CN116669181A

CN116669181A - Indoor personnel positioning method and system based on WiFi multi-reflection path image

Info

Publication number: CN116669181A
Application number: CN202310693357.1A
Authority: CN
Inventors: 梁泉泉; 吴庆哲; 张琨; 王文华; 于智杰
Original assignee: Shandong University of Science and Technology
Current assignee: Shandong University of Science and Technology
Priority date: 2023-06-13
Filing date: 2023-06-13
Publication date: 2023-08-29
Anticipated expiration: 2043-06-13
Also published as: CN116669181B

Abstract

The invention belongs to the technical field of human body sensing and signal processing, and discloses an indoor personnel positioning method and system based on WiFi multi-reflection path images. Preprocessing the received CSI data, and eliminating phase errors caused by the space between receiving antennas; applying a segmentation algorithm to the data packet in the preprocessed CSI data to obtain more sub-data packets; and (3) performing simulation expansion on the three-antenna equipment into multi-antenna equipment by using a matrix smoothing algorithm on each sub-data packet, acquiring multi-reflection signal path information, calculating angle information and flight time information corresponding to all reflection paths and direct paths transmitted by the target position by using a two-dimensional MUSIC algorithm, and generating a two-dimensional reflection path information image serving as a characteristic image by using the angle information and the flight time information. According to the invention, the testing accuracy of 93% and above can be achieved under two experimental environments through two different classification models respectively.

Description

Indoor personnel positioning method and system based on WiFi multi-reflection path image

Technical Field

The invention belongs to the technical field of human body perception and signal processing, and particularly relates to an indoor personnel positioning method and system based on WiFi multi-reflection path images.

Background

Indoor positioning systems play an increasingly important role in a number of emerging applications such as indoor navigation, augmented reality, disaster relief, and senior citizens. In recent years, wi-Fi based positioning systems have been considered most promising because of the ubiquitous nature of WiFi in public places, businesses, universities and homes.

Existing indoor positioning systems are mainly based on sensors, video methods and radio frequency-based methods. These traditional methods rely mainly on dedicated equipment, have many disadvantages, sensor-based deployment must wear related sensor equipment, which can cause many inconveniences to users, video-based methods need to be implemented in places with better light, and have problems of infringement of user privacy, high cost of radio-frequency-based methods, and inconvenience for popularization.

With the increasing popularity of wireless devices, wiFi has grown very rapidly. An important technology for WiFi success is multiple input multiple output (MIMO, multiple input multiple output, which is an antenna system that uses multiple antennas at both the transmitting end and the receiving end and forms multiple channels between transceivers) that provides high throughput to meet the ever-increasing demand for wireless data traffic. Along with Orthogonal Frequency Division Multiplexing (OFDM), MIMO provides channel state information (Channel State Information, CSI for short, which is a channel property of the communication link) for each transmit and receive antenna pair on each carrier frequency. More recently, CSI measurements from WiFi systems are used for different sensing purposes. WiFi sensing reuses the infrastructure for wireless communications and is therefore easy to deploy and low cost. Furthermore, unlike sensor-based and video-based solutions, wiFi sensors are not invasive or insensitive to lighting conditions.

In recent years, wiFi-based positioning systems mainly rely on received signal strength (Received Signal Strength, RSS), angle of Arrival (AOA), and fingerprints. RSS was the earliest method used for indoor positioning due to its simple receiving device and easy acquisition characteristics. However, the positioning system based on the RSS has the following defects that the received RSS value only contains rough information between the environment and the measured signal, no more multipath information can be provided, the RSS has a larger change with time, and the signal strength is greatly influenced by the environment, namely, the multipath effect. AOA-based positioning systems, which mainly estimate AOA by using multi-signal classification algorithms, face the main challenge of being affected by multipath factors, and the main idea is to find the direct paths between transceiver devices from multiple path information, but the method needs three or more devices for positioning. With the rapid development of machine learning, a positioning system based on a combination of fingerprint and machine learning is proposed. The conventional fingerprint locating method uses RSS as characteristic information, but the RSS information is extremely unstable and can be significantly different with time.

Through the above analysis, the problems and defects existing in the prior art are as follows: in the prior art, only the system using AOA for positioning needs to find direct path information from a plurality of pieces of reflection path information, and the characteristics of the multi-reflection path image generated in the prior art are not obvious, so that the distinction of the reflection path image and the identification of the position characteristics are not facilitated.

Disclosure of Invention

In order to overcome the problems in the related art, the disclosed embodiments of the invention provide an indoor personnel positioning method and system based on WiFi multi-reflection path images.

The technical scheme is as follows: the indoor personnel positioning method based on the WiFi multi-reflection path image comprises the following steps:

s1, transmitting and receiving wireless data by utilizing a plurality of three-antenna devices;

s2, preprocessing the received CSI data, and eliminating phase errors caused by the space between receiving antennas;

s3, a segmentation algorithm is applied to the data packet in the preprocessed CSI data, and more sub-data packet groups are obtained through cutting;

s4, each sub-data packet uses a matrix smoothing algorithm to simulate and expand the three-antenna equipment into multi-antenna equipment, and multi-reflection signal path information is obtained;

s5, for the multi-reflection signal path information, calculating angle information and flight time information corresponding to all reflection paths and direct paths transmitted by the target position by using a two-dimensional MUSIC algorithm, and generating a two-dimensional reflection path information image serving as a characteristic image by using the angle information and the flight time information;

S6, importing pictures of all reflection paths in a plurality of positions into a network classification model for training, and generating a training model;

and S7, classifying the two-dimensional reflection path information image generated by the new position information by using the generated training model, and positioning personnel.

In step S2, preprocessing the received CSI data includes: and a power distributor is used for connecting the transmitting end and the receiver, phase differences among the second antenna, the third antenna and the first antenna of the receiving equipment are calculated respectively, and the phase differences are eliminated by performing linear addition and subtraction operation on the phases through collected data.

In step S3, applying a segmentation algorithm to the data packet in the preprocessed CSI data, to obtain more sub-data packets includes: in the data acquisition process, a data file sets the transmission rate of received data, the total length of a data packet and the transmission interval of the data when transmitting; the total number of data packets transmitted by the data file is 1500 packets; generating an image for every 100 packets of data required to generate the reflected path image; the data file packet length of each transmitted data file is set to 1500 packets, and each data file is divided into 100 packets and one file, so that 15 data files with 100 data packets are obtained.

In step S4, the matrix smoothing algorithm performs analog expansion of the three-antenna device into a multi-antenna device, including: the multi-reflection signal path information is obtained by reorganizing a 3×30 channel state information matrix into a 30×32 data matrix using a matrix smoothing algorithm.

In one embodiment, obtaining multi-reflection signal path information includes: n incoherent far-field narrowband signals are incident on an array consisting of M antennas, and the expression of the received signals is:

；

in the method, in the process of the invention,for the reception signals of the array,/->For transmitting signals +.>For M-dimensional noise data, < >>Is a guide vector;

wherein, the CSI value of the mth antenna is expressed as:

；

in the method, in the process of the invention,for joining arraysReceive signal (s)/(s)>For transmitting signals +.>For M-dimensional noise data, < >>For spatial array->A dimension guide matrix;

the expression of the CSI matrix consisting of 3 antennas and 30 subcarriers for the Intel 5300 wireless network card is as follows:

；

the general formula of the CSI values formed by different antennas and different subcarriers is as followsThe CSI value of the nth subcarrier of the mth antenna is represented as a complex form +.>；/>The value is 1-3; />The value is 1-30;

the method comprises the steps of performing initial phase linear compensation on 30 subcarriers of each antenna by calculating initial phase difference values among different antennas of a receiving device, calibrating a second antenna, a third antenna and a first antenna of a receiver, and correcting phases of different antennas of the receiver;

Performing structural recombination on the input CSI measurement value to obtain a smooth CSI matrix; the smooth CSI matrix generated by recombination directly obtains the information of all reflection propagation paths through a MUSIC algorithm.

In one embodiment, obtaining information of all reflected propagation paths through the MUSIC algorithm includes:

the method comprises the steps of expanding a classical MUSIC algorithm into a two-dimensional 2D-MUSIC algorithm, combining arrival angles and flight times of different signals, displaying information of different reflection paths at different positions in an image, combining the arrival angles and the flight times with spectrogram peaks of the MUSIC, wherein each peak represents a path, the transverse axis of each peak represents arrival angle information, and the vertical axis represents corresponding flight time information;

the spectrum estimation formula in the classical MUSIC algorithm is:

；

in the method, in the process of the invention,is a spectral value obtained by a MUSIC algorithm;

if X is the data matrix obtained after the smoothing process, thenThe covariance matrix representing the data matrix X is:

；

the method comprises the following steps of:

；

wherein E represents an autocorrelation matrix, H represents a conjugate matrix,is->Is>The hermite matrix for positive is +.>A is the signal correlation matrix of the spatial array +.>Dimension-oriented matrix, p->The M eigenvalues obtained by the eigenvalue decomposition are all nonnegative real numbers +. >There are D large eigenvalues and +.>Small eigenvalues and space composed of eigenvectors corresponding to large eigenvalues +.>For signal subspace, the space composed of feature vectors corresponding to small feature valuesIs a noise subspace.

In step S5, calculating angle information and time of flight information corresponding to all the reflection paths and the direct paths emitted by the target position by using the two-dimensional MUSIC algorithm includes:

the one-dimensional MUSIC algorithm is promoted to the two-dimensional MUSIC algorithm, andexpansion to->；

；

In the method, in the process of the invention,，/>，/>for subcarrier frequency spacing, m is the number of antennas, +.>For the time of flight of the p-th path, +.>For the angle information of the p-th reflection path, < >>For the phase of the m dimension of the p-th path, +.>P is the p-th reflection path,>is phase information, d is antenna spacing, +.>Is the wavelength of the signal; />For a matrix of steering vectors carrying angle information and time of flight information +.>Phase shift matrices for different paths;

the angle information and the flight time information are obtained by a two-dimensional MUSIC algorithm spectrum estimation formula, wherein the two-dimensional MUSIC algorithm spectrum estimation formula is as follows:

；

in the method, in the process of the invention,to obtain angle information; />A steering vector representing a matrix obtained by smoothing the received signal, which contains angle-of-arrival information and time-of-flight information, " >Is the noise subspace of the signal,>is->Conjugate matrix of>Is the spectral value obtained by MUSIC algorithm.

In step S6, importing the pictures of all the reflection paths at the plurality of positions into the network classification model for training includes: testing by using an Alexnet lightweight neural network and a classification network Support Vector Machine (SVM); firstly inputting a characteristic image, dividing a data set to divide the data into a proportion of 7:3, using 70% of the data for training, using 30% of the data for testing, optimizing an Alexnet lightweight neural network and a classification network by using a momentum random gradient descent algorithm SGDM, using a GPU (graphic processing unit) acceleration model for training, and comparing a prediction label with a real label;

the network parameter learning optimization principle during training comprises the following steps: the gradient descent principle is that the initial learning rate is set to 0.0001, the number of single packed images is set to 16, the iteration number is 15, and training of a GPU acceleration model is used;

the training model contains the characteristic information of all training position images, and the newly received test photo extracts the characteristic and compares the data of the model with the predicted position.

In step S7, classifying the two-dimensional reflection path information image generated from the new position information using the generated training model includes:

Performing feature extraction training on the collected multi-reflection path images at a plurality of positions by using a neural network/machine learning model, storing an Alexnet lightweight neural network and a classification network, inputting a test sample into the trained Alexnet lightweight neural network and classification network, determining the positions according to the proportion of the predicted categories, and testing 15 pictures generated by data collection once by using a majority voting principle, wherein the 15 pictures are classified as positions with highest occupation ratio and predicted as final positions; the average error of the erroneously identified position coordinates is used to analyze the performance of the system;

the expression of the average error of the error recognition coordinates is:

average error；

Wherein, the liquid crystal display device comprises a liquid crystal display device,representing the calculation of the cumulative sum +.>Is the position estimated by the network model, +.>Is the position where the data is actually collected, N is the maximum value of the abscissa and the ordinate of the position where E, F are set respectively, the number of misclassifications.

Another object of the present invention is to provide an indoor personnel positioning system based on WiFi multi-reflection path image, implementing the indoor personnel positioning method based on WiFi multi-reflection path image, the system includes:

the three-antenna device is used for receiving and transmitting wireless data;

The phase error elimination module is used for preprocessing the received CSI data at first and eliminating phase errors caused by the space between the receiving antennas;

the sub-data packet acquisition module is used for applying a segmentation algorithm to the data packets to acquire more sub-data packet groups;

a data matrix reorganizing module, configured to reorganize a 3×30 channel state information matrix into a 30×32 data matrix by using a matrix smoothing algorithm for each sub-packet;

the two-dimensional reflection path information image generation module is used for carrying out two-dimensional MUSIC algorithm on the data matrix, calculating angle information and flight time information corresponding to all reflection paths and direct paths transmitted by the target position, and combining the angle information and the flight time information to generate a two-dimensional reflection path information image serving as a characteristic image;

the training module of the model is used for importing the multi-reflection path pictures at a plurality of positions into the network classification model for training, and generating a training model;

and the personnel positioning module is used for classifying the two-dimensional reflection path information image generated by the new position information by using the generated training model so as to position personnel.

In one embodiment, the system is installed on a computer configured with the linux14.04 system and installed with CSI tools (Channel State Information Tool, channel state information toolkit, CSITool is an open source toolkit for measuring channel state information CSI based on the 802.11n protocol), and the computer is configured to send and receive 2.4G/5G CSI (mainly referred to as WiFi frequency band, that is, a mobile phone supports 2.4G and 5G WiFi signals) signals.

By combining all the technical schemes, the invention has the advantages and positive effects that: the reflection path image obtained by using the phase information of the CSI is obviously different from the reflection path image obtained by using the phase information of the CSI in the prior art. The method provided by the invention does not change the phase with time. And this feature combines angle of arrival information and time of flight information, more of which are shown in the reflected path image. Unlike the system using AOA only for positioning, which needs to find out the direct information from a plurality of reflection path information, the more the spectrum peaks corresponding to the arrival angle and the flight time are, the more obvious the generated multi-reflection path image features are, which is more beneficial to the distinction of the reflection path images and the identification of the position features.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the disclosure and together with the description, serve to explain the principles of the disclosure;

FIG. 1 is a flowchart of an indoor personnel positioning method based on WiFi multi-reflection path images provided by an embodiment of the invention;

FIG. 2 is a schematic view of a laboratory environment 1 provided by an embodiment of the present invention to which the positioning method is applied;

FIG. 3 is a schematic view of a laboratory environment 2 using the positioning method according to an embodiment of the present invention;

FIG. 4 is a schematic diagram of a power divider connection provided by an embodiment of the present invention;

FIG. 5 is a schematic diagram of three antenna phases prior to the use of a linear phase correction algorithm according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of three antenna phases after using a linear phase correction algorithm according to an embodiment of the present invention;

FIG. 7 is a two-dimensional multi-reflection signal path image one combining angle of arrival and time of flight generated using an extended 2D-MUSIC algorithm provided by embodiments of the present invention;

FIG. 8 is a two-dimensional multi-reflection signal path image two combining angle of arrival and time of flight generated using an extended 2D-MUSIC algorithm provided by embodiments of the present invention;

FIG. 9 is a three-dimensional multi-reflection signal path image combining angle of arrival and time of flight generated using an extended 2D-MUSIC algorithm provided by embodiments of the present invention;

FIG. 10 is a schematic diagram of a confusion matrix of test results for laboratory environment 1 using convolutional neural network using the method of the present invention, provided by an embodiment of the present invention;

FIG. 11 is a graph of cumulative probability distribution obtained in laboratory environment 2 using the method of the present invention using two network configurations provided by an embodiment of the present invention.

Detailed Description

In order that the above objects, features and advantages of the invention will be readily understood, a more particular description of the invention will be rendered by reference to the appended drawings. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. The invention may be embodied in many other forms than described herein and similarly modified by those skilled in the art without departing from the spirit or scope of the invention, which is therefore not limited to the specific embodiments disclosed below.

Based on reflection path information of an object to a wireless signal, the embodiment of the invention provides an indoor personnel positioning method based on WiFi multi-reflection path images. The method comprises the steps of receiving and transmitting data through a software radio device or a receiving and transmitting CSI data device provided with a CSI Tool box, preprocessing the received CSI data, generating an image of reflection path information of all signals existing in an experimental environment by using a 2D-MUSIC (Multiple Signal Classification Algorithm, multiple signal classification) algorithm which is expanded to two dimensions, and positioning personnel indoors by distinguishing reflection path images of different positions. And model training of data is carried out before personnel positioning, so that high-precision personnel positioning is realized. The method can be realized according to the following steps: data acquisition is carried out on WiFi equipment carried by indoor personnel; performing a linear phase elimination algorithm, a data packet segmentation algorithm, a matrix smoothing algorithm and a 2D-MUSIC algorithm on the received data; generating a two-dimensional reflected path image (an unused feature image) using a 2D-MUSIC algorithm on the reorganized data channel state information (Channel State Information, CSI) matrix; training the reflection path information images at different positions by using a neural network training mode to obtain a training model; the received CSI data is then tested and location classified. The method combines the reflection path information image characteristics, so that the model after training realizes high-precision positioning during testing.

In embodiment 1, as shown in fig. 1, the embodiment of the invention provides an indoor personnel positioning method based on WiFi multi-reflection path images, which uses a brand new deep learning feature as an input of network training, and includes the following steps:

s101, receiving and transmitting wireless data by utilizing two three-antenna devices with modified designs;

s102, preprocessing the received CSI data, and eliminating phase errors caused by the space between receiving antennas;

s103, applying a segmentation algorithm to the data packet to obtain more sub-data packet groups;

in the data acquisition process, a data file sets the transmission rate of received data, the total length of a data packet and the transmission interval of the data during transmission. The total number of transmitted data packets of the data file used in the present invention is 1500 packets. The data required to generate the reflected path image generates an image for every 100 packets. In order to save trouble and time waste caused by collecting every 100 data packets during data collection, the length of each transmitted data file data packet is set to 1500 data packets, and each data file is divided into 100 data packets to be a file, so that the data file of 1 1500 data packets can obtain 15 data files of 100 data packets at most, and 14 images are generated after receiving and transmitting each time;

S104, using a matrix smoothing algorithm for each sub-data packet in the step S103 to reorganize the channel state information matrix of 3×30 into a data matrix of 30×32;

illustratively, the matrix smoothing algorithm includes: the received original CSI data is taken as a first row of a reset matrix by a 3X 30 matrix, wherein 1-16 subcarriers of a first antenna and 1-16 subcarriers of a second antenna of the original CSI data are intercepted; taking 2-17 subcarriers of a first antenna and 2-17 subcarriers of a second antenna of original CSI data as a second row of a reset matrix; the data passing through the first antenna and the second antenna are arranged through the same subcarrier; lines 16-30 are simultaneously composed of the same subcarriers through the second antenna and the third antenna;

s105, performing two-dimensional MUSIC algorithm on the data matrix passing through the step S104, calculating angle information and flight time information corresponding to all reflection paths and direct paths transmitted by the target position, and combining the angle information and the flight time information to generate a two-dimensional reflection path information image serving as a characteristic image;

s106, importing all reflection path images in a plurality of positions into a network classification model to train the network classification model; generating a training model;

Illustratively, importing all reflected path images at a plurality of locations into a network classification model for training of the network classification model includes:

the original image input 224×224, the pooling kernel size 3×3, and the dropout operation will set the output of each hidden layer neuron with probability less than 0.5 to 0, i.e. some neural nodes are removed, so as to prevent overfitting. The images are added with operations such as rotation and translation in the training process, so as to enhance the robustness of the training result. Network parameter learning optimization principle during training: the gradient descent principle is that the initial learning rate is set to 0.0001, the number of single packed images is set to 16, the iteration number is 15, and training of a GPU acceleration model is used;

the training model contains the characteristic information of all training position images, and the newly received test photo extraction characteristics and the data of the model can be compared and predicted to position;

and S107, classifying the two-dimensional reflection path information image generated by the new position information by using the training model generated in the step S106 so as to position personnel.

In step S102 of the embodiment of the present invention, a power distributor is used to connect one port of the transmitting end and three ports of the receiver, to calculate the phase differences between the second antenna, the third antenna and the first antenna of the receiving device, and to compensate for the phase differences by performing the linear addition and subtraction of the phases on the data collected subsequently.

In step S102 of the embodiment of the present invention, since the phases of the receiving antennas are different due to the different crystal oscillator frequencies generated by the switching on/off of the transmitting device each time, it is necessary to eliminate the phase influence by calculating the phase difference between the initial antennas. The linear phase elimination method takes the phase mainly used by the subsequent MUSIC algorithm as an access point, and the phase difference can seriously influence the precision of the MUSIC algorithm, so that the linear phase elimination method is necessary.

It will be appreciated that existing linear phase cancellation algorithms are based on the spacing between antennas and a calculated analog value, not the true value. The present invention, however, obtains a true value by using the power divider, namely: the actual measurement results in phase error, and the initial system phase difference measurement is shown in fig. 4.

In step S104 of the embodiment of the present invention, since the signal path information obtained by using the MUSIC algorithm is smaller than the number of antennas, the number relationship is as follows: the obtained signal path number is less than the antenna number, so the invention uses matrix smoothing algorithm to simulate and expand the three-antenna equipment into multi-antenna equipment, thereby achieving the purpose of obtaining multi-reflection signal path information. The image obtained in the 2D-MUSIC algorithm will reveal more path information.

In step S105 of the embodiment of the present invention, the spectrum estimation formula in the classical MUSIC algorithm is:

（4）

in the method, in the process of the invention,a steering vector representing the smoothing matrix; if X is the data matrix obtained after the smoothing process, +.>Covariance matrix for received data +.>Decomposed into->The method comprises the steps of carrying out a first treatment on the surface of the Wherein->The hermite matrix is positive +.>A is M x N dimensional steering matrix of the spatial array, i.e. Van der Waals matrix, p>The M eigenvalues obtained by the eigenvalue decomposition are all nonnegative real numbers +.>There are D large eigenvalues and M-D small eigenvalues, the space consisting of eigenvectors corresponding to the large eigenvalues ∈>The space is formed by feature vectors corresponding to small feature values and is a signal subspace; />Is the noise subspace of the signal,>is->Conjugate matrix of>Is the spectral value obtained by MUSIC algorithm.

In step S105 of the embodiment of the present invention, in order to obtain a two-dimensional MUSIC algorithm (2D-MUSIC) multi-reflection path image, it is necessary to raise the one-dimensional MUSIC algorithm to the two-dimensional MUSIC algorithmExpansion to->；

Wherein, the liquid crystal display device comprises a liquid crystal display device,

；

in the method, in the process of the invention,，/>，/>m is the number of antennas for the subcarrier frequency interval; />For a matrix of steering vectors carrying angle information and time of flight information +.>Values for each row of elements of the steering vector matrix;

The two-dimensional MUSIC algorithm spectrum estimation formula is:

；

in the method, in the process of the invention,to obtain angle information; />A steering vector representing a matrix obtained by smoothing the received signal, which contains angle-of-arrival information and time-of-flight information, ">Is the noise subspace of the signal,>steering vector representing smooth matrix covariance matrix, < +.>Is->Conjugate matrix of>Is the spectral value obtained by MUSIC algorithm.

And obtaining angle information and flight time information by a two-dimensional MUSIC algorithm spectrum estimation formula.

In step S105 of the embodiment of the present invention, a two-dimensional MUSIC algorithm is used to generate a multi-reflection path image using the smoothed CSI matrix for training a model.

In the embodiment of the present invention, in step S106, a machine learning model is constructed, the model is trained and stored by using the acquired spectrograms of a plurality of positions, and the test sample is input into the trained classification model to determine the final position.

Example 2 as another implementation of the present example, a computer configured with a linux14.04 system and installed with CSI Tool is provided, which can be used to implement the step S101-step S107 step commands of example 1, and which is configured with equipment available for transceiving 2.4G/5G CSI signals.

Embodiment 3, as another implementation manner of the embodiment of the present invention, the embodiment of the present invention provides an indoor personnel positioning method based on WiFi multi-reflection path image, in which information received by an Intel 5300 wireless network card through 802.11n protocol is called CSI, and the value is in a complex form, and includes channel information on each subcarrier of each antenna and overall attenuation and phase shift due to environmental introduction. Assume that N incoherent far-field narrowband signals are incident on an array of M antennas, where the received signals are:

（1）

wherein, the CSI value of the mth antenna is expressed as:

（2）

in the method, in the process of the invention,for the reception signals of the array,/->For transmitting signals +.>For M-dimensional noise data, < >>For spatial array->A dimension guide vector;

（3）

the CSI values composed of different antennas and different subcarriers in the formula (3) are expressed as the following general formulaThe CSI value of the nth subcarrier of the mth antenna is represented as a complex form +.>；/>The value is 1-3; />The value is 1-30.

Intel 5300The wireless network card is provided with three channels, and the channels complete the down-conversion process by combining the frequency generated by the crystal oscillator after receiving the signals. The down-conversion frequency is generated by the same crystal oscillator in the host, and is distributed to all channels after passing through the phase-locked loop, and the locked frequency has certain difference in the process of powering on again after the host is powered off. The error has a certain randomness, which causes an error in the initial phase between channels. Therefore, in order to accurately complete the estimation, it is necessary to cancel the initial phase error between the channels after the power-on. Taking a uniform linear array as an example, assume that the distance between two adjacent antennas isThe signal entrance angle is +.>，/>Wavelength is: />. Wherein->Resulting in the addition of the phase of the CSI value for the nth subcarrier。

The additional phase caused by the sample time offset (sampling time offset, STO) under the same WiFi network card is the same between antennas of a particular subcarrier. Since the error between the antennas of the receiving device cannot be directly obtained by the machine, the initial phase difference value between the different antennas of the receiving device can be calculated, the initial phase of 30 subcarriers of each antenna can be linearly compensated, and the second antenna, the third antenna and the first antenna of the receiver can be calibrated, so that the phases of the different antennas of the receiver can be corrected. But care should be taken: the feeders of the power splitter to the three different interfaces of the Wi-Fi NIC card need to be the same, which will introduce the same phase shift to the signals. And directly obtaining the phases of different Wi-Fi NIC interfaces, and calculating the phase difference between ports to perform phase compensation.

Since the received data is of a 3×30 matrix type, the relationship between the acquired maximum number of paths and the number of antennas is the acquired maximum number of paths=the number of antennas-1. In order to obtain more path information, the input CSI measurement values are subjected to structural reorganization to obtain a smoothed CSI matrix. The smooth CSI matrix generated by recombination directly obtains the information of all reflection propagation paths through a MUSIC algorithm.

In the embodiment of the present invention, the obtaining information of all reflection propagation paths through the MUSIC algorithm includes:

the spectrum estimation formula in the classical MUSIC algorithm is:

；

the method comprises the following steps of:

；

In the middle ofE is an autocorrelation matrix, H is a conjugate matrix,is->Is>The hermite matrix for positive is +.>A is the signal correlation matrix of the spatial array +.>Dimension-oriented matrix, p->The M eigenvalues obtained by the eigenvalue decomposition are all non-negative real +.>There are D large eigenvalues and +.>Small eigenvalues and space composed of eigenvectors corresponding to large eigenvalues +.>For signal subspace, space composed of eigenvectors corresponding to small eigenvalues +.>Is a noise subspace.

Embodiment 4, as another implementation manner of the embodiment of the present invention, an indoor personnel positioning method implementation manner based on WiFi multi-reflection path images provided by the embodiment of the present invention includes:

intercepting 1-16 subcarriers of a first antenna and 1-16 subcarriers of a second antenna of original CSI data as a first row of a reset matrix; taking 2-17 subcarriers of a first antenna and 2-17 subcarriers of a second antenna of original CSI data as a second row of a reset matrix; the data passing through the first antenna and the second antenna are arranged through the same subcarrier; lines 16-30 are simultaneously formed by the same subcarriers for the second antenna and the third antenna.

In a classical MUSIC algorithm, firstly, obtaining a noise subspace through space matrix decomposition to find an arrival angle, then, finding a path position corresponding to a spectrum peak value of a MUSIC spectrum as a direct path, and finding a signal end on the angle path by taking the arrival angle corresponding to the spectrum peak value of the spectrum as an angle of the direct path. However, this method has certain drawbacks, and the method can find the true angle of the known source. Since there are a large number of reflected signals in the actual environment, the intensity of the reflected signals may be far greater than that of the direct path signals, and finding the position of the actual signals by searching for the maximum spectral peak is not feasible, and the maximum angle corresponding to the spectral peak may be the path position of the reflected signals. Based on this idea the present invention proposes to locate a position using images generated from all direct and reflected signal path information corresponding to that position. The arrival angle and the flight time in the image correspond to a peak value, which is the information of different reflection paths. The method can cancel finding the direct path from the complex multi-reflection path, and avoids the complicated path finding process.

The spectrum estimation formula in the classical MUSIC algorithm is:

（4）

In the method, in the process of the invention,for the obtained MUSIC spectral values, +.>A steering vector representing the smoothed matrix; if X is the data matrix of the acquisition and reception, then +.>Covariance matrix for received data/>Decomposed into->The method comprises the steps of carrying out a first treatment on the surface of the Wherein->Hermite matrix positive, is +.>Is a signal correlation matrix of a spatial arrayDimension-oriented matrix, p->The eigenvalues obtained by the eigenvalue decomposition are all nonnegative real numbers, wherein the eigenvalue comprises D large eigenvalues and M-D small eigenvalues, and the space formed by eigenvectors corresponding to the large eigenvalues is->For signal subspace, space composed of eigenvectors corresponding to small eigenvalues +.>Is a noise subspace.

In order to obtain a 2D-MUSIC reflected path image, it is necessary to boost the one-dimensional MUSIC algorithm to the two-dimensional MUSIC algorithm, wherein,

；

in the method, in the process of the invention,，/>，/>m is the number of antennas for the subcarrier frequency interval; />For the time of flight of the p-th path, +.>For the angle information of the p-th reflection path, < >>For the phase of the m dimension of the p-th path, +.>P is the p-th reflection path,>is phase information, d is antenna spacing, +.>Wavelength of signal,/->Phase shift matrices for different paths;

；

In the method, in the process of the invention,to obtain angle information; />A steering vector representing a matrix obtained by smoothing the received signal, which contains angle-of-arrival information and time-of-flight information, ">Is the noise subspace of the signal,>is->Conjugate matrix of>Is the spectral value obtained by MUSIC algorithm.

The angle information and the flight time information can be obtained from the angle information. From a succession of 100 data packets drawn as a reflected path image, several relatively strong paths can be acquired.

In the embodiment of the invention, as the software radio equipment is expensive, a flat-replacing device which can also send the CSI data is used, namely, two associated desktop computers which are provided with a Linux 802.11n CSI Tool kit are used as a transmitter and a receiver. The method uses a 64-bit Ubuntu 14.04.03 LTS (Main cloud computing) operating system provided with an Intel 5300 network card (a network card model specified by a CSI software official), the reliability, performance and interactivity of a cloud platform and a telescopic environment are greatly improved, and meanwhile support and maintenance service for 5 years is provided.

One PC is configured as a transmitter, the other PC is configured as a receiver, all PCs work at 5GHz, and the signal bandwidth is 40MHz. The receiving device remains stationary and the transmitting device simulates a human body making a change in position.

In the embodiment of the invention, one index of the measurement result is an average error of the error identification coordinates which occur, and the average error is used for calculating the average value of the sum of errors of all classification error points which occur from the (1, 1) point to the (E, F) point;

average error；

In addition, the invention also carries out error comparison test through a 50% confidence interval error and an 80% confidence interval error respectively.

And (3) performing feature extraction training on the collected multi-reflection path images at a plurality of positions by using a neural network/machine learning model, storing the neural network/machine learning model, inputting a test sample into the trained Alexnet light-weight neural network and classification network, determining an action category according to the proportion of the prediction category, and determining the highest prediction occupancy rate as the current position category.

Embodiment 5, as another implementation manner of the embodiment of the present invention, the embodiment of the present invention provides an indoor personnel positioning method based on WiFi multi-reflection path images, including the following steps:

First, data acquisition: when the software defined radio or other alternative device can be used to perform the CSI data required for data acquisition training, the transmitter and receiver are desktops, and three receiving antennas must be configured. Because of the high cost of software radio, the invention uses a desktop host which can also transmit and receive CSI data and is provided with a commercial network card, the transceiver needs to be provided with a Linux 14.04.03 system, a CSI Tool kit is provided for transmitting and receiving data, and the kernel is replaced by a version 4.2.

When in transmitting test, the receiving and transmitting equipment needs to be adjusted to a monitoring mode, so that the invention is beneficial to changing the size and the length of data transmission, and a transmitting end sets the packet length and the size through a terminal command. The receiver may receive a data file containing the data used for the experiment. (note: transceiver devices should be at the same level) at multiple locations within the experimental environment for data acquisition and transmission. And the tester sends data at the appointed place by carrying the WiFi equipment, and the receiving equipment is arranged at a certain fixed point to receive the personnel position data information. For the requirement of transmitting data, the invention sets that 1500 packets are transmitted each time, and 30 times of data are transmitted at each position to collect the CSI information. The invention splits every 1500 packets, and generates a CSI data file every 100 packets according to the actual number of the packets, so that a plurality of reflection path images can be obtained by only transmitting and receiving data once. Experimental data of the present invention were obtained in two experimental environments as shown in fig. 2 and 3. The invention was tested in two laboratories, FIG. 2 is an open laboratory with two rows of table chairs in alignment and two laboratory machines; the laboratory in fig. 3 has two semi-annular iron tables, and there are many signal reflection paths due to the presence of the iron tables.

Second, data processing: the data file is a dat type file, and MATLAB software is needed to analyze the file once to obtain the CSI data matrix. The data matrix is of the matrix type 3 x 30. The method specifically comprises the following steps:

(1) The data is divided, and each data packet is intercepted into a file for every 100 data packets. Subsequent processing is then performed on every 100 packets. The correction of the phase is then performed for each file. The phase difference between the No. 2 antenna, the No. 3 antenna and the No. 1 antenna of the receiving antenna is calculated by connecting three channels of the receiving device and one channel of the transmitting device through the power distributor. The connection schematic diagram of the power divider is shown in fig. 4, the true value of the phase difference can be obtained by directly calculating the phase difference between the antennas, then the phase recovery can be performed by performing linear addition and subtraction of the phase on each 100-packet file, the influence of the phase deviation on the 2D-MUSIC algorithm is avoided, and the following is the phase comparison of the data before and after the phase correction, as shown in fig. 5 and 6.

Fig. 5 shows the original phases of the three antennas before the linear phase correction algorithm, and it can be seen that there is a significant phase difference between the phases of the three antennas, and in fig. 5, there are a total of 3×30 phase values, which are the phase values of 30 subcarriers of the three antennas, and this is the phase before the phase correction.

As shown in fig. 6, the processed phase data is obviously improved, and the phases of the three antennas are obviously overlapped, which shows that the phase of the device is effectively improved. In fig. 6, the total of 3×30 phase values is the phase values of 30 subcarriers of three antennas, and it can be seen that all 30 subcarriers of three antennas overlap together to indicate that the phases of 30 subcarriers of three antennas after correction remain consistent. This is the phase after the phase correction.

(2) And carrying out smooth expansion on the data of the received signal. Since the received device can only obtain the data of three antennas, in order to obtain more antenna data information, the invention expands the data matrix, and expands the data matrix of 3×30 into the data matrix of 30×32, the method is that the data is rearranged and combined, and the data matrix is changed as follows:

。

thirdly, spectrogram generation: the two-dimensional steering vector is obtained by using the CSI matrix expanded by the matrix smoothing algorithmWherein, the method comprises the steps of, wherein,

；

，/>，/>the subcarrier frequency interval is m, which antenna is the number of antennas.

It can be understood that the flight time of the signal is introduced into the MUSIC algorithm, one-dimensional data is expanded to a two-dimensional space, and the arrival angle information of the signal is obtained only through the MUSIC algorithm from the original information, so that the joint information of the arrival angle and the flight time is obtained. The abscissa information of each reflection path image is corresponding arrival angle and flight time information. The arrival angle and the flight time of each position correspond to a peak value, and the peak value is information of different reflection paths.

Obtaining a two-dimensional MUSIC algorithm after expansion:

；

in the method, in the process of the invention,is the noise subspace of the signal,>steering vector representing smooth matrix covariance matrix, < +.>Is->Conjugate matrix of>The spectrum value obtained by the MUSIC algorithm is obtained by applying a two-dimensional MUSIC algorithm to the CSI matrix after smooth expansion, so that a two-dimensional MUSIC (2D-MUSIC) multi-reflection path image can be obtained, and three image information obtained by one position can be obtained as shown in fig. 7-9. The three reflected paths obtained by the 2D-MUSIC algorithm are shown in fig. 7-9, where the peaks are shown as one path, and two distinct paths and one weaker path can be seen. The three images of FIGS. 7-9 are 1500 dataOf the 15 images obtained, 3 were wrapped and one data file at each location could be seen to be very similar.

Fourth, model training classification: in order to verify the feasibility of the image method, the invention uses two methods of machine learning algorithm/neural network to carry out comparison test when constructing the classification model. The following invention is tested in a manner using an Alexnet lightweight neural network and a classification network support vector machine SVM. Firstly, inputting a characteristic image, dividing a data set into a proportion of 7:3, using 70% of data for training, using 30% of data for testing, optimizing a model by using a random gradient descent algorithm (Stochastic Gradient Descent Algorithm, SGDM) of momentum, and using a GPU to accelerate the training of the model. And obtaining the final accuracy through comparing the predicted label with the real label.

Embodiment 5 of the present invention provides an indoor personnel positioning system based on WiFi multi-reflection path image, including:

the three-antenna device is used for receiving and transmitting wireless data;

the sub-data packet acquisition module is used for applying a segmentation algorithm to the data packet to acquire more sub-data packets;

the training module of the model is used for importing pictures of all reflection paths at a plurality of positions into the network classification model for training, and generating a training model;

In the foregoing embodiments, the descriptions of the embodiments are emphasized, and in part, not described or illustrated in any particular embodiment, reference is made to the related descriptions of other embodiments.

The content of the information interaction and the execution process between the devices/units and the like is based on the same conception as the method embodiment of the present invention, and specific functions and technical effects brought by the content can be referred to in the method embodiment section, and will not be described herein.

It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions. The functional units and modules in the embodiment may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit, where the integrated units may be implemented in a form of hardware or a form of a software functional unit. In addition, the specific names of the functional units and modules are only for distinguishing from each other, and are not used for limiting the protection scope of the present invention. For specific working processes of the units and modules in the system, reference may be made to corresponding processes in the foregoing method embodiments.

Based on the technical solutions described in the embodiments of the present application, the following application examples may be further proposed.

According to an embodiment of the present application, there is also provided a computer apparatus including: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, which when executed by the processor performs the steps of any of the various method embodiments described above.

Embodiments of the present application also provide a computer readable storage medium storing a computer program which, when executed by a processor, performs the steps of the respective method embodiments described above.

The embodiment of the application also provides an information data processing terminal, which is used for providing a user input interface to implement the steps in the method embodiments when being implemented on an electronic device, and the information data processing terminal is not limited to a mobile phone, a computer and a switch.

The embodiment of the application also provides a server, which is used for realizing the steps in the method embodiments when being executed on the electronic device and providing a user input interface.

Embodiments of the present application also provide a computer program product which, when run on an electronic device, causes the electronic device to perform the steps of the method embodiments described above.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the present application may implement all or part of the flow of the method of the above-described embodiments, and may be implemented by a computer program to instruct related hardware, where the computer program may be stored in a computer readable storage medium, and the computer program may implement the steps of the method embodiments described above when executed by a processor. Wherein the computer program comprises computer program code which may be in source code form, object code form, executable file or some intermediate form etc. The computer readable medium may include at least: any entity or device capable of carrying computer program code to a photographing device/terminal apparatus, recording medium, computer Memory, read-Only Memory (ROM), random access Memory (Random Access Memory, RAM), electrical carrier signals, telecommunications signals, and software distribution media. Such as a U-disk, removable hard disk, magnetic or optical disk, etc.

To further demonstrate the positive effects of the above embodiments, the present invention was based on the above technical solutions to perform the following experiments.

Analysis of experimental results: by testing the final data, the experimental data result of the confusion matrix shown in fig. 10 is obtained, and according to the confusion matrix, the data at all positions can be seen to realize 90% of testing accuracy, so that the usability of the 2D-MUSIC multi-reflection path image method can be illustrated.

As shown in fig. 11, a probability distribution graph is obtained using an Alexnet network and an SVM classification structure, wherein the lower part represents a curve obtained using the SVM structure and the upper part represents an arrival curve using the Alexnet neural network. It can be seen that the two networks have about 1.5m of 50% error, and the difference of classification results obtained by using the reflection path image method is very little, so that good accuracy and similar test errors can be obtained even if different networks are used.

Table 1 shows the data obtained using two different classification algorithms, wherein the average error calculation is obtained by the following formula:

Average error；

Where N represents the number of all errors that occur.Representative is the calculation of the running sum. The formula means: an average value of the sum of errors of all the classification error points occurring from the (1, 1) point to the (E, F) point is calculated. />Is the position where the data are actually collected, N is the number of misclassifications, E and F are respectively setMaximum value of the abscissa.

In the experimental process of the invention, not only the average error is used as a standard of the test precision, but also a 50% confidence interval error is used as a standard test of the training result. The data in table 1 more intuitively shows that the position estimation obtained by the method using the 2D-MUSIC reflected signal image still reaches an acceptable range even if different networks are used, and different learning models can reach similar test levels, thereby proving the usability of the method. From table 2, it can be seen that the test accuracy of 93% and above can be achieved under two experimental environments by using the method through two different classification models, which is equivalent to the accuracy of the current mainstream fingerprint positioning method, and the usability of the method is illustrated.

Table 1 test accuracy using convolutional neural network using the method of the present invention in laboratory environment 1

。

Table 2 test accuracy for two network structures using the method of the present invention in two different laboratory environments

。

In the experimental process of the invention, not only the average error is used as a standard of the test precision, but also the errors of the 50% confidence interval and the 80% confidence interval are used as standard tests of the training result. The data in table 1 more intuitively shows that the position estimation obtained by the method using the 2D-MUSIC reflected signal image still reaches an acceptable range even if different networks are used, and different learning models can reach similar test levels, thereby proving the usability of the method. It can be seen from table 2 that the test accuracy of 93% and above can be achieved under two experimental environments by using the method through two different classification models, and the method has the same precision as the current mainstream fingerprint positioning method, thus demonstrating the usability of the method.

The invention can provide convenience for detection and activity recognition of users with inconvenient actions. The technical scheme of the invention combines the MUSIC algorithm and the fingerprint positioning scheme, and provides a brand new solution for the indoor positioning method. The original positioning method based on the MUSIC algorithm is to perform triangular positioning by searching the direct path angle, and the traditional fingerprint positioning method based on the signal strength is unstable, and the signal strength is easy to change along with the change of time; the invention combines the MUSIC algorithm with the fingerprint, and does not use the traditional triangle positioning and signal intensity positioning method any more, but the image of the reflection path is hardly changed due to almost unchanged indoor environment, and the image of the reflection path is more stable by the fingerprint method.

While the invention has been described with respect to what is presently considered to be the most practical and preferred embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but on the contrary, is intended to cover various modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

Claims

1. The indoor personnel positioning method based on the WiFi multi-reflection path image is characterized by comprising the following steps of:

s3, applying a segmentation algorithm to the data packet in the preprocessed CSI data to obtain more sub-data packet groups;

2. The indoor personnel positioning method based on the WiFi multi-reflection path image according to claim 1, wherein in step S2, preprocessing the received CSI data includes: the power distributor is used for connecting a transmitting end and a receiver, phase differences among a second antenna, a third antenna and a first antenna of receiving equipment are calculated respectively, and the phase differences are eliminated by performing linear addition and subtraction of the phases on collected data;

in step S3, applying a segmentation algorithm to the data packet in the preprocessed CSI data, obtaining a sub-data packet includes: in the data acquisition process, a data file sets the transmission rate of received data, the total length of a data packet and the transmission interval of the data when transmitting; the total number of data packets transmitted by the data file is 1500 packets; generating an image for every 100 packets of data required to generate the reflected path image; the data file packet length of each transmitted data file is set to 1500 packets, and each data file is divided into 100 packets and one file, so that 15 data files with 100 data packets are obtained.

3. The indoor personnel positioning method based on the WiFi multi-reflection path image according to claim 1, wherein in step S4, the matrix smoothing algorithm performs analog expansion of the three-antenna device into the multi-antenna device includes: the multi-reflection signal path information is obtained by reorganizing a 3×30 channel state information matrix into a 30×32 data matrix using a matrix smoothing algorithm.

4. The indoor personnel location method based on WiFi multi-reflection path images according to claim 3, wherein obtaining multi-reflection signal path information includes: n incoherent far-field narrowband signals are incident on an array consisting of M antennas, and the expression of the received signals is:

；

wherein, the CSI value of the mth antenna is expressed as:

；

in the method, in the process of the invention,for the reception signals of the array,/->For spatial array->A vector matrix is guided in a dimension;

；

in the formula, the CSI values formed by different root antennas and different subcarriers are expressed as the general formulaThe CSI value of the nth subcarrier of the mth antenna is represented as a complex form +. >；/>The value is 1-3; />The value is 1-30;

5. The indoor personnel positioning method based on the WiFi multi-reflection path image according to claim 4, wherein obtaining information of all reflection propagation paths through MUSIC algorithm includes:

the spectrum estimation formula in the classical MUSIC algorithm is:

；

the method comprises the following steps of:

；

wherein E represents an autocorrelation matrix, H represents a conjugate matrix,is->Is>Hermitian matrix to positiveA is the signal correlation matrix of the spatial array +.>Dimension-oriented matrix, p->The M eigenvalues obtained by the eigenvalue decomposition are all nonnegative real numbers +.>There are D large eigenvalues and +.>Small eigenvalues and space composed of eigenvectors corresponding to large eigenvalues +.>For signal subspace, space composed of eigenvectors corresponding to small eigenvalues +.>Is a noise subspace.

6. The indoor personnel positioning method based on the WiFi multi-reflection path image according to claim 1, wherein in step S5, calculating angle information and flight time information corresponding to all reflection paths and direct paths emitted by the target position using a two-dimensional MUSIC algorithm includes:

；

In the method, in the process of the invention,for a matrix of steering vectors carrying angle information and time of flight information +.>For the phase shift matrix of the different paths, +. >For the time of flight of the p-th path, +.>，/>，/>For subcarrier frequency spacing, m is the number of antennas, +.>For the angle information of the p-th reflection path, < >>For the phase of the m dimension of the p-th path, +.>P is the p-th reflection path,>is phase information, d is antenna spacing, +.>Is the wavelength of the signal;

；

7. The indoor personnel positioning method based on WiFi multi-reflection path image according to claim 1, wherein in step S6, importing pictures of all reflection paths in a plurality of positions into a network classification model for training includes: testing by using an Alexnet lightweight neural network and a classification network Support Vector Machine (SVM); firstly inputting a characteristic image, dividing a data set to divide the data into a proportion of 7:3, using 70% of the data for training, using 30% of the data for testing, optimizing an Alexnet lightweight neural network and a classification network by using a momentum random gradient descent algorithm SGDM, using a GPU (graphic processing unit) acceleration model for training, and comparing a prediction label with a real label;

8. The indoor personnel positioning method based on WiFi multi-reflection path image according to claim 1, wherein classifying the two-dimensional reflection path information image generated with the generated training model in step S7 includes:

performing feature extraction training on the collected multi-reflection path images at a plurality of positions by using a neural network/machine learning model, storing an Alexnet lightweight neural network and a classification network, inputting a test sample into the trained Alexnet lightweight neural network and classification network, determining the positions according to the proportion of the predicted categories, testing 15 pictures generated by data collection once by using a majority voting principle, and classifying the 15 pictures into the position with the highest proportion and predicting the position as a final position; the average error of the erroneously identified coordinates is used to analyze the performance of the system;

The expression of the average error of the error recognition coordinates is:

average error；

In the method, in the process of the invention,representing the calculation of the cumulative sum +.>Is the position estimated by the network model, +.>Is actually collecting dataThe position, N, is the maximum value of the abscissa of the position where E, F are set, respectively, for the number of misclassifications.

9. An indoor personnel positioning system based on a WiFi multi-reflection path image, wherein the indoor personnel positioning method based on the WiFi multi-reflection path image according to any one of claims 1 to 8 is implemented, the system includes:

the three-antenna device is used for receiving and transmitting wireless data;

10. The WiFi multi-reflection path image based indoor personnel positioning system according to claim 9, wherein the system is mounted on a computer configured with a linux14.04 system and installed with CSI Tool, the computer being configured with a device for transceiving 2.4G/5G CSI signals.