence degree of each abnormal state on the total abnormal value, and can be adjusted according to the actual situation or dynamically adjusted. The present invention does not make specific limitations on it.

[0223] Accordingly, the present invention calculates the duration of the abnormal state of the breathing rate (low frequency, high frequency, mutation) by accumulation, and fuses them with weights into a comprehensive abnormal value to avoid misjudgment caused by instantaneous noise interference.

[0224] The behavior abnormality judgment unit 5 is used to execute step S5: judge whether the abnormal value of the breathing rate of the current elevator rider meets a behavior abnormality condition: if so, send an alarm message; if not, continue to execute step S1.

[0225] Specifically, the specific judgment of the behavior abnormality condition is as follows:

[0226]

[0227] where I(t) > th noraml is used to represent that the abnormal value of the breathing rate of the current elevator rider is greater than a normal threshold th normal ; It is used to indicate that the breathing signal in the current elevator is valid, that is, it is determined that there are passengers in the current elevator. Among them, the abnormal operation of the elevator is indicated by the elevator door not opening and the floor selection button not being triggered, that is, it can be considered that there is a passenger in a coma in the elevator, which belongs to abnormal riding behavior. At the same time, the specific conditions for abnormal operation can be adjusted by referring to the fault states defined in the elevator safety standard, and its actual specific conditions can be expanded or adjusted according to requirements. The present invention does not specifically limit this here.

[0228] Accordingly, the present invention combines the abnormal breathing value with the elevator operation state to ensure that the alarm rule covers various abnormal behavior scenarios.

[0229] Compared with the prior art, the present invention realizes high-precision anti-noise estimation of the breathing frequency through a breathing fitting model, that is, through the minimization of the Euclidean distance based on the discretized sine template, effectively improving the detection robustness of weak breathing signals in complex environments. In addition, through the non-negative matrix factorization (NMF) algorithm, the normalized displacement data is decomposed into low rank, and the potential breathing signal is extracted from the time pattern, so that subsequent variance analysis is used to directly extract the time pattern with significant differences as the most significant breathing signal, thereby improving the accuracy of breathing signal extraction. And, by combining the continuous abnormal accumulation with the duration of the abnormal breathing frequency of instantaneous mutation, a multi-dimensional breathing risk quantification assessment is realized to improve the clinical adaptability and scenario generalization of breathing state assessment.

[0230] Based on the same inventive concept, the present application also provides an electronic device, which can be a server, a desktop computing device or a mobile computing device (such as a laptop computing device, a handheld computing device, a tablet computer, a netbook, etc.) and other terminal devices. The device includes one or more processors and a memory, wherein the processor is used to execute a program to implement the method for detecting abnormal breathing state of passengers in the elevator based on millimeter waves in the embodiments of the present invention; the memory is used to store a computer program executable by the processor.

[0231] Based on the same inventive concept, the present application also provides a computer-readable storage medium, corresponding to the embodiments of the method for detecting abnormal breathing state of passengers in the elevator based on millimeter waves described above. The computer-readable storage medium stores a computer program thereon, and when the program is executed by a processor, it realizes the steps of the method for detecting abnormal breathing state of passengers in the elevator based on millimeter waves recorded in any of the above embodiments.

[0232] The present application may take the form of a computer program product implemented on one or more storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) that contain program code. Computer-usable storage media include both permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information may be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include but are not limited to: phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0233] The above-described embodiments merely represent several implementation manners of the present invention, and the description thereof is relatively specific and detailed. However, it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and the present invention also intends to include these changes and modifications.

Claims

1. A method for detecting abnormal breathing status of elevator passengers based on millimeter waves, characterized in that: The following steps are involved: S1. Obtaining body micro-motion data of passengers; wherein the body micro-motion data of passengers is collected by a millimeter wave radar; S2. Using a millimeter wave micro-motion feature algorithm to extract breathing features from the body micro-motion data of the elevator passengers, and obtain the breathing data of the current elevator passengers; S3, using a breathing fitting model to perform correlation matching on the breathing data of the current elevator passenger to obtain an estimated value of the breathing frequency of the current elevator passenger; S4, using a breathing abnormality frequency weighted algorithm to calculate the breathing frequency estimation value of the current elevator passenger, and obtain the breathing frequency abnormality value of the current elevator passenger; S5. Determine whether the abnormal breathing rate value of the current elevator passenger meets an abnormal behavior condition: if so, send an alarm message; if not, continue to execute step S1.

2. The method for detecting abnormal breathing state of elevator passengers according to claim 1, characterized in that: The millimeter wave micro-motion feature algorithm comprises the following steps: S21, construct feature points based on the current body micro-motion data of the elevator passenger, and calculate its offset based on the feature points to obtain the displacement data of the elevator passenger; wherein, for the displacement data of the i-th feature point at time t The specific calculation is as follows: Where λ represents the wavelength of the millimeter wave emitted by the millimeter wave radar; Δp(t,r i ) means at time t, the distance is r i The signal strength of the body micro-motion data of the elevator passengers at the echo; n represents the total number of feature points, and its specific calculation is as follows: n=4u+1 In the formula, u represents the distribution number of feature points in different directions; S22, normalize the displacement data of the passengers, and use the non-negative matrix decomposition algorithm to decompose and reconstruct the normalized data to obtain the potential breathing data set to be extracted; wherein, for the normalized displacement data z of the i-th feature point at the t-th moment i (t) The specific calculation is as follows: In the formula, μ(t) is the mean; σ is the standard deviation; C is the translation data; S23. Extract breathing data from the potential breathing data set to be extracted using variance analysis to obtain breathing data of the current passengers.

3. The method for detecting abnormal breathing state of elevator passengers according to claim 2, characterized in that: The non-negative matrix decomposition algorithm is used to decompose and reconstruct to obtain a potential respiratory data set to be extracted, which is specifically expressed as follows: In the formula, Z represents the normalized displacement data of the elevator passengers, and its dimension is S * and T * They represent the set of all optimal spatial pattern matrices and optimal temporal pattern matrices after decomposition and reconstruction by the Q-time non-negative matrix factorization algorithm, and are used to represent the potential respiratory data set to be extracted. Specifically, they are expressed as follows: In the formula, q represents the total number of reconstructions; The weight coefficient used to represent the contribution of the kth mode of the qth optimal spatial mode matrix to the respiratory signal at all feature points is The k-th mode time domain waveform, i.e., time series, used to represent the q-th optimal time mode matrix has the dimension r represents the rank of decomposition using the non-negative matrix factorization algorithm; Among them, for the potential respiratory data set S to be extracted * and T * The optimal spatial matrix S for the qth reconstruction (q) and the optimal time matrix T (q) The specific steps of decomposition and reconstruction are as follows: S221, randomly initialize the spatial pattern matrix and the temporal pattern matrix to generate a non-negative matrix S (0) and T (0) ; S222, updating the non-negative spatial mode matrix and the temporal mode matrix to obtain the spatial mode matrix and the temporal mode matrix of the current iteration, which are specifically expressed as follows: In the formula, S (q,m) and T (q,m) They represent the m-th iterative spatial pattern matrix and temporal pattern matrix of the q-th reconstruction respectively; ⊙ represents element-by-element multiplication; ∈ represents zero removal; S223, determine whether the spatial mode matrix and the temporal mode matrix of the current iteration meet a convergence condition; if so, obtain the spatial mode matrix and the temporal mode matrix with the minimum reconstruction error, and use them as the optimal spatial mode matrix and the optimal temporal mode matrix of the qth reconstruction; if not, continue to execute step S222; The specific judgment of the convergence condition is as follows: Where ε represents the minimum reconstruction error change rate threshold, which is used as a convergence condition and its value range is [0,1]; m max Indicates the maximum number of iterations, used as a convergence condition; It is used to represent the rate of change of reconstruction error. The reconstruction error of the mth iteration is The specific calculation is as follows: The specific representation of the spatial mode matrix and the temporal mode matrix for obtaining the minimum reconstruction error is as follows: In the formula, Indicates getting all current iterations The spatial pattern matrix S and the temporal pattern matrix T corresponding to the minimum reconstruction error are used as the optimal spatial pattern matrix and the optimal temporal pattern matrix for the qth reconstruction.

4. The method for detecting abnormal breathing state of elevator passengers according to claim 3, characterized in that: The specific expression of the breathing fitting model is as follows: In the formula, Template j (t) is the breathing fitting model, which is used to represent the sinusoidal template of the jth candidate frequency; F i represents the jth discretized frequency parameter, which is specifically expressed as: F j =f low +f interval ×(j-1) In the formula, f low Indicates the preset lower limit of normal respiratory rate; f interval Represents the breathing frequency increment interval; N represents the total number of discretized frequency parameters, and its specific calculation is as follows: In the formula, f up Indicates the upper limit of normal respiratory rate; represents the offset term; P represents the number of time periods for generating a sinusoidal signal of each frequency; its specific calculation is as follows: Where Time total is the signal duration; Indicates rounding up; The specific calculation of the correlation matching is as follows: In the formula, f pre represents the estimated breathing rate of the current passenger; k represents the index number, where k∈[1,N]; l j represents the Euclidean distance between the jth respiratory fitting model and the respiratory signal, which is specifically expressed as follows: In the formula, The time series representing the most significant mode, i.e. the main breathing signal.

5. The method for detecting abnormal breathing state of elevator passengers according to claim 4, characterized in that: The specific expression of the abnormal breathing frequency weighting algorithm is as follows: I(t)=w1·T low (t)+w2·T up (t)+w3·T sud (t) Where I(t) represents the abnormal breathing rate of the current passenger at the tth moment; T low (t) represents the duration of low respiratory rate at time t, and its specific calculation is as follows: Where Δt represents the step size of the time window; f pre (t) represents the estimated value of respiratory frequency corresponding to the tth moment; and T up It represents the duration of excessively high respiratory rate at time t, and its specific calculation is as follows: T sud (t) represents the duration of abnormal respiratory rate change at time t, and its specific calculation is as follows: Where Δf pre (t) represents the frequency change at adjacent moments, which is specifically calculated as: Δf pre (t)=|f pre (t)-fp pre (t-1)| θ th is the frequency mutation threshold; w1, w2, w3 represent the weight coefficients of too low frequency, too high frequency and mutation frequency change respectively.

6. The method for detecting abnormal breathing state of elevator passengers according to claim 5, characterized in that: The specific judgment of the abnormal behavior condition is as follows: Where, I(t)>th noraml It is used to indicate that the current abnormal value of the breathing rate of the elevator passengers is greater than a normal threshold value th normal ; Used to indicate that the breathing signal in the current elevator is valid.

7. A device for detecting abnormal breathing state of elevator passengers, characterized in that: It includes a body micro-movement data acquisition unit, a breathing feature extraction unit, a breathing frequency matching unit, a breathing frequency abnormal value extraction unit and a behavior abnormality judgment unit; The body micro-motion data acquisition unit is used to acquire body micro-motion data of passengers; wherein the body micro-motion data of passengers are collected by a millimeter wave radar; The breathing feature extraction unit is used to extract breathing features from the body micro-motion data of the elevator passengers using a millimeter wave micro-motion feature algorithm to obtain the breathing data of the current elevator passengers; The breathing frequency matching unit is used to use a breathing fitting model to perform correlation matching on the breathing data of the current elevator passenger to obtain an estimated value of the breathing frequency of the current elevator passenger; The abnormal breathing frequency value extraction unit is used to calculate the estimated value of the breathing frequency of the current passenger by using an abnormal breathing frequency weighting algorithm to obtain the abnormal breathing frequency value of the current passenger; The abnormal behavior judgment unit is used to judge whether the abnormal breathing rate value of the current elevator passenger meets an abnormal behavior condition: if so, an alarm message is sent; if not, the body micro-movement data acquisition unit is continued to be called.

8. The device for detecting abnormal breathing state of elevator passengers according to claim 7, characterized in that: The millimeter wave micro-motion feature algorithm comprises the following steps: S21, construct feature points based on the current body micro-motion data of the elevator passenger, and calculate its offset based on the feature points to obtain the displacement data of the elevator passenger; wherein, for the displacement data of the i-th feature point at time t The specific calculation is as follows: Where λ represents the wavelength of the millimeter wave emitted by the millimeter wave radar; Δp(t,r i ) means at time t, the distance is r i The signal strength of the body micro-motion data of the elevator passengers at the echo; n represents the total number of feature points, and its specific calculation is as follows: n=4u+1 In the formula, u represents the distribution number of feature points in different directions; S22, normalize the displacement data of the passengers, and use the non-negative matrix decomposition algorithm to decompose and reconstruct the normalized data to obtain the potential breathing data set to be extracted; wherein, for the normalized displacement data z of the i-th feature point at the t-th moment i (t) The specific calculation is as follows: In the formula, μ(t) is the mean; σ is the standard deviation; C is the translation data; S23, extracting respiratory data from the potential respiratory data set to be extracted using variance analysis to obtain respiratory data of the current passenger; The non-negative matrix decomposition algorithm is used for decomposition and reconstruction to obtain a potential respiratory data set to be extracted, which is specifically expressed as follows: In the formula, Z represents the normalized displacement data of the elevator passengers, and its dimension is S * and T * They represent the set of all optimal spatial pattern matrices and optimal temporal pattern matrices after decomposition and reconstruction by the Q-time non-negative matrix factorization algorithm, and are used to represent the potential respiratory data set to be extracted. Specifically, they are expressed as follows: In the formula, q represents the total number of reconstructions; The weight coefficient used to represent the contribution of the kth mode of the qth optimal spatial mode matrix to the respiratory signal at all feature points is The k-th mode time domain waveform, i.e., time series, used to represent the q-th optimal time mode matrix has the dimension r represents the rank of decomposition using the non-negative matrix factorization algorithm; Among them, for the potential respiratory data set S to be extracted * and T * The optimal spatial matrix S for the qth reconstruction (q) and the optimal time matrix T (q) The specific steps of decomposition and reconstruction are as follows: S221, randomly initialize the spatial pattern matrix and the temporal pattern matrix to generate a non-negative matrix S (0) and T (0) ; S222, updating the non-negative spatial mode matrix and the temporal mode matrix to obtain the spatial mode matrix and the temporal mode matrix of the current iteration, which are specifically expressed as follows: In the formula, S (q,m) and T (q,m) They represent the m-th iterative spatial pattern matrix and temporal pattern matrix of the q-th reconstruction respectively; ⊙ represents element-by-element multiplication; ∈ represents zero removal; S223, determine whether the spatial mode matrix and the temporal mode matrix of the current iteration meet a convergence condition; if so, obtain the spatial mode matrix and the temporal mode matrix with the minimum reconstruction error, and use them as the optimal spatial mode matrix and the optimal temporal mode matrix of the qth reconstruction; if not, continue to execute step S222; The specific judgment of the convergence condition is as follows: Where ε represents the minimum reconstruction error change rate threshold, which is used as a convergence condition and its value range is [0,1]; m max Indicates the maximum number of iterations, used as a convergence condition; It is used to represent the rate of change of reconstruction error. The reconstruction error of the mth iteration is The specific calculation is as follows: The specific representation of the spatial mode matrix and the temporal mode matrix for obtaining the minimum reconstruction error is as follows: In the formula, Indicates getting all current iterations The spatial pattern matrix S and the temporal pattern matrix T corresponding to the minimum reconstruction error are used as the optimal spatial pattern matrix and the optimal temporal pattern matrix for the qth reconstruction.

9. The device for detecting abnormal breathing state of elevator passengers according to claim 8, characterized in that: The specific expression of the breathing fitting model is as follows: In the formula, Template j (t) is the breathing fitting model, which is used to represent the sinusoidal template of the jth candidate frequency; F i represents the jth discretized frequency parameter, which is specifically expressed as: F j =f low +f interval ×(j-1) In the formula, f low Indicates the preset lower limit of normal respiratory rate; f interval Represents the breathing frequency increment interval; N represents the total number of discretized frequency parameters, and its specific calculation is as follows: In the formula, f up Indicates the upper limit of normal respiratory rate; represents the offset term; P represents the number of time periods for generating a sinusoidal signal of each frequency; its specific calculation is as follows: Where Time total is the signal duration; Indicates rounding up; The specific calculation of the correlation matching is as follows: In the formula, f pre represents the estimated breathing rate of the current passenger; k represents the index number, where k∈[1,N]; l j represents the Euclidean distance between the jth respiratory fitting model and the respiratory signal, which is specifically expressed as follows: In the formula, The time series representing the most significant pattern, i.e. the main respiratory signal; The specific expression of the abnormal breathing frequency weighting algorithm is as follows: I(t)=w1·T low (t)+w2·T up (t)+w3·T sud (t) Where I(t) represents the abnormal breathing rate of the current passenger at the tth moment; T low (t) represents the duration of low respiratory rate at time t, and its specific calculation is as follows: Where Δt represents the step size of the time window; f pre (t) represents the estimated value of respiratory frequency corresponding to the tth moment; and T up It represents the duration of excessively high respiratory rate at time t, and its specific calculation is as follows: T sud (t) represents the duration of abnormal respiratory rate change at time t, and its specific calculation is as follows: Where Δf pre (t) represents the frequency change at adjacent moments, which is specifically calculated as: Δf pre (t)=|f pre (t)-f pre (t-1)| θ th is the frequency mutation threshold; w1, w2, w3 represent the weight coefficients of too low frequency, too high frequency and mutation frequency change respectively; The specific judgment of the abnormal behavior condition is as follows: Where, I(t)>th niraml It is used to indicate that the current abnormal value of the breathing rate of the elevator passengers is greater than a normal threshold value th normal ; Used to indicate that the breathing signal in the current elevator is valid.

10. A system for monitoring abnormal behavior of elevator passengers, characterized in that: It includes millimeter wave radar, abnormal breathing state detection device for passengers and alarm device; The millimeter wave radar is arranged inside the elevator car, and is used to transmit a frequency modulated continuous wave signal, i.e., FMCW, to the inside of the elevator car. The FMCW is reflected upon encountering the body of the elevator passenger, and generates a millimeter wave signal of the elevator passenger's body micro-movement, and transmits the millimeter wave signal to the elevator passenger's abnormal breathing state detection device wirelessly or wiredly; The abnormal breathing state detection device for elevator passengers is used to receive the millimeter wave signal of the elevator passenger's body micro-movement, and perform breathing state analysis, and determine whether the elevator passenger is in an abnormal state according to the analysis result: if so, send an alarm signal to the alarm device; if not, continue to receive the millimeter wave signal of the millimeter wave radar to perform breathing state analysis; The alarm device is arranged inside the elevator car, and is used to receive the alarm signal of the abnormal breathing state detection device of the elevator passengers, and send out an audible and visual alarm to remind the elevator passengers that they are currently in an abnormal state, and send an alarm message to the property management center or the rescue department through a wired or wireless network; Wherein, the abnormal breathing state detection device for elevator passengers is the abnormal breathing state detection device for elevator passengers as described in any one of the preceding claims 7-9.

Images