Non-contact human respiration data real-time measurement method oriented to Wi-Fi complex interference environment

Abstract

The invention discloses a non-contact real-time monitoring method for human respiration in a Wi-Fi (Wireless Fidelity) complex interference environment. And analyzing the CSI interference intensity and the CSI active ratio, constructing a Wi-Fi interference feature mapping matrix, and calculating each channel interference index by using the matrix to realize interference discrimination. Then, through a dynamic subcarrier selection algorithm CSI-DSSA based on an interference index, a subcarrier combination with the weakest cross correlation in interference data is selected for interference processing, and time domain feature information of multiple data streams in data which is not interfered in a multi-link data fusion method aggregation mode is analyzed. And finally, extracting a time domain characteristic value and constructing an SVM multi-activity classification model in a Wi-Fi interference environment to obtain a tumble activity identification result. According to the method, the human respiratory activity recognition accuracy in the Wi-Fi interference environment can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent devices, and in particular to a non-contact real-time measurement method and system for human respiratory parameters in a Wi-Fi interference environment. Background Art

[0002] With the increasing popularity of wireless networks, contactless identification technology represented by wireless sensing technology has become a hot research topic in the field of ubiquitous computing. Wi-Fi-based behavior recognition technology usually includes two types of received signals: Channel State Information (CSI) and Radio Signal Strength Index (RSSI). RSSI is susceptible to narrowband and multipath interference, has low recognition accuracy, and limited performance. In contrast, CSI can present the amplitude and phase of multipath propagation at different frequencies, thereby providing richer and more stable channel parameters. Therefore, CSI is more suitable for human behavior recognition than RSSI.

[0003] Existing human detection methods mainly rely on video analysis and wearable sensors. The former faces the problem of privacy and security invasion and is easily affected by light, which poses the risk of user privacy leakage, while the latter is easily forgotten by users and has a certain hardware cost that limits its popularization and application. Wi-Fi signals are insensitive to light, low cost, easy to install and protect privacy, so they are more practical than the above methods.

[0004] However, most of the current CSI detection models only detect a single item (such as sitting breathing or supine breathing), and a comprehensive detection model that can distinguish between sitting breathing, supine breathing, prone breathing, and side breathing has not yet been developed. In addition, the existing models can usually only do classification, but cannot combine the indoor environment data after obtaining the classification results to infer the user's next behavior habits, thereby providing "automated task suggestions and other online service functions" for smart homes, further improving the user experience and quality of life. Summary of the invention

[0005] In order to overcome the shortcomings of existing breathing detection methods, such as lack of comfort, potential privacy and security risks, and high price, the present invention provides a non-contact real-time measurement method for human breathing parameters in a Wi-Fi interference environment. By merging multiple Wi-Fi channels to expand the signal bandwidth, the multipath caused by breathing and the multipath caused by other movements are identified in a more refined time domain spectrum, and the breathing parameters of stationary targets in dynamic and static environments are detected in a non-contact manner.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is:

[0007] A non-contact human respiration real-time monitoring method for a Wi-Fi complex interference environment includes the following steps:

[0008] S1: Build a channel state information (CSI) data acquisition hardware platform;

[0009] S2: Preprocessing the collected data, including eliminating phase offset, signal filtering, and denoising the waveform.

[0010] S3: synchronously switch channels through the Wi-Fi transceiver, obtain breathing-related data in the indoor area under the Wi-Fi interference environment, calculate the interference strength P and the activity ratio A through the CSI signal strength vector and the CSI noise threshold, and build the interference feature mapping matrix F(P, A, PRR) in combination with the CSI packet reception rate PRR, and use the matrix to calculate the interference index I-index for interference discrimination;

[0011] S4: Solve the signal power spectrum and measure the breathing parameters in the signal amplitude sequence corresponding to the multipath caused by breathing. Cut the data fragments by using a sliding window, and use a dynamic subcarrier selection algorithm based on the interference index to select the subcarrier combination with the weakest mutual correlation in the interference data to achieve Wi-Fi interference filtering.

[0012] S5: Based on step S4, extract time domain feature values, including mean, standard deviation, mean absolute deviation, maximum value, and percentile, and classify the feature data by label according to the activity type.

[0013] S6: Using the SVM-based classification algorithm and taking the feature data set of step S3 as input, a SVM multi-activity classification model suitable for the Wi-Fi interference environment is constructed, and the model is tested and optimized to obtain the current breathing condition recognition result.

[0014] Further, step S2 specifically includes the following steps:

[0015] S2_1: First, in the indoor area under Wi-Fi interference environment, data is collected according to the types of no one, sitting breathing, supine breathing, prone breathing, and side-lying breathing. Then, the amplitude, time stamp, CSI signal strength vector, CSI noise threshold, and CSI packet reception rate PRR in the CSI data are extracted according to the activity type and time sequence.

[0016] S2_2: The interference strength P and the active ratio A are calculated by using the CSI signal strength vector and the CSI noise threshold to obtain the interference feature F(P, A). The Wi-Fi interference feature mapping matrix S(P, A, PRR) is constructed in combination with the CSI packet reception rate PRR. The matrix is ​​then used to calculate the interference index of each channel.

[0017] S2_3: The interference index I_index is used to reflect the Wi-Fi interference of the channel. The data is detected for interference and the interference degree is divided according to the interference index. The CSI data with interference and the CSI data without interference are marked according to the activity type, time sequence, and interference degree to construct an interference data set and a non-interference data set.

[0018] S2_4: Extract the interference filtered data set and the undisturbed data set, and use Butterworth low-pass filtering combined with wavelet transform denoising to process the data and filter out irrelevant information such as noise and outliers in the data.

[0019] Further, step S3 specifically includes the following steps:

[0020] H rssi is a vector composed of RSSI values ​​in CSI data collected at a period of Ts, and |H rssi |Indicates the number of RSSIs collected, passing the noise threshold Noise thr Determine whether the extracted RSSI value is noise, that is, when the RSSI value is less than or equal to Noise thr When , this value is regarded as channel noise;

[0021] (1)H rem_rssi H rssi After removing the noise, H rssi vector, the interference intensity is:

[0022]

[0023] (2)H actived is a 0 / 1 vector if and only if H rssi (i)>Noise thr , H rssi (i)=1, otherwise H rssi (i) = 0, then the activity ratio is:

[0024]

[0025] (3) Construct the Wi-Fi interference feature mapping matrix S(P, A, PRR) based on interference intensity, active ratio, and packet reception rate;

[0026] ① Use the grid to divide the value space of parameters P and A. The value range in the P axis direction is [P min , P max ], the unit width is ΔP; the value range in the A-axis direction is [0, A max ], the unit width is ΔA; let each grid point correspond to an interference feature, and establish a feature mapping matrix F to store the PRR estimates corresponding to all interference features (P, A); given a set of measurement values ​​of feature (P, A) PRR1, PRR2, ..., PRR s The PRR estimate corresponding to this feature is

[0027] ② After collecting a sufficient number of measurements for each interference feature (P, A), establish the feature mapping matrix S;

[0028] ③ According to the feature mapping matrix S, the interference index of each channel is calculated. The interference characteristic of the current Wi-Fi channel is (P, A), and the interference characteristic of channel i is (P i , A i ), interference characteristics (P i , A i The distance between (P, A) and (P, A) is:

[0029]

[0030] Let the K interference features closest to (P, A) in F be (P1, A2), (P2, A2), … (P k , A k ), the PRR estimation value of the interference state corresponding to the interference feature (P, A) is:

[0031]

[0032] ④ Assuming that the Wi-Fi interference source channel is m and the transmitter channel is n, the interference index I_index is calculated as:

[0033]

[0034] Further, step S4 specifically includes the following steps:

[0035] S4_1: According to the CSMA / CA mechanism of the IEEE 802.11n standard, the overlap of subcarriers in the spectrum will lead to uneven subcarrier power distribution and weaken the correlation between subcarriers. Therefore, a dynamic subcarrier selection algorithm based on interference index is used to filter the interference data set. The algorithm steps are as follows:

[0036] (1) Set the sliding window size w fThe interfered CSI data stream is partitioned, and the total number of sliding windows N is as follows:

[0037]

[0038] Where L represents the length of the data stream, w f represents the sliding window size, i represents the i-th CSI data packet;

[0039] (2) The number of cycles is determined according to the total number of windows N, the number of subcarrier selections k is determined according to the interference index I-index, and the subcarrier combination C to be selected is output.

[0040] (3) Find the mode of the data in set C and arrange them in descending order, and add the first k subcarriers to set C. k , judge C k Are the subcarriers distributed in the intervals [1, 10], [11, 20], and [21, 30]? If so, let C s =C k Output the result; otherwise, reselect the subcarrier for judgment until the above conditions are met;

[0041] S4_2_1: The noise components of the wavelet detail coefficients are removed from the data by using the threshold processing method; since the frequencies of activities such as sitting breathing, supine breathing, prone breathing, and side-lying breathing are relatively low, their characteristics are mainly contained in the approximate components. By extracting the approximate components containing low-frequency information, the characteristics of different human body movements are analyzed later; finally, the multi-link data fusion method is used to process the undisturbed data. The specific steps are as follows:

[0042] (1) Initially, all data streams of the first action are given the same weight set to 1, and mean aggregation is performed when forming the final collected data, where N represents the number of data streams.

[0043] (2) Extract the feature value of the action fingerprint from the original training set, obtain its cluster center through processing, and use the feature value e0 corresponding to this center point as the action standard. In addition, after performing this action classification, the feature value e0 of this group of actions is obtained. m , where n represents the dimension of the eigenvalues, as follows:

[0044]

[0045]

[0046] (3) Calculate the characteristic value e of the N data streams in this action m The Euclidean distance from the standard action feature value e0 is obtained as the Euclidean distance L m As shown below:

[0047]

[0048] (4) The Euclidean distance L obtained based on N data streams m Get the weight of each link, then normalize it to get the Euclidean distance of the link As shown below:

[0049]

[0050] (5) Using the normalized Euclidean distance The weight of this data stream is fed back to each data link, and the time sequence k(t)′ of the aggregated CSI is calculated by taking the average, as shown in the following formula:

[0051]

[0052] S4_2_2: All processed data sets are classified and stored according to activity types for feature extraction.

[0053] S4_3: The frequencies of the subcarriers in the combined channel H are non-uniformly distributed. The IFFT algorithm is not suitable for processing such data. The optimal solution of a is solved by the least squares estimation in the complex domain, and then t is obtained. n The power delay spectrum P(t n , τ)

[0054] The least squares estimation in the complex domain is realized by converting the optimal problem of solving a in the matrix equation H=Fa in the complex domain into an equivalent matrix equation in the real domain:

[0055]

[0056] Among them, H Re , H im Represent the real and imaginary data of H respectively.

[0057] According to the least squares estimation formula in the real number domain:

[0058]

[0059] Then t n The power delay spectrum P(t n , τ) = |a Re +i*a 1m |, i is the imaginary unit.

[0060] Further, step S5 specifically includes the following steps:

[0061] S5_1: According to the result of step S4, extract the time domain feature values ​​of the activity, the feature values ​​include mean, standard deviation, mean absolute deviation, maximum value, and percentile, wherein the percentile part is specifically:

[0062] By calculating the P percentile of CSI to determine the position and approximate distribution of a sample in the entire data set Na, the speed of different body parts and legs during human activities is evaluated. The 25th percentile and 75th percentile are shown below:

[0063]

[0064]

[0065] In the above formula, s represents the index number of the data packet, k represents the kth subcarrier, Na represents the number of active continuous data packets, and H k Indicates that a CSI sequence containing human activities is extracted;

[0066] S5_2: The above characteristic data are sorted and classified according to activity type and time series, and used as input values ​​of the respiratory activity classification and recognition model based on the SVM algorithm.

[0067] Further, step S6 specifically includes the following steps:

[0068] S6_1: The extracted effective mean, standard deviation, mean absolute deviation, maximum value, and percentile feature data are mapped to the high-dimensional space, which becomes a linear separable problem in the high-dimensional space. The kernel function is introduced as

[0069]

[0070] The human respiratory activity recognition and classification problem is expressed as:

[0071]

[0072] subject to you i (w T φ(x i )+b)≥1-ε i

[0073] ε i =0, i=1, ..., l

[0074] Where w represents the weight coefficient of the optimal classification hyperplane; b represents the classification threshold;

[0075] ε i is the relaxation factor introduced to solve the linear inseparable problem. Only outliers have relaxation variables, ε iThe larger it is, the farther away it is from the group. Considering that abandoning these outliers will lose classification accuracy, a penalty factor C (C>0) is introduced.

[0076] The final decision function is:

[0077]

[0078] According to the data characteristics of no one, sitting breathing, supine breathing, prone breathing, and side-lying breathing, a suitable kernel function and penalty parameter C are selected to determine the SVM. The kernel function selects the radial basis RBF kernel function as shown below:

[0079] K(x i , x j )=exp(-γ‖x i -x j || 2 ),γ>0

[0080] Where γ is the kernel parameter;

[0081] Combined with the characteristics of CSI fall activity data in Wi-Fi interference environment, a SVM algorithm model based on RBF kernel function is constructed. Its performance is mainly determined by the error penalty factor C and the kernel function width g. The parameters C and g are selected from N and M values ​​respectively. SVM is trained for N×M combinations of (C, g) respectively, and then the optimal network parameters of the final training SVM are determined by searching for the highest value.

[0082] S6_2: Since the support vector machine itself does not support multi-class problems, this method involves the classification of 5 types of activities. Therefore, a combination of 5 SVM classifiers is required to complete the classification of multi-class samples. Therefore, the "one-to-one" classification mode in multi-classification is adopted; in the actual model training, 80% of the data of the 5 characteristic values ​​of no one, sitting breathing, supine breathing, prone breathing, and side-lying breathing types are selected as the pre-training set, and 20% of the data is used as the test set; then two activities are selected from the pre-training set each time as the training set, and they are trained by SVM classifiers respectively, and a total of 10 training results can be obtained; in the test verification, the corresponding test sets are tested on the 10 results respectively, and finally voting is adopted to obtain the classification results of the 5 activities.

[0083] The beneficial effects of the present invention are: by combining the CSI data of multiple Wi-Fi channels and calculating the time domain spectrum, the time resolution capability of the Wi-Fi signal is improved, and the accuracy of non-contact respiratory detection in dynamic environments is improved. In addition, in the processing of undisturbed data, a multi-data link fusion method is proposed to perform weighted fusion of data features, enriching the diversity of the data.

[0084] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0086] Figure 1 This is an overall flow chart of the method for human breathing recognition in a Wi-Fi interference environment described in the present invention;

[0087] Figure 2 The human breathing condition recognition model in the Wi-Fi interference environment described in the present invention;

[0088] Figure 3 This is a flow chart of the dynamic subcarrier selection algorithm of the present invention;

[0089] Figure 4 The original collected CSI waveform and the processed waveform described in the present invention;

[0090] Figure 5 7 kinds of behavior confusion matrices for self-collection are identified in the embodiment of the present invention, and the 1st to 5th rows / columns correspond to 5 breathing modes, namely, no one, sitting breathing, supine breathing, prone breathing, and side-lying breathing. DETAILED DESCRIPTION

[0091] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0092] In order to facilitate the understanding of the present invention, the terms involved in the present invention are first explained:

[0093] See also Figure 1 to Figure 3 ,A human breathing action recognition method for WIFI CSI dynamic subcarrier selection under Wi-Fi interference environment. Figure 1 FIG. 1 is an overall flow chart of the method for identifying human falls in a Wi-Fi interference environment according to the present invention, as shown in FIG. Figure 1 As shown, the method is mainly divided into the following steps:

[0094] S1: Use port y on the receiving end network card to connect to the wireless antenna;

[0095] S2: After the system is started, the initial channel of the transceiver is set to #102. The sending device then sends a packet of data, switches to channel #110 and starts the timer. The receiving device receives the data and collects the CSI under channel #102, and switches to channel #110 at the same time. Each time the timer of the sending end expires, the transceiver repeats the above process to complete the CSI sampling under channels #102, #110, #118, #126, and #134.

[0096] S3: Obtain data related to activities such as falls in indoor areas under Wi-Fi interference environments, calculate interference strength P and activity ratio A through CSI signal strength vector and CSI noise threshold, and construct interference feature mapping matrix F(P, A, PRR) in combination with CSI packet reception rate PRR. Use this matrix to calculate interference index I-index for interference discrimination.

[0097] S4: The dynamic subcarrier selection algorithm CSI-DSSA based on interference index is used to select the subcarrier combination with the weakest mutual correlation in the interference data to achieve Wi-Fi interference filtering; the multi-link data fusion method CSI-MLDF is used to aggregate the time domain feature information of multiple data streams in the non-interfered data to improve data reliability;

[0098] S5: Extract time domain feature values ​​based on S4. Feature values ​​include mean, standard deviation, mean absolute deviation, maximum value, and percentile. Classify feature data according to activity types.

[0099] S6: Based on the SVM classification algorithm, the feature data set of S5 is used as input to build an SVM multi-activity classification model in a Wi-Fi interference environment, and the model is tested and optimized to obtain the breathing activity recognition results.

[0100] After obtaining the specific CSI data values ​​in each group, the online platform uses the wireless fingerprint database set to match the data set obtained by accurate identification. That is, the online platform uses the wireless fingerprint database set to match the CSI data values ​​subdivided from each group, and finally accurately obtains the state of human breathing.

[0101] To summarize, in the embodiment of the present invention, the behavior recognition device transmits a Wi-Fi signal, and then collects the Wi-Fi signal reflected by an object or a human body. First, the collected Wi-Fi signals are divided into perimeters, and Wi-Fi signals in a preset area are selected. Then, CSI data values ​​in these Wi-Fi signals are obtained, and then these CSI data values ​​are continuously sent to the computing platform.

[0102] The present invention can realize the recognition of human behavior without using any video imaging equipment or wearable sensor equipment. It is not limited by factors such as user privacy and lighting conditions, and does not require users to wear sensors at all times. It is not affected by the high energy consumption and resource waste problems under frequent information interaction. It can realize the recognition of human behavior without complicated operations, is very convenient, and greatly improves the user experience.

[0103] The contents described in the embodiments of this specification are merely enumerations of implementation forms of the inventive concept and are for illustrative purposes only. The protection scope of the present invention should not be considered to be limited to the specific forms described in this embodiment, and the protection scope of the present invention also extends to equivalent technical means that can be thought of by ordinary technicians in this field based on the inventive concept.

Claims

1. A non-contact human breathing real-time monitoring method for Wi-Fi complex interference environment, characterized in that: The method comprises the following steps: S1: Build a channel state information (CSI) data acquisition hardware platform. S2: Preprocessing the collected data, including eliminating phase offset, signal filtering, and denoising the waveform. S3: synchronously switch channels through the Wi-Fi transceiver, obtain breathing-related data in the indoor area under the Wi-Fi interference environment, calculate the interference strength P and the activity ratio A through the CSI signal strength vector and the CSI noise threshold, and build the interference feature mapping matrix F(P, A, PRR) in combination with the CSI packet reception rate PRR, and use the matrix to calculate the interference index I-index for interference discrimination; S4: Solve the signal power spectrum and measure the breathing parameters in the signal amplitude sequence corresponding to the multipath caused by breathing. Use the dynamic subcarrier selection algorithm based on the interference index to select the subcarrier combination with the weakest mutual correlation in the interference data to achieve Wi-Fi interference filtering; S5: Based on step S4, the data segments are cut by using a sliding window method, and time domain feature values ​​are extracted. The feature values ​​include mean, standard deviation, mean absolute deviation, maximum value, and percentile. The feature data are labeled and classified according to the activity type. S6: Using the SVM-based classification algorithm and taking the feature data set of step S3 as input, a SVM multi-activity classification model suitable for the Wi-Fi interference environment is constructed, and the model is tested and optimized to obtain the current breathing condition recognition result.

2. The non-contact human respiratory parameter real-time measurement method for Wi-Fi interference environment according to claim 1, characterized in that: Step S2 specifically includes the following steps: S2_1: First, in the indoor area under Wi-Fi interference environment, data is collected according to the types of no one, sitting breathing, supine breathing, prone breathing, and side-lying breathing. Then, the amplitude, time stamp, CSI signal strength vector, CSI noise threshold, and CSI packet reception rate PRR in the CSI data are extracted according to the activity type and time sequence. S2_2: The interference strength P and the active ratio A are calculated by using the CSI signal strength vector and the CSI noise threshold to obtain the interference feature F(P, A). The Wi-Fi interference feature mapping matrix S(P, A, PRR) is constructed in combination with the CSI packet reception rate PRR. The matrix is ​​then used to calculate the interference index of each channel. S2_3: The interference index I_index is used to reflect the Wi-Fi interference of the channel. The data is detected for interference and the interference degree is divided according to the interference index. The CSI data with interference and the CSI data without interference are marked according to the activity type, time sequence, and interference degree to construct an interference data set and a non-interference data set.

3. The non-contact human respiratory parameter real-time measurement method in a Wi-Fi interference environment according to claim 1, characterized in that: Step S3 specifically includes the following steps: is a vector composed of RSSI values ​​in CSI data collected at a period of Ts, Indicates the number of RSSIs collected, passing the noise threshold Noise thr Determine whether the extracted RSSI value is noise, that is, when the RSSI value is less than or equal to Noise thr When , this value is regarded as channel noise; ①Let P min and P max are the minimum and maximum possible interference levels, A max The maximum possible active ratio is obtained by dividing the value space of parameters P and A into grids. The value range in the P axis direction is [P min , P max ], the unit width is AP; the value range in the A-axis direction is [0,A max ], the unit width is ΔA; let each grid point correspond to an interference feature, and establish a feature mapping matrix F to store the PRR estimates corresponding to all interference features (P, A); given a set of measurement values ​​of feature (P, A) PRR1, PRR2, ..., PRR s The PRR estimate corresponding to this feature is ② After collecting a sufficient number of measurements for each interference feature (P, A), establish the feature mapping matrix S; ③ According to the feature mapping matrix S, the interference index of each channel is calculated. The interference characteristic of the current Wi-Fi channel is (P, A), and the interference characteristic of channel i is (P i , A i ), interference characteristics (P i , A i ) and (P, A) is: d((P i , A i ), (P, A)) Let the K interference features closest to (P, A) in F be (P1, A2), (P2, A2), ... (P k , A k ), the PRR estimation value of the interference state corresponding to the interference feature (P, A) is: ④ Assuming that the Wi-Fi interference source channel is m and the transmitter channel is n, the interference index I_index is calculated as:

4. The non-contact human respiratory parameter real-time measurement method in a Wi-Fi interference environment according to claim 1, characterized in that: Step S5 specifically includes the following steps: S5_1: According to the result of step S4, extract the time domain feature values ​​of the activity, the feature values ​​include mean, standard deviation, mean absolute deviation, maximum value, and percentile. S5_2: The above characteristic data are sorted and classified according to activity type and time series, and used as input values ​​of the respiratory activity classification and recognition model based on the SVM algorithm.

5. The non-contact real-time measurement method of human respiratory parameters in a Wi-Fi interference environment according to claim 1, characterized in that: Step S6 specifically includes the following steps: The classification kernel function is The human respiratory activity recognition and classification problem is expressed as: subject to y i (w T φ(x i )+b)≥1-ε i e i =0,i=1,...,l Where w represents the weight coefficient of the optimal classification hyperplane; b represents the classification threshold; ε i is the relaxation factor introduced to solve the linear inseparable problem. Only outliers have relaxation variables, ε i The larger the value, the farther away from the group. Considering that abandoning these outliers will lose classification accuracy, a penalty factor C (C>0) is introduced. The dual problem of the above problem is expressed as: subject to y T ɑ=0 0≤ɑi≤C,i=1,...,l Where Q ij =y i y j k(x i , x j ), e is a vector of all 1s, and the final decision function is: According to the data characteristics of no one, sitting breathing, supine breathing, prone breathing, and side-lying breathing, a suitable kernel function and penalty parameter C are selected to determine the SVM, and the kernel function selects the radial basis RBF kernel function. Combined with the characteristics of CSI fall activity data in Wi-Fi interference environment, a SVM algorithm model based on RBF kernel function is constructed. Its performance is mainly determined by the error penalty factor C and the kernel function width g. Parameters C and g are selected from N and M values ​​respectively. SVM is trained for N×M combinations of (C, g) respectively, and then the optimal network parameters of the final trained SVM are determined by searching the highest value.

Images