CN108548545A - A kind of non-contact more people's step-recording methods and system based on commercial Wi-Fi - Google Patents

Abstract

The invention discloses a non-contact multi-person step counting method and system based on commercial Wi-Fi. The method includes the following steps: collecting the CSI signals of participants through commercial Wi-Fi; defining d(t) as a time sliding window The sum of the average absolute deviations of all amplitude data, calculate the noise level N(t) within the time period t according to d(t), and filter out the noise signal from the CSI signal; remove each subcarrier in the CSI signal through a Hamel filter Static components of the static component, use the Savitsky-Gray filter to further remove other noises; introduce tensor decomposition, through signal decomposition and signal fusion, to obtain the running signal generated by a single person, and then measure the number of steps of each person through peak monitoring estimate. The system includes: a microprocessor, a Wi‑Fi transmitter, and a Wi‑Fi receiver.

Description

A non-contact multi-person step counting method and system based on commercial Wi-Fi

technical field

The present invention relates to a computer network, in particular to the field of feature extraction and data mining, in particular to a method and system for non-contact multi-person step counting based on commercial Wi-Fi.

Background technique

With the development of wireless technology, wireless signals can not only be used to transmit data, but also can be used to sense the environment. In indoor environments, wireless signals generated by WiFi devices are directly irradiated, reflected and scattered on different objects around them, and finally reach the receiving device , so the wireless signal carries the information of the surrounding environment.

By establishing the relationship between the waveform characteristics of the channel state information in the physical layer of the receiving device and the target task, indoor positioning, behavior recognition, security monitoring, medical monitoring, etc. can be performed. Previous behavior recognition systems generally use cameras, wearable sensors, and software radio devices to track motion information. Although these devices capture motions with high accuracy, they are expensive and have low universality.

Other systems use machine learning methods to train CSI (Channel State Information) signals in commercial WiFi devices, and finally use models to identify relevant behaviors. However, the model is not only time-consuming to train, but also more dependent on features. Large, so it is difficult to develop a lightweight and robust user interface using the training method.

The present invention will use conventional commercial Wi-Fi equipment to estimate the number of steps, which is cheap and highly universal; technically, it will use an unsupervised method to estimate the number of steps of multi-person movement, which is efficient and does not require training.

Contents of the invention

The present invention provides a non-contact multi-person step counting method based on commercial Wi-Fi. The present invention uses a series of signal processing and data mining technologies to process Wi-Fi signals, and realizes that conventional Wi-Fi devices can be used to Many people carry out step counting, see the following description for details: 1, a kind of non-contact multi-person step counting method based on commercial Wi-Fi, described method comprises the following steps:

Collect participants' CSI signals via commercial Wi-Fi;

Define d(t) as the sum of the average absolute deviations of all amplitude data in a time sliding window, calculate the noise level N(t) in the time period t according to d(t), and filter out the noise signal from the CSI signal;

The static component of each subcarrier in the CSI signal is removed by the Hampel filter, and other noises are further removed by the Savitsky-Gray filter;

Tensor decomposition is introduced, and the running signal generated by a single person is obtained through signal decomposition and signal fusion, and then the number of steps of each person is estimated through peak monitoring.

said

Among them, a _p (n) represents the amplitude value at which the data packet index corresponding to subcarrier p is n, P is the maximum value of the subcarrier index (90 here), and N is the set of all data packet indexes in the sliding window , L is the length of the sliding window, and E is the mean value of the amplitude of all data packets in the sliding window.

The method expands the CSI amplitude matrix obtained after removing other noises into a CSI tensor by using a Hankelization method.

The signal fusion is specifically:

Use autocorrelation to enhance the periodicity of the decomposed signal;

Use the average Fresche distance to measure the similarity between two signals. The average Fresche distance considers both the position and order of the points along the curve. It can identify the shift in the autocorrelation signal and is very suitable for measuring the curve. the similarity between

Use the stable roommate matching algorithm to match the decomposition signals generated by each person with the Fresche distance between the autocorrelation signals as the metric;

Finally, two similar signals are fused into one signal by averaging. On the one hand, averaging can reduce the deviation of the decomposed signals, and on the other hand, it can ensure that the fusion of signals is performed at the same time.

A non-contact multi-person pedometer system based on commercial Wi-Fi, said system comprising: a microprocessor, a transmitting end of WiFi, and a receiving end of WiFi,

The transmitting end and the receiving end of the WiFi are placed on the ground, and the transmitting end and the receiving end are in a straight line;

After collecting the CSI data on the receiving end, the receiving end sends the CSI data to the microprocessor through the TCP/IP protocol, and the microprocessor processes the CSI data through MATLAB;

That is, the static component of each subcarrier in the CSI signal is removed by the Hampel filter, and the Savitsky-Gray filter is used to further remove other noises;

Tensor decomposition is introduced, and the running signal generated by a single person is obtained through signal decomposition and signal fusion, and then the number of steps of each person is estimated through peak monitoring.

The beneficial effects of the technical solution provided by the invention are:

1. The present invention uses conventional Wi-Fi equipment to realize step counting by multiple people indoors, which is cheap and highly universal.

2. The data processing method proposed by the present invention can also be used in other fields and has good versatility; the present invention provides new ideas and concepts for the development of immersive game devices.

3. On the one hand, the significance of step counting is to standardize the action of running in situ. On the other hand, it can increase the fun of running in situ and turn fitness into an entertainment game. Set the number of steps and steps, and then complete the predetermined number of steps within the specified time. White-collar workers can play games for 3 minutes in the office to relieve stress. Family members can use their spare time to jog and count in place to enhance their relationship.

4. The present invention can also be used for medical rehabilitation, for example: patients do quantitative exercise every day to achieve the purpose of rehabilitation.

Description of drawings

Fig. 1 is a flow chart of a non-contact multi-person step counting method based on commercial Wi-Fi;

Figure 2 is a schematic diagram of the comparison before and after CSI amplitude noise reduction;

Among them, (a) is the original CSI signal; (b) is the signal after CSI noise reduction.

Figure 3 is a schematic diagram of the results of CP decomposition;

Fig. 4 is a schematic diagram of the result of signal fusion;

Figure 5 is a system structure diagram.

Detailed ways

In order to make the purpose, technical solution and advantages of the present invention clearer, the implementation manners of the present invention will be further described in detail below.

Example 1

A kind of non-contact multi-person step counting method based on commercial Wi-Fi, referring to Fig. 1, this method comprises the following steps:

101: Collecting CSI signals of participants through commercial Wi-Fi;

102: Define d(t) as the sum of the average absolute deviations of all amplitude data in a time sliding window, calculate the noise level N(t) within the time period t according to d(t), and filter out the noise signal from the CSI signal ;

103: Eliminate the static component of each subcarrier in the CSI signal through the Hampel filter, and use the Savitsky-Gray filter to further remove other noises;

104: Introduce tensor decomposition, through signal decomposition and signal fusion, obtain the running signal generated by a single person, and then estimate the number of steps of each person through peak monitoring.

Wherein, described in step 102:

Among them, a _p (n) represents the amplitude value at which the data packet index corresponding to subcarrier p is n, P is the maximum value of the subcarrier index (90 here), and N is the set of all data packet indexes in the sliding window , L is the length of the sliding window, and E is the mean value of the amplitude of all data packets in the sliding window.

Furthermore, this method uses the Hankelization method to expand the CSI amplitude matrix obtained after removing other noises into a CSI tensor.

Wherein, the signal fusion is specifically:

Use autocorrelation to enhance the periodicity of the decomposed signal;

Use the average Fresche distance to measure the similarity between two signals. The average Fresche distance considers both the position and order of the points along the curve. It can identify the shift in the autocorrelation signal and is very suitable for measuring the curve. the similarity between

Use the stable roommate matching algorithm to match the decomposition signals generated by each person with the Fresche distance between the autocorrelation signals as the metric;

Finally, two similar signals are fused into one signal by averaging. On the one hand, averaging can reduce the deviation of the decomposed signals, and on the other hand, it can ensure that the fusion of signals is performed at the same time.

To sum up, the embodiments of the present invention use a series of signal processing and data mining technologies to process Wi-Fi signals, and realize step counting of multiple people using conventional Wi-Fi devices.

Example 2

The embodiment of the present invention proposes a non-contact multi-person step counting method based on commercial Wi-Fi, referring to Fig. 1, the method includes the following steps:

1. Motion detection

First, the amplitude information is extracted from the collected WiFi signals, and then it is necessary to monitor whether jogging starts. During the experiment, it can be observed that the fluctuation of the time series of the CSI signal generated by human motion is larger than that of the background noise. In order to quantify this fluctuation, d(t) is defined as the average absolute deviation of all amplitude data within a time sliding window Sum:

Among them, a _p (n) represents the amplitude value at which the data packet index corresponding to subcarrier p is n, P is the maximum value of the subcarrier index (90 here), and N is the set of all data packet indexes in the sliding window , L is the length of the sliding window, and E is the mean value of the amplitude of all data packets in the sliding window.

Since the fluctuation of the CSI signal is mainly caused by noise when there is no movement in the room, and the noise level changes slowly over time, a dynamic threshold algorithm is used to track the change of the noise level.

First, calculate the average absolute deviation d(t) in a time sliding window according to formula (1). The time sliding window used in the embodiment of the present invention is 200 sampling points, about 0.2s. In this time sliding window, use the index A moving average is used to update the noise level N(t) over time period t:

N(t)=(1-α _n )N(t-1)+α _n ×d(t) (2)

Among them, α _n is a constant coefficient.

Here, the embodiment of the present invention sets the coefficient α _n to 0.15. Once the exercise starts or ends, the time series of the CSI signal will fluctuate violently at this moment. In addition, combined with the experimental data, when the embodiment of the present invention sets the monitoring value to be 3 times the noise level update value N'(t), it is considered that The jogging starts, and then the embodiment of the present invention performs noise reduction processing on the data.

2. Data noise reduction

The CSI data obtained from commercial WiFi devices contain static components, low-frequency interference and impulse noise. On the one hand, it is caused by clock asynchrony, radio wave interference, and changes in the transmit power of the transmitter. On the other hand, in the actual environment In the process of jogging, due to the shaking of various parts of the body, the CSI data will also be mixed with different degrees of low-frequency noise and high-frequency noise. The existence of these noises increases the difficulty of extracting motion features. In addition, the subsequent steps need The process of matching multi-person signals is very sensitive to noise. This requires not only to effectively remove noise, but also to keep the waveform unchanged as much as possible after denoising. The traditional IIR (infinite impulse response) low-pass filter cannot After filtering, it is very good to keep the waveform unchanged, which is not applicable here. Because the pulse noise bandwidth in CSI data is large and the energy is high, band-pass filtering cannot effectively remove noise.

In order to solve these problems, the embodiment of the present invention firstly uses a Hampel Filter (Hampel filter) to remove the static component of each subcarrier in the CSI data. Specifically, it can be divided into the following two steps:

1) Set a large window and a small threshold Hampel Filter to obtain the static component of each subcarrier;

The specific operation steps of setting a large window and a small threshold Hampel Filter are well known to those skilled in the art, and will not be described in detail in the embodiment of the present invention.

2) Subtract the above static components from the original CSI data; then, use the Savizky-Golay filter (Savizky-Gray filter) to further remove other noises;

Among them, the Savizky-Golay filter is a filtering method based on local polynomial least squares fitting in the time domain. It can remove noise while keeping the CSI amplitude signal waveform unchanged. The specific filtering steps are provided by those skilled in the art. As is well known, this embodiment of the present invention does not describe it in detail.

During specific implementation, other filters or filtering methods may also be used to implement the above steps, which is not limited in this embodiment of the present invention.

3. Multi-person signal extraction

For the extraction of multi-person running signals, since the CSI signal is affected by the independent movement of multiple people at the same time, first of all, factor analysis obviously cannot separate the running signals of each person from the mixed signal of multi-person running, and secondly, when multiple people run in situ , the frequency of the legs comes from multiple people, and time-frequency transformation cannot separate the movement frequencies of multiple people's legs at the same time point. In order to separate the running signals of each person from the CSI amplitude information generated by multiple people running in situ and further refine The granularity describes the movement details of the legs during in-situ running. The embodiment of the present invention introduces the method of tensor decomposition. This part of processing is mainly divided into two parts: signal decomposition and signal fusion. After tensor decomposition, the embodiment of the present invention can Obtain the running signal generated by a single person, and then estimate the number of steps for each person.

1. Signal decomposition

A. Tensor construction

After data noise reduction, the embodiment of the present invention obtains the CSI amplitude matrix, whose dimension is the number of data packets × the number of subcarriers, and then the embodiment of the present invention utilizes the method of Hankel (Hankel) to convert the CSI amplitude The matrix is expanded into a CSI tensor. Specifically, the embodiment of the present invention uses a 2-dimensional Hankel matrix to store the signals of each column of subcarriers in the CSI amplitude matrix, so that 60 columns of subcarrier signals can form a 3-dimensional tensor. Define H _r is the Hankel matrix that the size that the subcarrier r of signal length is N constructs is the Hankel matrix of I * J, and I, J and N satisfy the condition: I+J-1=N, here the embodiment of the present invention sets I=J=(N +1)/2, so for the subcarrier r, after Hankelization, the Hankel matrix H _r as shown below can be constructed:

Here, h _r (i) represents the amplitude value of the data packet index i at the rth subcarrier in the Hr matrix, and N=5000 is set in the experiment, and I=J=2500.

Theory 1. If there are R in-situ running signals in the indoor detection environment, then under the condition that other interferences are ignored, the rank corresponding to the Hankel matrix H _r constructed by the subcarrier r is 2R.

Proof: When analyzing the data structure of the leg signal of jogging in place, assuming that the noise is negligible, the leg signal of the i-th person can be expressed as: The observed signal from each subcarrier can be expressed as:

Here K _i is the coefficient of the leg signal of the i-th person jogging in place, where The legstep signal of the ith person in Y(t) It can be further decomposed using Euler's formula, then:

Each leg signal can be split independently into two exponential signals with different parameters. For R leg signals, there are:

Here, the new signal its coefficient For packets received in discrete time, the receive signal can be used To represent. Note that Y(n) can be considered as an exponential polynomial composed of 2R different exponential terms, where n=1,2,3,...,N, and Y(n) is mapped to a size of The Hankel matrix, so there are:

It is already known that the Hankel matrix can be decomposed by Vandermode (Vandermont), namely:

Here the Vandermode matrix and

Since the Vandermonde (Vandermonde) matrix is composed of different series, it is of full rank, so the rank of the Hankel matrix is 2R.

According to theory 1, 2R signal components need to separate out R leg signals of running in situ. Next, consider the influence of noise on the Hankel matrix H _r , due to the influence of noise, the Hankel matrix H _r is actually full rank, but theory 1 shows that the rank of H _r is 2R, which means that as long as the signal-to-noise ratio of the signal is not too high Low, the decomposition components of the first 2R weights are more robust than the remaining components. This also shows that the structure of the Hankel matrix can effectively separate the running signal from the white noise. In fact, through tensor decomposition, different signals can be well denoised and separated. Subsequently, the present invention performs CP decomposition on the tensor to extract the jogging signals of multiple people.

2. CP decomposition

After the construction of the CSI tensor is completed, CP decomposition is used to estimate the running signals of multiple people. According to theory 1, the number of CSI tensor decomposition components has been determined to be 2R, and the CP decomposition of the tensor is performed using the alternating least squares method, and the CP decomposition is completed. Afterwards, 2R rank 1 tensors can be obtained, and each rank 1 tensor is composed of the outer product of 3 vectors, and 3 × 2R vectors are composed of 3 matrices represented by A, B, and C respectively. In order to ensure the effective decomposition of components Next, we will prove the uniqueness of CP decomposition. Regarding the uniqueness of CP decomposition, its basic theory is as follows:

Fact 1: For tensor χ of rank L, the CP decomposition of tensor χ is unique if k _A + k _B + k _C ≥ 2L+2. Here k _A , k _B and k _C respectively represent the matrices A, B and C with rank k, where rank k means that the maximum number of linearly independent columns of a matrix is k.

Based on the above facts, for the created CSI tensor, there are the following theories:

Theory 2: For a created CSI tensor χ of rank 2R, the CP decomposition of the tensor χ is unique.

Proof: The CSI tensor χ is created using K Hankel matrices, according to theory 1, the rank of the rth Hankel matrix H _r is 2R, for the rank k _A =2R of A matrix and B matrix, k _B =2R, and the other On the one hand, since the amplitude data of the 60 subcarriers come from two antennas, the two antennas are independent of each other, so the rank k _c ≥ 2 of the matrix C. Therefore, the expression k _A +k _B +k _C ≥ 2R+2R+2=2(2R)+2 satisfies fact 1, and the proof is complete.

Theorem 2 shows that the result of CP decomposition of the created CSI tensor is unique, and it can effectively extract the leg signals of multiple people running. The system uses matrix A as the decomposition signal a ₁ , a ₂ , a ₃ , …, a _2R .

3. Signal Fusion

After the constructed CSI tensor data is decomposed by CP, 2R signals can be obtained from the A matrix, that is, S ₁ , S ₂ , S ₃ ,...,S _2R , but their indexes are randomly arranged, and the decomposed signals are not There is no guarantee that similar signals are located in adjacent positions. Therefore, the decomposed signals must be pairwise matched. A signal matching algorithm is designed to make two similar signals belonging to the same person match into a pair.

First, use autocorrelation to strengthen the periodicity of the decomposed signal. There are two main reasons for using the autocorrelation function. On the one hand, autocorrelation of the decomposed signal can increase the data length and improve the accuracy of peak detection. On the other hand, due to The amplitude signal obtained after CP decomposition has an offset, and the autocorrelation can reduce this offset and increase the periodicity of the signal.

Second, the average Fréchet distance is used to measure the similarity between two pairs of signals. The average Fréchet distance takes into account both the position and order of points along the curve, which can identify the offset in the autocorrelation signal, Very suitable for measuring the similarity between curves, it is usually better than the famous Hausdorff (Hausdorff) distance.

Third, use Stable Roommate Matching (stable roommate matching algorithm) to perform pairwise matching on the decomposition signals generated by each person using the Fréchet distance between autocorrelation signals as the metric. Finally, two similar signals are fused into one signal by averaging. On the one hand, averaging can reduce the deviation of the decomposed signals, and on the other hand, it can ensure that the fusion of signals is performed at the same time.

4. Peak monitoring

It is only necessary to constrain the peak (trough) spacing and the peak (trough) width based on the interval time between steps when a normal person is jogging and the time it takes for a normal person to lift a single leg up and down when jogging. In fact, due to the use of Hankel matrix and CP decomposition are used to smooth the motion curve, so the curve rarely contains false peaks, and the final number of legal peaks and troughs is the estimated value of each person's steps.

Example 3

The functions and effects of Embodiments 1 and 2 of the present invention will be shown below in conjunction with the accompanying drawings.

This example uses CSI data processing as an example to give the implementation of the invention, and the specific steps are as follows:

Use a laptop as the WiFi access point, that is, the sending end, and another laptop as the receiving end. Both laptops are installed with Intel 5300NIC and Ubuntu 14.04LTS desktop system. There are 3 antennas on the receiving end and 3 antennas on the transmitting end. Antennas, the distance between the three antennas at each end is a wavelength (5.2cm), and they are located in a straight line. The transmitting end and receiving end of the WiFi are placed on the ground, and the distance between the two is 3m. The transmission rate of the data packet is 1024HZ. The transmitting end and the receiving end are in a straight line, and the transmission link works on the 165 channel with a frequency band of 5.825 GHz. The reason for choosing the 5 GHz frequency band instead of the 2.4 GHz frequency band in the embodiment of the present invention is that the wavelength of the 5 GHz frequency band is relatively short and short Wavelength has a higher resolution to the speed of motion. After collecting the CSI data using the Linux CSI tool on the receiving end, the receiving end sends the CSI data to a computer configured as Intel i7-5600U 2.6GHz through the TCP/IP protocol, and finally processes the CSI data through MATLAB.

First, the present invention monitors when the activity starts. According to the formula (2), once the activity is detected, the data will be denoised.

Figure 2 presents an example of data denoising. By comparing the waveforms of the CSI 30 subcarrier amplitude data before and after denoising, it can be seen that the original CSI amplitude data contains many high-frequency noise and static components. When using our proposed denoising method Afterwards, the static components and most of the noise were effectively removed. In addition, from the overall shape, width and smoothness of the signal, it shows that our noise reduction method not only effectively removes the noise in the CSI amplitude data, but also The waveform of the original CSI amplitude data can be well maintained. In the next processing, this data will be used for single-person signal extraction.

After the noise reduction is completed, construct a tensor for the denoised data. For the technical details of the tensor construction, please refer to the tensor construction part of Technical Solution A. The CSI tensor is then decomposed using CP decomposition. The number of components decomposed is the number of people.

Figure 3 shows the results of the CSI tensor containing 3 people (R=3) after CP decomposition. In order to reflect the accuracy of the method designed in the embodiment of the present invention, during the experiment, 3 experimenters were purposely arranged, The first person steps slowly on the spot, the second person jogs normally on the spot, and the third person runs fast on the spot. From Figure 4, we can see 6 signals, which can be divided into 3 groups. The first group is signal 1 and signal 1. 3. The second group is signal 2 and signal 6, and the third group is signal 4 and signal 5.

First of all, it can be observed that the first group of signal peaks is the sparsest, the second group of signals is denser than the first group of signal peaks, and the third group of signal peaks is the most dense, which is caused by the different step frequencies of the three experimenters. Secondly, it can also be observed that similar signals are not located in adjacent positions. This is because the signal index output by CP decomposition is random, so similar signals need to be identified and fused into one signal to represent a person’s in-situ location running signal.

Signal fusion is then performed. Figure 4 shows the fusion results of the signals after the above processing steps. It can be seen that three signals have been obtained to represent the movement characteristics of the legs of three people running in situ. Taking the third experimenter as an example, the system starts to estimate the number of steps , the tester first leaves the ground with one leg, reaches the highest point at 0.12s, then begins to fall, falls back to the ground at 0.31s, then lifts the other leg, and reaches the highest point at 0.40s, 0.62s Fall back to the ground at the same time, thereby completing 2 steps. It can be seen that the method designed in the embodiment of the present invention not only meticulously depicts the change characteristics of the leg signal when running in situ, but also achieves the purpose of estimating the number of steps.

Counting the fused peaks is the number of steps for each person.

Example 4

A non-contact multi-person pedometer system based on commercial Wi-Fi, as shown in Figure 5, the system includes: a microprocessor, a transmitter and a receiver.

In this paper, an unmodified commercial WiFi device is used for the first time to design and implement an immersive game system similar to parkour, which can perform multi-person step counting. The system includes two WiFi devices, and one device (such as a router) continuously transmits Wireless signal, another device (such as a laptop) is constantly receiving wireless signals. When the game starts, the participants first need to stand upright, with their feet shoulder-width apart, and their arms drooping naturally on both sides of the body. Then bend one leg and lift the foot up to knee height. Swing the arm on the same side backward, and the arm on the opposite side swings forward. The other leg is slightly bent, leaving the foot on the ground, with the weight on the toes. Then put the lifted leg back to the starting position, then lift the other leg, repeat the above action, repeat the action alternately on both sides to the recommended number of steps. In the process of the participants playing the game, the WiFi signal is reflected by the participants to generate a unique and fluctuating CSI signal. At the receiving end, because the WiFi signal is affected by the multipath effect, the system can use signal processing technology to obtain and estimate multiple participants. The number of steps taken by the person jogging in place.

In the embodiments of the present invention, unless otherwise specified, the models of the devices are not limited, as long as they can complete the above functions.

Those skilled in the art can understand that the accompanying drawing is only a schematic diagram of a preferred embodiment, and the serial numbers of the above-mentioned embodiments of the present invention are for description only, and do not represent the advantages and disadvantages of the embodiments.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (5)

1. a kind of non-contact multi-person step counting method based on commercial Wi-Fi, it is characterized in that, described method comprises the following steps: Collect participants' CSI signals via commercial Wi-Fi; Define d(t) as the sum of the average absolute deviations of all amplitude data in a time sliding window, calculate the noise level N(t) in the time period t according to d(t), and filter out the noise signal from the CSI signal; The static component of each subcarrier in the CSI signal is removed by the Hampel filter, and other noises are further removed by the Savitsky-Gray filter; Tensor decomposition is introduced, and the running signal generated by a single person is obtained through signal decomposition and signal fusion, and then the number of steps of each person is estimated through peak monitoring. 2. a kind of non-contact multi-person step counting method based on commercial Wi-Fi according to claim 1, is characterized in that, described Among them, a _p (n) represents the amplitude value at which the data packet index corresponding to subcarrier p is n, P is the maximum value of the subcarrier index (90 here), and N is the set of all data packet indexes in the sliding window , L is the length of the sliding window, and E is the mean value of the amplitude of all data packets in the sliding window. 3. A kind of non-contact multi-person step counting method based on commercial Wi-Fi according to claim 1, characterized in that, said method utilizes the method of Hankelization to expand the CSI amplitude matrix obtained after removing other noises is the CSI tensor. 4. a kind of non-contact multi-person step counting method based on commercial Wi-Fi according to claim 1, is characterized in that, described signal fusion is specifically: Use autocorrelation to enhance the periodicity of the decomposed signal; Use the average Fresche distance to measure the similarity between two signals. The average Fresche distance considers both the position and order of the points along the curve. It can identify the shift in the autocorrelation signal and is very suitable for measuring the curve. the similarity between Use the stable roommate matching algorithm to match the decomposition signals generated by each person with the Fresche distance between the autocorrelation signals as the metric; Finally, two similar signals are fused into one signal by averaging. On the one hand, averaging can reduce the deviation of the decomposed signals, and on the other hand, it can ensure that the fusion of signals is performed at the same time. 5. A non-contact multi-person pedometer system based on commercial Wi-Fi, characterized in that, said system includes: a transmitting end of a microprocessor, a WiFi, and a receiving end of a WiFi, The transmitting end and the receiving end of the WiFi are placed on the ground, and the transmitting end and the receiving end are in a straight line; After collecting the CSI data on the receiving end, the receiving end sends the CSI data to the microprocessor through the TCP/IP protocol, and the microprocessor processes the CSI data through MATLAB; That is, the static component of each subcarrier in the CSI signal is removed by the Hampel filter, and the Savitsky-Gray filter is used to further remove other noises; Tensor decomposition is introduced, and the running signal generated by a single person is obtained through signal decomposition and signal fusion, and then the number of steps of each person is estimated through peak monitoring.