CN110337066B - Indoor personnel activity identification method based on channel state information and man-machine interaction system - Google Patents

Indoor personnel activity identification method based on channel state information and man-machine interaction system Download PDF

Info

Publication number
CN110337066B
CN110337066B CN201910425638.2A CN201910425638A CN110337066B CN 110337066 B CN110337066 B CN 110337066B CN 201910425638 A CN201910425638 A CN 201910425638A CN 110337066 B CN110337066 B CN 110337066B
Authority
CN
China
Prior art keywords
activity
data
csi
personnel
indoor
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910425638.2A
Other languages
Chinese (zh)
Other versions
CN110337066A (en
Inventor
王勇
丁建阳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN201910425638.2A priority Critical patent/CN110337066B/en
Publication of CN110337066A publication Critical patent/CN110337066A/en
Application granted granted Critical
Publication of CN110337066B publication Critical patent/CN110337066B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/309—Measuring or estimating channel quality parameters
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
    • H04W16/18—Network planning tools
    • H04W16/20—Network planning tools for indoor coverage or short range network deployment
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/029—Location-based management or tracking services
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30—Services specially adapted for particular environments, situations or purposes
    • H04W4/33—Services specially adapted for particular environments, situations or purposes for indoor environments, e.g. buildings

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Quality & Reliability (AREA)
  • Physics & Mathematics (AREA)
  • Electromagnetism (AREA)
  • Complex Calculations (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

The invention belongs to the technical field of wireless communication, and discloses an indoor personnel activity identification method and a man-machine interaction system based on channel state information; preprocessing the collected CSI data to extract environment change information including correlation coefficients and variances; sensing the surrounding environment by using a decision tree, monitoring the activity of the personnel, continuously collecting CSI data if the activity of the personnel is not detected, keeping the activity monitoring state, and triggering an activity feature extraction module if the activity of the personnel is not detected; carrying out extreme value removing processing and low-pass filtering processing on the acquired CSI data in the module; and then carrying out activity matching and recognition of the heaviest personnel activity by using a dynamic time programming algorithm based on principal component analysis. According to the invention, the indoor personnel activity monitoring and activity recognition are combined, and the indoor personnel activity recognition is carried out through PCA-DTW, so that the step of offline acquisition data training in the prior art is omitted, and the system performance and the recognition rate are improved.

Description

Indoor personnel activity identification method based on channel state information and man-machine interaction system
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to an indoor personnel activity identification method based on channel state information and a man-machine interaction system.
Background
Currently, the closest prior art: e-eyes and WiFinger, and WiGest and CARM.
The prior art has the defects that: e-eyes and WiFinger can not effectively adapt to different environments or need to correct system parameters in different environments, the transportability is poor, and the identification efficiency is reduced. This is because E-eyes and WiFinger lack matching models that can quantitatively correlate CSI statistics with indoor human activity; WiGest and CARM are very sensitive to the effects of environmental noise and indoor channel variations when performing activity recognition. This is because they require capturing small changes in activity to be able to perform the matching of actions, but the captured CSI data also contains noise components. When the noise component is large or the channel changes, the activity recognition effect is poor and the recognition rate is low.
The scheme solves the defects that: 1. a matching model which quantitatively associates the CSI statistical characteristics with the indoor personnel activities is established, the method can be effectively suitable for different environments, set parameters do not need to be changed, and the activity recognition efficiency is improved. 2. The anti-interference capability of the system to environmental noise and indoor channel change is improved, and the anti-interference capability comprises extreme value processing, low-pass filtering and the like, so that the activity recognition rate is improved.
Currently, the closest prior art: in recent years, with the rapid development of wireless communication technology and the great popularization of WiFi devices, passive indoor activity recognition has attracted extensive attention in many fields such as human-computer interaction, security monitoring, old people care, emergency rescue, and the like, and has become a research hotspot in the technical field of wireless communication. The existing technologies applied to indoor personnel Identification are mainly based on computer vision, infrared and Radio Frequency Identification (RFID), and these methods can achieve expected effects, but have certain application defects. They can only work in a single scenario or require special sensor equipment and are costly. If widely deployed WiFi infrastructure can be utilized, not only can the cost be saved and the use be facilitated, but also the wide application can be realized. This approach does not require the deployment of a specific sensor network nor does it require special equipment to be worn by the personnel in the room. In a conventional sensor network, a large number of sensor nodes are deployed in a specific area, and the sensor nodes are responsible for sensing signals and transmitting collected signals through a special channel. The method not only consumes a great deal of manpower to deploy the equipment, but also consumes a great deal of labor and needs long-term maintenance, so that the method is not suitable for wide deployment. Conventional indoor personnel activity identification techniques do not meet the high efficiency, low cost and increasing demand. Due to the rapid development of wireless communication technology, WiFi networks are widely available, and indoor activity identification using WiFi networks has become a research hotspot.
At present, a physical quantity characteristic commonly used by a passive human body recognition system designed by using a WIFI Signal and irrelevant to equipment is Received Signal Strength (RSSI), which is convenient to obtain, but the RSSI is the superposition of Signal strengths of a plurality of paths and is greatly influenced by other noises in the environment. For example, the technology utilizes the abnormal fluctuation of the RSSI to capture the environmental change to realize human body identification, and has the defect of unreliable identification result when performing activity identification, because the activity identification needs to detect the environmental characteristic change situation in real time, and the RSSI comes from a Media Access Control (MAC) layer, which is the superposition of the signal intensity of multiple paths, and the influence of multipath and noise in an indoor environment can cause the RSSI accuracy to be seriously interfered, thereby having strong randomness and dynamic property. Channel State Information (CSI) is finer-grained physical layer Information, and Channel Information of multiple Orthogonal Frequency Division Multiplexing (OFDM) subcarriers can be measured simultaneously, so that more accurate detection can be provided for wireless environment sensing, and applications developed by using CSI, such as gesture detection, activity recognition, indoor positioning, security monitoring, and the like, are increasing. These applications are helpful for child and geriatric care, smart home, and security monitoring. The CSI may provide a channel frequency response, including amplitude and phase, for each subcarrier. By studying some of the unique features exhibited in CSI, the basic activities of a person, even gestures and typed keys, can be identified.
Currently, CSI-based indoor activity recognition techniques have been proposed in large numbers. Some functions are like E-eyes and WiFinger. The E-eye performs activity recognition by using the amplitude characteristics extracted by the CSI and a related matching algorithm (including Dynamic Time Warping (DTW) and bulldozer Distance (EMD)), and compares the amplitude characteristics with the related characteristics of known activities. WiFinger is a lightweight algorithm for gesture behavior recognition using Discrete Wavelet Transform (DWT) and DTW algorithms. Therefore, WiFinger can recognize gestures by detecting unique features exhibited in CSI. Although E-eyes and wifi finger can guarantee a certain recognition accuracy, they are limited by the lack of a model that can quantitatively correlate CSI statistics and indoor human activities. Still other indoor human activity recognition, such as WiGest and CARM, model the propagation of WiFi signals in physical space under indoor human activity. WiGest utilizes the effect of hand motion on wireless signals to recognize gestures. CARM uses quantitative correlation as an analysis mechanism and identifies it by activity matching. Their main advantage is that they both establish a quantitative correspondence between CSI statistics and indoor human activity. Unfortunately, they are also very sensitive to indoor environmental noise and variations.
In summary, the problems of the prior art are as follows: the existing indoor personnel activity recognition technology lacks a matching model capable of quantitatively correlating CSI statistical characteristics with indoor personnel activities, so that a system cannot be well adapted to different environments, or parameters need to be reset when the system is transplanted to different environments, namely, the environment adaptability is poor; the method is very sensitive to the influence of environment random noise and indoor channel variation, so that the activity recognition effect is poor and the recognition rate is low.
The difficulty of solving the technical problems is as follows: how to effectively utilize the statistical characteristics of the CSI data to establish a matching model and how to solve the influence of identifying process environmental noise and indoor channel variation.
The significance of solving the technical problems is as follows: the indoor personnel identification has very important significance for production and life, and can be widely applied to places such as families, markets, hospitals and the like. The technical problem is solved, and the passive indoor personnel activity recognition can meet the requirements of low cost and high precision.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides an indoor personnel activity identification method based on channel state information and a man-machine interaction system.
The invention is realized in such a way that the indoor personnel activity identification method based on the channel state information preprocesses the collected CSI data to extract the environment change information including the correlation coefficient and the variance; then, sensing the surrounding environment by using a decision tree to monitor the activities of the personnel; secondly, if no personnel activity is detected, continuously collecting CSI data and keeping the activity monitoring state, otherwise triggering an activity feature extraction module, and carrying out extreme value removal processing and low-pass filtering processing on the collected CSI data in the activity feature extraction module; and then carrying out activity matching and recognition of the heaviest personnel activity by using a dynamic time programming algorithm based on principal component analysis.
Further, the indoor personnel activity identification method based on the channel state information specifically comprises the following steps:
the method comprises the steps that firstly, Channel State Information (CSI) is collected through existing WiFi equipment;
secondly, preprocessing the collected CSI data, and respectively extracting correlation coefficients and variances among subcarriers;
thirdly, constructing a decision tree based on the correlation coefficient and the variance, monitoring personnel activities by using the decision tree, triggering an indoor personnel activity identification module when monitoring the indoor personnel activities, and carrying out the fourth step, otherwise, continuously collecting CSI data and keeping monitoring of the personnel activities;
fourthly, further processing the collected CSI original data to obtain activity characteristic information, firstly removing extreme values in the CSI data by using an extreme value detection algorithm based on K-means clustering, and then removing environmental noise and high-frequency components in the CSI data by using a low-pass filter;
fifthly, extracting activity principal component information through a dynamic time programming matching algorithm based on principal component analysis, and then performing activity matching;
and sixthly, identifying the indoor personnel activities, then ending the activity identification process, entering the next personnel activity identification state, and returning to the first step.
Further, the step one of acquiring channel state information CSI through the existing WiFi device specifically includes:
at time t, acquiring channel state information H of the l-th link between the transmitting terminal TX and the receiving terminal TR, including subcarrier information of multiple groups of links:
Figure BDA0002067397790000041
wherein [ ·]TDenotes transposition, NsubIs the number of OFDM subcarriers, g represents all the links, h (i) can be expressed as:
H(i)＝||H(i)||ej∠H(i)；
wherein h (i) is channel state information of the ith subcarrier, | h (i) and ═ h (i) represent amplitude and phase information, respectively;
selecting one of the links from
Figure BDA0002067397790000051
The acquisition of continuous unprocessed CSI is started and then passed through a sliding window of length m, denoted as:
Figure BDA0002067397790000052
wherein the content of the first and second substances,
Figure BDA0002067397790000053
is a number NsubXm matrix, t is the start index of the sliding window;
in the time direction from
Figure BDA0002067397790000054
The ith subcarrier is extracted, and is expressed as:
Figure BDA0002067397790000055
wherein
Figure BDA0002067397790000056
Is a 1 x m matrix, ExsubRaw-i(. represents a slave matrix
Figure BDA0002067397790000057
The ith subcarrier is extracted.
Further, the step of preprocessing the collected CSI data to extract environment change information, that is, to extract correlation coefficients and variances between subcarriers, specifically includes:
(1) preprocessing original CSI data;
1) calculating the variance of the CSI subcarriers;
the variance of each subcarrier over the sliding window constitutes NsubA x 1 vector, expressed as;
Figure BDA0002067397790000058
wherein the content of the first and second substances,
Figure BDA0002067397790000059
viis the variance of the ith subcarrier, and Var (-) is the variance symbol;
secondly, measuring the average variance of all subcarriers;
V＝Var(Vw)；
wherein V represents the degree of channel fluctuation;
2) calculating correlation coefficients among the CSI subcarriers;
for NsubSub-carriers, measuring the correlation coefficient as follows;
Figure BDA00020673977900000510
where ρ isi，jRepresenting the correlation coefficient between the ith subcarrier and the jth subcarrier, wherein Corr (-) is a correlation symbol;
repeating the above calculation to obtain a group of correlation coefficients, performing extreme point detection through the least median square, and screening out incorrect correlation coefficients, wherein the definition of the objective function is as follows:
Figure BDA0002067397790000061
wherein:
Figure BDA0002067397790000062
ρi,j＝{ρi,j,i<j∈[1,Nsub]}；
Γ＝Nsub(Nsub-1)/2；
where min (-) and med (-) are min and median operations, respectively, ri,jIs the remainder of the process,
Figure BDA0002067397790000063
is the least mean square estimate, and Γ is ρi，jThe number of (2);
(2) outlier detection was as follows:
Figure BDA0002067397790000064
Figure BDA0002067397790000065
Figure BDA0002067397790000066
Figure BDA0002067397790000067
after incorrect correlation coefficients are determined, they are filtered out and the remaining coefficients are then averaged as a final estimate of a set of correlation coefficients:
Figure BDA0002067397790000068
where ρ and I (ρ)i,j) Respectively is the final phaseAnd finally obtaining a correlation coefficient rho for identifying the indoor personnel activities by using the correlation coefficient and the outlier index.
Further, the third step is to construct a decision tree based on the correlation coefficient and the variance, monitor the activity of the people by using the decision tree, trigger the indoor people activity recognition module after monitoring the activity of the indoor people, and otherwise continue to collect CSI data and maintain the activity monitoring state, and specifically includes:
according to the result of the second step, on the basis of environmental change, namely subcarrier correlation coefficient and variance, constructing a decision tree and monitoring personnel activities by using the decision tree, triggering an indoor personnel activity identification module after monitoring the indoor personnel activities, and otherwise, continuously collecting CSI data and keeping an activity monitoring state;
indoor human activity monitoring with correlation coefficients and variances is expressed as:
Figure BDA0002067397790000071
further, the fourth step is to further process the collected CSI raw data to obtain activity characteristic information, remove extreme values in the CSI data by using an extreme value detection algorithm based on K-means clustering, and remove environmental noise and high-frequency components in the CSI data by using a low-pass filter;
(1) the K-means clustering method is very sensitive to abnormal values, finds strongly related data point groups, and detects the abnormality to find data points which are not strongly related to other data points; the extreme point processing of the original CSI data comprises the following steps;
1) the data was subjected to a normalization operation, expressed as:
Figure BDA0002067397790000072
wherein
Figure BDA0002067397790000073
h is the output of the data normalization, Z-score (. cndot.) is the data normalization operationMaking a symbol;
2) randomly selecting k data points from a data set h as an original centroid:
u＝{u1，u2，...uω...，uk},(ω＝1,2,...,k)；
3) the maximum number of iterations is set to Ω, and the iterations are from 1 to Ω, as follows:
Figure BDA0002067397790000074
wherein C isχIs the original cluster;
3.1) calculating data points
Figure BDA0002067397790000075
And each centroid uωThe distance between
Figure BDA0002067397790000076
Figure BDA0002067397790000077
3.2) Next step, labeling
Figure BDA0002067397790000078
Category corresponding as minimum
Figure BDA0002067397790000079
At this time, the cluster is updated, which is expressed as:
Figure BDA00020673977900000710
3.3) for all data points Cω，ω∈[1，k]Recalculating new centroids:
Figure BDA00020673977900000711
3.4) if the k centroids buzai2 are changed, entering the step four;
4) the final C cluster outputs are:
C＝{C1,C2,...,Ck}；
5) after clustering, extremum detection in CSI data:
u＝{u1，u2，...uε...，uk},ε∈[1，k]；
C＝{C1,C2,...Cε...,Ck},ε∈[1,k]；
where u is the final cluster centroid and C is the corresponding cluster:
Figure BDA0002067397790000081
wherein
Figure BDA0002067397790000082
Distance error between the centroid and the data point in the corresponding cluster;
5.1) Final, comparison
Figure BDA0002067397790000083
And the known threshold ζ, expressed as:
Figure BDA0002067397790000084
5.2) after the extreme value detection algorithm based on the K-means clustering is processed, obtaining:
Figure BDA0002067397790000085
where EVD (-) is an extremum detecting opcode;
(2) the low-pass filter eliminates random environmental noise and high-frequency components caused by any non-human activities; meanwhile, the special statistical characteristics caused by the human activities in the CSI data are reserved, and the statistical characteristics are obtained through a low-pass filter:
Figure BDA0002067397790000086
where LPF (-) is the low pass filter operator.
Further, the fifth step of extracting activity features, namely activity principal component information, by a dynamic time programming matching algorithm based on principal component analysis, and then performing activity matching specifically comprises:
(1) the step of extracting the active principal component information in the CSI data is as follows;
1) collecting filtered CSI, i.e.
Figure BDA0002067397790000087
Will be provided with
Figure BDA0002067397790000088
Dividing into data blocks containing 2 second intervals, arranging the data blocks of g links in rows to form a matrix
Figure BDA0002067397790000089
2) Correlation estimation, calculating a correlation matrix with dimension m × m:
Figure BDA0002067397790000091
3) and (3) feature decomposition, namely performing feature decomposition on the correlation matrix, and calculating a feature vector:
Figure BDA0002067397790000092
wherein
Figure BDA0002067397790000093
υiIs a feature vector, sigma is the number of principal components to be obtained, deceigen (·) is a feature decomposition operation symbol;
4) and (3) reconstructing an activity signal, wherein the main components are constructed as follows:
Figure BDA0002067397790000094
Figure BDA0002067397790000095
wherein
Figure BDA0002067397790000096
And
Figure BDA0002067397790000097
are respectively the τ -th eigenvector and the τ -th component, PiIs a principal component matrix;
(2) according to the characteristics of indoor personnel activities, carrying out similarity measurement on two activity characteristics, namely unknown activities and activity characteristics in a known database by adopting dynamic time warping; the specific algorithm is as follows:
1) given two activity characteristics, an unknown PiA is known
Figure BDA0002067397790000098
Then find a function of the point-to-point distance:
Figure BDA0002067397790000099
2) finding the correspondence between the points:
Figure BDA00020673977900000910
wherein:
Figure BDA00020673977900000911
Figure BDA00020673977900000912
then, considering ψ (w), the solution for the cumulative distance between two active features is as follows:
Figure BDA00020673977900000913
3) the final DTW estimate is to find the optimum ψ (w) that minimizes the cumulative distance:
Figure BDA00020673977900000914
further, step six carries out indoor personnel's activity discernment, after personnel's activity discernment, ends this identification process, enters next personnel's activity discernment state, will return to step one promptly and specifically include:
by computing the unknown activity features and the activity features in the pre-constructed database, the process of human activity recognition is as follows:
Figure BDA0002067397790000101
Figure BDA0002067397790000102
where argmin (·) is the minimum operator symbol and κ represents the category of human activity. Only when
Figure BDA0002067397790000103
When the minimum is taken, the activity of the person is correctly identified.
The invention also aims to provide a human-computer interaction control system applying the indoor human activity identification method based on the channel state information.
Another object of the present invention is to provide an intelligent home control system using the method for identifying indoor human activities based on channel state information.
In summary, the advantages and positive effects of the invention are: the invention solves the problems that the existing indoor personnel activity identification technology is easily influenced by environmental noise and channel change and the statistical characteristics of CSI data are effectively utilized. The implementation scheme is as follows: collecting CSI data by using WiFi equipment; and preprocessing the CSI data, and extracting environment change information, namely subcarrier correlation coefficients and variances. Sensing the environment of a target area by using a decision tree through environment change information, monitoring personnel activities, triggering a personnel activity feature extraction module if the human activities are monitored, and continuing to collect CSI data and keeping an activity monitoring state if the human activities are monitored; if the activity exists in the target area, performing data processing on the collected original CSI data to obtain activity characteristic information, namely principal component information, firstly removing extreme values in the data by using an extreme value detection algorithm based on K-means clustering, and then removing environmental noise and high-frequency components in the data by using a low-pass filter; and (3) performing activity feature extraction and personnel activity matching by using a Principal Component Analysis based on Dynamic Time programming matching algorithm (PCA-DTW). And identifying the indoor personnel activities, finishing the identification process after identifying the personnel activities, and entering the next personnel activity identification state.
The invention overcomes the defect of low reliability of the recognition result in the prior art and improves the activity recognition precision. Because the invention adopts the environment change information and the indoor personnel activity monitoring method based on the decision tree, the invention can sense the indoor environment and monitor the activity change of the indoor personnel, thereby avoiding the system from missing the activity.
The invention carries out extreme value elimination and low-pass filtering in the CSI data processing stage, screens and de-noizes the CSI data, overcomes the defects of low detection accuracy and poor reliability caused by data original errors, and improves the identification rate.
According to the invention, the indoor personnel activity monitoring and activity recognition are combined, and the indoor personnel activity recognition is carried out through PCA-DTW, so that the step of offline acquisition data training in the prior art is omitted, and the system performance and the recognition rate are improved.
The following table shows the comparison of the present invention with the prior art
Method Activity monitoring Extremum processing Filtering process Principal component analysis Matching
WiAR Is provided with Is provided with Is provided with Is provided with Is provided with
E-eyes Is free of Is free of Is provided with Is free of Is provided with
WiFinger Is free of Is free of Is provided with Is free of Is provided with
Drawings
Fig. 1 is a flowchart of an indoor human activity identification method based on channel state information according to an embodiment of the present invention.
Fig. 2 is a flowchart of an implementation of the method for identifying indoor human activities based on channel state information according to an embodiment of the present invention.
Fig. 3 is a sub-flow diagram of an active feature database construction module provided by an embodiment of the present invention.
Fig. 4 is a sub-flowchart of an indoor environment change information extraction module according to an embodiment of the present invention.
FIG. 5 is a sub-flow diagram of an activity monitoring module provided by an embodiment of the present invention.
Fig. 6 is a sub-flow diagram of an activity feature extraction module provided by an embodiment of the present invention.
Fig. 7 is a sub-flow diagram of an activity recognition module provided by an embodiment of the present invention.
Fig. 8 is a comparison graph of simulation results of two existing indoor human activity recognition methods according to the embodiment of the present invention.
Fig. 9 is a comparison diagram for simulating whether to recognize human activities based on principal component analysis when the experimental environment is unchanged according to the embodiment of the present invention.
Fig. 10 is a diagram illustrating collected raw CSI data according to an embodiment of the present invention.
Fig. 11 is a schematic diagram of CSI data after K-means extremum detection according to an embodiment of the present invention.
Fig. 12 is a schematic diagram of CSI data processed by a low-pass filter according to an embodiment of the present invention.
Fig. 13 is a schematic diagram of the finally collected principal component information provided by the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention effectively utilizes the characteristics of environmental noise, channel change and CSI data statistics to overcome the defect of low reliability of the recognition result in the prior art and improve the activity recognition precision. In particular to an indoor personnel activity identification method based on channel state information, which can be used for man-machine interaction, smart families, nursing of old people and emergency recourse.
The following detailed description of the principles of the invention is provided in connection with the accompanying drawings.
As shown in fig. 1, the indoor human activity identification method based on channel state information according to an embodiment of the present invention includes the following steps:
s101: acquiring Channel State Information (CSI) through existing WiFi equipment;
s102: preprocessing the collected CSI data, and respectively extracting correlation coefficients and variances among subcarriers;
s103: constructing a decision tree based on the correlation coefficient and the variance, monitoring personnel activities by using the decision tree, triggering an indoor personnel activity identification module after monitoring the indoor personnel activities, and carrying out S104, otherwise, continuously collecting CSI data and keeping monitoring of the personnel activities;
s104: further processing the collected CSI original data to obtain activity characteristic information, firstly removing extreme values in the CSI data by using an extreme value detection algorithm based on K-means clustering, and then removing environmental noise and high-frequency components in the CSI data by using a low-pass filter;
s105: extracting active Principal Component information by a Principal Component Analysis based on Dynamic Time programming matching algorithm (PCA-DTW), and then performing active matching;
s106: and identifying the indoor personnel activities, then ending the activity identification process, entering the next personnel activity identification state, and returning to the S101.
The application of the principles of the present invention will now be described in further detail with reference to the accompanying drawings.
As shown in fig. 2, the method for identifying indoor personnel activities based on channel state information provided by the embodiment of the present invention specifically includes the following steps:
step one, acquiring Channel State Information (CSI) through existing WiFi equipment, and specifically comprising the following steps:
at time t, acquiring channel state information H of the l-th link between the transmitting terminal TX and the receiving terminal TR, including subcarrier information of multiple groups of links:
Figure BDA0002067397790000131
wherein [ ·]TDenotes transposition, NsubIs the number of OFDM subcarriers, g represents all the links, h (i) can be expressed as:
H(i)＝||H(i)||ej∠H(i)；
where h (i) is channel state information of the ith subcarrier, | h (i) and ═ h (i) represent amplitude and phase information, respectively.
Typically, one of the links is selected from
Figure BDA0002067397790000132
The acquisition of continuous unprocessed CSI is started and then passes through a sliding window of length m, which can be expressed as:
Figure BDA0002067397790000133
wherein the content of the first and second substances,
Figure BDA0002067397790000134
is a number NsubXm matrix, t is the start index of the sliding window.
Then in the time direction from
Figure BDA0002067397790000135
The ith subcarrier is extracted, which can be expressed as:
Figure BDA0002067397790000136
wherein
Figure BDA0002067397790000137
Is a 1 x m matrix, ExsubRaw-i(. represents a slave matrix
Figure BDA0002067397790000138
The ith subcarrier is extracted.
Step two, preprocessing the collected CSI data, and extracting environment change information, that is, extracting correlation coefficients and variances between subcarriers, as shown in fig. 4, specifically including:
(1) preprocessing original CSI data;
1a) calculating the variance of the CSI subcarriers;
the variance of each subcarrier over the sliding window constitutes NsubA x 1 vector, which can be expressed as;
Figure BDA0002067397790000141
wherein the content of the first and second substances,
Figure BDA0002067397790000142
viis the variance of the ith subcarrier, and Var (-) is the variance symbol.
Secondly, the average variance of all subcarriers can be conveniently measured;
V＝Var(Vw)；
where V represents the degree of channel fluctuation. Thus, when a person moves in the target area, it is more likely to become larger.
1b) Calculating correlation coefficients among the CSI subcarriers;
in particular, for NsubSub-carriers, measuring the correlation coefficient as follows;
Figure BDA0002067397790000143
where ρ isi,jRepresents the correlation coefficient between the ith subcarrier and the jth subcarrier, and Corr (-) is the correlation symbol.
A group of correlation coefficients can be obtained by repeating the calculation, and then the incorrect correlation coefficients can be screened out by carrying out extreme point detection through the least mean square. Specifically, the objective function is defined as:
Figure BDA0002067397790000144
wherein:
Figure BDA0002067397790000145
ρi,j＝{ρi,j,i<j∈[1,Nsub]}；
Γ＝Nsub(Nsub-1)/2；
where min (-) and med (-) are min and median operations, respectively, ri,jIs the remainder of the process,
Figure BDA0002067397790000146
is the least mean square estimate, and Γ is ρi,jThe number of (2).
(2) Outlier detection was as follows:
Figure BDA0002067397790000151
Figure BDA0002067397790000152
Figure BDA0002067397790000153
Figure BDA0002067397790000154
after incorrect correlation coefficients are determined, they can be filtered out and the remaining coefficients averaged as a final estimate of a set of correlation coefficients:
Figure BDA0002067397790000155
where ρ and I (ρ)i，j) And finally obtaining a correlation coefficient rho which can be used for indoor personnel activity identification.
Step three, constructing a decision tree based on the correlation coefficient and the variance, monitoring the personnel activities by using the decision tree, triggering an indoor personnel activity recognition module after monitoring the indoor personnel activities, and if not, continuously collecting CSI data and keeping an activity monitoring state, as shown in FIG. 5, specifically comprising:
and according to the result of the second step, on the basis of environmental change, namely subcarrier correlation coefficient and variance, constructing a decision tree and monitoring personnel activities by using the decision tree, triggering an indoor personnel activity identification module after monitoring the indoor personnel activities, and otherwise, continuously collecting CSI data and keeping an activity monitoring state.
Indoor human activity monitoring using correlation coefficients and variances can be expressed as:
Figure BDA0002067397790000156
step four, further processing the collected CSI original data to obtain activity characteristic information, namely principal component information, firstly removing extreme values in the CSI data by using an extreme value detection algorithm based on K-means clustering, and then removing environmental noise and high-frequency components in the CSI data by using a low-pass filter; the method specifically comprises the following steps:
the collected CSI original data are further processed to obtain activity characteristic information, an extreme value detection algorithm based on K-means clustering is used for removing an extreme value in the CSI data, and then a low-pass filter is used for removing environmental noise and high-frequency components in the CSI data.
(1) K-means is a classical clustering method that is very sensitive to outliers and can be used to find groups of data points that are strongly correlated, while anomaly detection is used to find data points that are not strongly correlated with other data points. Therefore, K-means is very suitable for the detection of abnormal values. The extreme point processing of the original CSI data includes the following five steps.
1) The data is subjected to a normalization operation, which can be expressed as:
Figure BDA0002067397790000161
wherein
Figure BDA0002067397790000162
h is the output of the data normalization and Z-score (. cndot.) is the data normalization operator.
2) Randomly selecting k data points from a data set h as an original centroid:
u＝{u1，u2，...uω...，uk},(ω＝1,2,...,k)；
3) the maximum number of iterations can be set to Ω and the iterations are from1To Ω, as follows:
Figure BDA0002067397790000163
wherein C isχIs the original cluster.
3.1) calculating data points
Figure BDA0002067397790000164
And each centroid uωThe distance between
Figure BDA0002067397790000165
Figure BDA0002067397790000166
3.2) Next step, labeling
Figure BDA0002067397790000167
Category corresponding as minimum
Figure BDA0002067397790000168
The cluster is updated at this time, which can be expressed as:
Figure BDA0002067397790000169
3.3) for all data points Cω，ω∈[1，k]Recalculating new centroids:
Figure BDA00020673977900001610
3.4) if the k centroids buzai2 change, go to step four.
4) Finally, the C cluster outputs are:
C＝{C1,C2,...,Ck}；
5) after clustering, the detection of extrema in the CSI data is as follows:
u＝{u1，u2，...uε...，uk},ε∈[1,k]；
C＝{C1,C2,...Cε...,Ck},ε∈[1，k]；
where u is the final cluster centroid and C is the corresponding cluster:
Figure BDA0002067397790000171
wherein
Figure BDA0002067397790000172
Is the distance error of the centroid from the data point within the corresponding cluster.
5.1) Final, comparison
Figure BDA0002067397790000173
With the known threshold ζ, it can be expressed as:
Figure BDA0002067397790000174
5.2) after the extreme value detection algorithm based on the K-means clustering is processed, the following can be obtained:
Figure BDA0002067397790000175
where EVD (-) is an extremum detecting operator.
(2) The purpose of the low-pass filter is to eliminate any random environmental noise and high frequency components caused by non-human activity, considering that human activity usually has a relatively low frequency range. Meanwhile, special statistical characteristics caused by personnel activities in the CSI data can be effectively reserved. After passing through a low-pass filter, the following can be obtained:
Figure BDA0002067397790000176
where LPF (-) is the low pass filter operator.
Step five, extracting activity characteristic information, namely principal component information, through PCA-DTW, and then performing activity matching, as shown in fig. 6, specifically including:
and extracting activity characteristics, namely activity principal component information, by a dynamic time programming matching algorithm based on principal component analysis, and then performing activity matching.
(1.1) extracting active principal component information in the CSI data as follows;
(1) collecting filtered CSI, i.e.
Figure BDA0002067397790000177
Then, will
Figure BDA0002067397790000178
Dividing into data blocks containing 2 second intervals, arranging the data blocks of g links in rows to form a matrix
Figure BDA0002067397790000181
(2) Correlation estimation, calculating a correlation matrix with dimension m × m:
Figure BDA0002067397790000182
(3) and (3) feature decomposition, namely performing feature decomposition on the correlation matrix, and calculating a feature vector:
Figure BDA0002067397790000183
wherein
Figure BDA0002067397790000184
υiIs the eigenvector, σ is the number of principal components to be obtained, and deceigen (·) is the eigen-decomposition operator.
(4) Activity signal reconstruction the main components can be constructed as follows:
Figure BDA0002067397790000185
Figure BDA0002067397790000186
wherein
Figure BDA0002067397790000187
And
Figure BDA0002067397790000188
are respectively the τ -th eigenvector and the τ -th component, PiIs a principal component matrix.
And (1.2) according to the characteristics of indoor personnel activities, performing similarity measurement on two activity characteristics, namely unknown activities and activity characteristics in a known database by adopting dynamic time warping. The specific algorithm is as follows:
(1) given two activity characteristics, an unknown PiA is known
Figure BDA0002067397790000189
Then find a function of the point-to-point distance:
Figure BDA00020673977900001810
(2) finding the correspondence between the points:
Figure BDA00020673977900001811
wherein:
Figure BDA00020673977900001812
Figure BDA00020673977900001813
then, considering ψ (w), the solution for the cumulative distance between two active features is as follows:
Figure BDA00020673977900001814
(3) the final DTW estimate is to find the optimum ψ (w) to minimize the cumulative distance:
Figure BDA0002067397790000191
step six, identifying indoor personnel activities, finishing the identification process after identifying the personnel activities, and entering the next personnel activity identification state, namely returning to the step one; as further shown in fig. 7, the method specifically includes:
by computing the unknown activity features and the activity features in the pre-constructed database, the process of human activity recognition is as follows:
Figure BDA0002067397790000192
Figure BDA0002067397790000193
where argmin (·) is the minimum operator symbol and κ represents the category of human activity. Only when
Figure BDA0002067397790000194
When the minimum is taken, the person activity can be correctly identified.
The application effect of the present invention will be described in detail with reference to the simulation.
Firstly, simulation conditions: the transmitting and receiving end nodes are deployed in an indoor space with a reachable view distance of 10m by 10m, a notebook computer with an Intel5300 wireless network card is arranged as a transmitting end, and the same notebook with the Intel5300 is selected as a receiving end. The CSI collecting tool is an open source drive on a Linux platform, and parameters can be adjusted to collect CSI data after equipment is configured.
Second, simulation content and results
Simulation 1, comparing the invention with the existing CSI-based indoor personnel activity recognition systems E-eyes and WiFinger, the invention selects four activities including standing, sitting, squatting and lying for recognition, which are respectively represented by A, B, C, D. The invention uses the following indexes to measure the system identification accuracy:
(1) true Positive Rate (TPR): the TPR for activity a is defined as the proportion of class a activities that are correctly identified as activity a.
(2) False Positive Rate (FPR): the FPR for activity A is defined as the proportion of all activities other than A that are misrecognized as A.
The results are shown in FIG. 8. As can be seen from fig. 8, under the same environment, compared with the CSI-based indoor human activity recognition systems E-eyes and WiFinger, the recognition rate of WiFinger activity is 95%, 93%, 94%, and 95%, the recognition rate of E-eyes activity is 98%, 97%, 95%, and 97%, and the recognition rate of invention activity is 100%, 98%, and 100%. Therefore, compared with the former two methods, the method has higher correct identification rate.
Simulation 2, when the experimental environment is not changed, whether or not the simulation is based on Principal Component Analysis (PCA) is performed, and the result is shown in fig. 9.
As can be seen from fig. 9, when the indoor human activity recognition is performed based on PCA, the recognition rate is 100%, 98%, 100%, otherwise the recognition rate is 87%, 75%, 72%, 82%. The performance based on PCA is better, and the recognition rate is higher.
The effect of the present invention will be described in detail with reference to experiments.
The system performance is evaluated through a series of experimental simulations, the provided indoor personnel activity recognition is subjected to experimental design and verification, and the experimental result is subjected to detailed analysis. Under the same environment, compared with the indoor human activity recognition systems E-eyes and WiFinger based on CSI, the WiFinger activity recognition rate is 95%, 93%, 94% and 95%, the E-eyes activity recognition rate is 98%, 97%, 95% and 97%, and the E-eyes activity recognition rate is 100%, 98% and 100%. Therefore, compared with the former two methods, the method has higher correct identification rate. When the experimental environment is not changed, simulation is carried out on the basis of the PCA, when indoor personnel activity recognition is carried out on the basis of the PCA, the recognition rate is 100%, 98% and 100%, otherwise, the recognition rate is 87%, 75%, 72% and 82. The performance based on PCA is better, and the recognition rate is higher.
Experimental data
Table 1 sample data collected
Movement of A B C D
Number of 100 100 100 100
Motion matching data
WiAR Table 2. action matching Effect of WiAR
Figure BDA0002067397790000201
Figure BDA0002067397790000211
WiFinger Table 3. action matching Effect of WiFinger
Movement of A B C D
A 95 3 2 0
B 4 93 2 1
C 2 3 94 1
D 0 2 3 95
E-eyes Table 4. action matching Effect of E-eyes
Movement of A B C D
A 98 2 0 0
B 1 97 2 0
C 0 2 97 1
D 0 0 1 99
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Claims (7)

1. The indoor personnel activity identification method based on the channel state information is characterized in that the indoor personnel activity identification method based on the channel state information is used for preprocessing collected CSI data to extract environment change information comprising correlation coefficients and variances; then, sensing the surrounding environment by using a decision tree to monitor the activities of the personnel; secondly, if no personnel activity is detected, continuously collecting CSI data and keeping the activity monitoring state, otherwise triggering an activity feature extraction module, and carrying out extreme value removal processing and low-pass filtering processing on the collected CSI data in the activity feature extraction module; then, performing activity matching and finally achieving personnel activity recognition by using a dynamic time planning algorithm based on principal component analysis;
the indoor personnel activity identification method based on the channel state information specifically comprises the following steps:
the method comprises the steps that firstly, Channel State Information (CSI) is collected through existing WiFi equipment;
secondly, preprocessing the collected CSI data, and respectively calculating correlation coefficients among subcarrier time sequences and variances of corresponding subcarrier time sequences; the method specifically comprises the following steps:
(1) preprocessing original CSI data;
1) calculating the variance of the CSI subcarriers;
the variance of each subcarrier over the sliding window constitutes NsubA x 1 vector, expressed as;
Figure FDA0002967091160000011
wherein the content of the first and second substances,
Figure FDA0002967091160000012
viis the variance of the ith subcarrier, and Var (-) is the variance symbol;
secondly, measuring the average variance of all subcarriers;
V＝Var(Vw)；
wherein V represents the degree of channel fluctuation;
2) calculating correlation coefficients among the CSI subcarriers;
for NsubSub-carriers, measuring the correlation coefficient as follows;
Figure FDA0002967091160000013
where ρ isi,jRepresenting the correlation coefficient between the ith subcarrier and the jth subcarrier, wherein Corr (-) is a correlation symbol;
repeating the above calculation to obtain a group of correlation coefficients, performing extreme point detection through the least median square, and screening out incorrect correlation coefficients, wherein the definition of the objective function is as follows:
Figure FDA0002967091160000021
wherein:
Figure FDA0002967091160000022
ρi,j＝{ρi,j,i<j∈[1,Nsub]}；
Γ＝Nsub(Nsub-1)/2；
where min (-) and med (-) are min and median operations, respectively, ri,jIs the remainder of the process,
Figure FDA0002967091160000023
is the least mean square estimate, and Γ is ρi,jThe number of (2);
(2) outlier detection was as follows:
Figure FDA0002967091160000024
Figure FDA0002967091160000025
Figure FDA0002967091160000026
Figure FDA0002967091160000027
after incorrect correlation coefficients are determined, they are filtered out and the remaining coefficients are then averaged as a final estimate of a set of correlation coefficients:
Figure FDA0002967091160000028
where ρ and I (ρ)i,j) Respectively obtaining a final correlation coefficient and the outlier index, and finally obtaining a correlation coefficient rho for indoor personnel activity identification;
thirdly, constructing a decision tree based on the correlation coefficient and the variance, monitoring personnel activities by using the decision tree, triggering an indoor personnel activity identification module when monitoring the indoor personnel activities, and carrying out the fourth step, otherwise, continuously collecting CSI data and keeping monitoring of the personnel activities;
fourthly, further processing the collected CSI original data to obtain activity characteristic information, firstly removing extreme values in the CSI data by using an extreme value detection algorithm based on K-means clustering, and then removing environmental noise and high-frequency components in the CSI data by using a low-pass filter;
fifthly, extracting activity principal component information through a dynamic time programming matching algorithm based on principal component analysis, and then performing activity matching; the method specifically comprises the following steps:
(1) the step of extracting the active principal component information in the CSI data is as follows;
1) collecting filtered CSI, i.e.
Figure FDA0002967091160000031
Will be provided with
Figure FDA0002967091160000032
Dividing into data blocks containing 2 second intervals, arranging the data blocks of g links in rows to form a matrix
Figure FDA0002967091160000033
2) Correlation estimation, calculating a correlation matrix with dimension m × m:
Figure FDA0002967091160000034
3) and (3) feature decomposition, namely performing feature decomposition on the correlation matrix, and calculating a feature vector:
Figure FDA0002967091160000035
wherein
Figure FDA0002967091160000036
υiIs a feature vector, sigma is the number of principal components to be obtained, deceigen (·) is a feature decomposition operation symbol;
4) and (3) reconstructing an activity signal, wherein the main components are constructed as follows:
Figure FDA0002967091160000037
Figure FDA0002967091160000038
wherein
Figure FDA0002967091160000039
And
Figure FDA00029670911600000310
are respectively the τ -th eigenvector and the τ -th component, PiIs a principal component matrix;
(2) according to the characteristics of indoor personnel activities, carrying out similarity measurement on two activity characteristics, namely unknown activities and activity characteristics in a known database by adopting dynamic time warping; the specific algorithm is as follows:
1) given two activity characteristics, an unknown PiA is known
Figure FDA00029670911600000311
Then find a function of the point-to-point distance:
Figure FDA00029670911600000312
2) finding the correspondence between the points:
Figure FDA00029670911600000313
wherein:
Figure FDA0002967091160000041
Figure FDA0002967091160000042
then, considering ψ (w), the solution for the cumulative distance between two active features is as follows:
Figure FDA0002967091160000043
3) the final DTW estimate is to find the optimum ψ (w) that minimizes the cumulative distance:
Figure FDA0002967091160000044
and sixthly, identifying the indoor personnel activities, then ending the activity identification process, entering the next personnel activity identification state, and returning to the first step.
2. The method for identifying indoor human activities based on channel state information according to claim 1, wherein the first step of collecting channel state information CSI through existing WiFi equipment specifically includes:
at time t, acquiring channel state information H of the l-th link between the transmitting terminal TX and the receiving terminal TR, including subcarrier information of multiple groups of links:
Figure FDA0002967091160000045
wherein [ ·]TDenotes transposition, NsubIs the number of OFDM subcarriers, g represents all the links, h (i) can be expressed as:
H(i)＝||H(i)||ej∠H(i)；
wherein h (i) is channel state information of the ith subcarrier, | h (i) and ═ h (i) represent amplitude and phase information, respectively;
selecting one of the links from
Figure FDA0002967091160000046
Start of acquisition of continuous no additionThe processed CSI, then passes through a sliding window of length m, represented as:
Figure FDA0002967091160000047
wherein the content of the first and second substances,
Figure FDA0002967091160000048
is a number NsubXm matrix, t is the start index of the sliding window;
in the time direction from
Figure FDA0002967091160000049
The ith subcarrier is extracted, and is expressed as:
Figure FDA00029670911600000410
wherein
Figure FDA0002967091160000051
Is a 1 x m matrix, ExsubRaw-i(. represents a slave matrix
Figure FDA0002967091160000052
The ith subcarrier is extracted.
3. The method for identifying indoor human activities based on channel state information according to claim 1, wherein the third step constructs a decision tree based on the correlation coefficient and the variance, monitors human activities by using the decision tree, triggers an indoor human activity identification module when the indoor human activities are monitored, and otherwise continues to collect CSI data and maintains an activity monitoring state, and specifically comprises:
according to the result of the second step, on the basis of environmental change, namely subcarrier correlation coefficient and variance, constructing a decision tree and monitoring personnel activities by using the decision tree, triggering an indoor personnel activity identification module after monitoring the indoor personnel activities, and otherwise, continuously collecting CSI data and keeping an activity monitoring state;
indoor human activity monitoring with correlation coefficients and variances is expressed as:
Figure FDA0002967091160000053
4. the method for identifying indoor human activity based on channel state information according to claim 1, wherein the fourth step further processes the collected CSI raw data to obtain activity characteristic information, removes extrema in the CSI data using an extremum detection algorithm based on K-means clustering, and removes environmental noise and high frequency components in the CSI data using a low pass filter;
(1) the K-means clustering method is very sensitive to abnormal values, finds strongly related data point groups, and detects the abnormality to find data points which are not strongly related to other data points; the extreme point processing of the original CSI data comprises the following steps;
1) the data was subjected to a normalization operation, expressed as:
Figure FDA0002967091160000054
wherein
Figure FDA0002967091160000055
h is the output of the data normalization, Z-score (. cndot.) is the data normalization operator;
2) randomly selecting k data points from a data set h as an original centroid:
u＝{u1,u2,...uω...,uk},(ω＝1,2,...,k)；
3) the maximum number of iterations is set to Ω, and the iterations are from 1 to Ω, as follows:
Figure FDA0002967091160000061
wherein C isχIs the original cluster;
3.1) calculating data points
Figure FDA0002967091160000062
And each centroid uωThe distance between
Figure FDA0002967091160000063
Figure FDA0002967091160000064
3.2) Next step, labeling
Figure FDA0002967091160000065
Category corresponding as minimum
Figure FDA0002967091160000066
At this time, the cluster is updated, which is expressed as:
Figure FDA0002967091160000067
3.3) for all data points Cω,ω∈[1,k]Recalculating new centroids:
Figure FDA0002967091160000068
3.4) if the k centroids buzai2 are changed, entering the step four;
4) the final C cluster outputs are:
C＝{C1,C2,...,Ck}；
5) after clustering, extremum detection in CSI data:
u＝{u1,u2,...uε...,uk},ε∈[1,k]；
C＝{C1,C2,...Cε...,Ck},ε∈[1,k]；
where u is the final cluster centroid and C is the corresponding cluster:
Figure FDA0002967091160000069
wherein
Figure FDA00029670911600000610
Distance error between the centroid and the data point in the corresponding cluster;
5.1) Final, comparison
Figure FDA00029670911600000611
And the known threshold ζ, expressed as:
Figure FDA00029670911600000612
5.2) after the extreme value detection algorithm based on the K-means clustering is processed, obtaining:
Figure FDA00029670911600000613
where EVD (-) is an extremum detecting opcode;
(2) the low-pass filter eliminates random environmental noise and high-frequency components caused by any non-human activities; meanwhile, the special statistical characteristics caused by the human activities in the CSI data are reserved, and the statistical characteristics are obtained through a low-pass filter:
Figure FDA0002967091160000071
where LPF (-) is the low pass filter operator.
5. The indoor human activity recognition method based on channel state information as claimed in claim 1, wherein the sixth step performs indoor human activity recognition, and after human activity recognition, ends the recognition process this time, and enters the next human activity recognition state, that is, it needs to return to step one to specifically include:
by computing the unknown activity features and the activity features in the pre-constructed database, the process of human activity recognition is as follows:
Figure FDA0002967091160000072
Figure FDA0002967091160000073
where argmin (·) is the minimum operator symbol, κ represents the category of human activity; when in use
Figure FDA0002967091160000074
When the minimum is taken, the activity of the person is correctly identified.
6. A human-computer interaction control system applying the method for identifying the indoor personnel activities based on the channel state information as claimed in any one of claims 1 to 5.
7. An intelligent home control system applying the method for identifying the activity of people in a room based on channel state information as claimed in any one of claims 1 to 5.
CN201910425638.2A 2019-05-21 2019-05-21 Indoor personnel activity identification method based on channel state information and man-machine interaction system Active CN110337066B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910425638.2A CN110337066B (en) 2019-05-21 2019-05-21 Indoor personnel activity identification method based on channel state information and man-machine interaction system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910425638.2A CN110337066B (en) 2019-05-21 2019-05-21 Indoor personnel activity identification method based on channel state information and man-machine interaction system

Publications (2)

Publication Number Publication Date
CN110337066A CN110337066A (en) 2019-10-15
CN110337066B true CN110337066B (en) 2021-04-16

Family

ID=68139059

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910425638.2A Active CN110337066B (en) 2019-05-21 2019-05-21 Indoor personnel activity identification method based on channel state information and man-machine interaction system

Country Status (1)

Country Link
CN (1) CN110337066B (en)

Families Citing this family (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110855382B (en) * 2019-12-02 2021-01-22 北京邮电大学 Method and device for detecting human body activity data based on CSI (channel State information), and electronic equipment
CN111030766B (en) * 2019-12-13 2022-05-06 福州大学 People flow detection method and system based on WIFI signal
CN110991559B (en) * 2019-12-19 2023-05-12 中国矿业大学 Indoor personnel behavior non-contact cooperative sensing method
CN111225354B (en) * 2020-02-14 2022-02-22 重庆邮电大学 CSI human body falling identification method under WiFi interference environment
CN111626174B (en) * 2020-05-22 2023-03-24 哈尔滨工程大学 Attitude robust motion recognition method based on channel state information
CN111683345B (en) * 2020-06-09 2021-12-14 江苏希塔信息科技有限公司 Wireless sensor network anti-interference positioning method based on CSI and RSSI
CN112333653A (en) * 2020-09-22 2021-02-05 西安电子科技大学 Identity intelligent identification method and system based on WiFi channel state information
CN112769721B (en) * 2020-12-25 2021-11-23 中国科学院计算技术研究所 Multi-feature-based multi-distribution cooperative OFDM signal detection method and system
CN112869734B (en) * 2021-01-11 2022-10-21 乐鑫信息科技（上海）股份有限公司 Wi-Fi human body detection method and intelligent device
CN112839327B (en) * 2021-01-21 2022-08-16 河北工程大学 Personnel validity detection method and device based on WiFi signals
CN113054491B (en) * 2021-03-08 2022-07-05 常州百芝龙智慧科技有限公司 AI socket system based on human motion of Wi-Fi is listened
CN112686221B (en) * 2021-03-11 2021-05-25 成都点泽智能科技有限公司 Pedestrian entrance and exit monitoring method, device and equipment and computer storage medium
CN113300750A (en) * 2021-05-24 2021-08-24 南京邮电大学 Personnel identity authentication and handwritten letter identification method based on WIFI signal
CN113326801A (en) * 2021-06-22 2021-08-31 哈尔滨工程大学 Human body moving direction identification method based on channel state information
CN113935373A (en) * 2021-10-11 2022-01-14 南京邮电大学 Human body action recognition method based on phase information and signal intensity
CN114079859A (en) * 2021-11-04 2022-02-22 中国矿业大学 Monitoring device and monitoring method based on CSI
CN114041784B (en) * 2021-11-11 2023-09-19 天津大学 Double gait disorder judging device based on commercial Wi-Fi
CN116205469B (en) * 2023-05-05 2023-07-25 武汉理工大学 Ship driver active state monitoring method and device based on physical channel
CN116304915B (en) * 2023-05-16 2023-08-29 山东科技大学 WiFi-based contactless action recognition method, system and laboratory device
CN116821778B (en) * 2023-08-30 2024-01-09 之江实验室 Gait recognition method and device based on WIFI signals and readable storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107154088A (en) * 2017-03-29 2017-09-12 西安电子科技大学 Activity personnel amount method of estimation based on channel condition information
CN107947874A (en) * 2017-11-21 2018-04-20 太原理工大学 Indoor map semanteme identification method based on WiFi channel condition informations
CN109474890A (en) * 2018-12-19 2019-03-15 重庆邮电大学 A kind of indoor objects passive tracking method based on WiFi Doppler frequency shift
CN109587645A (en) * 2018-11-12 2019-04-05 南京邮电大学 Personnel's recognition methods under indoor environment based on channel state information
CN109657572A (en) * 2018-12-04 2019-04-19 重庆邮电大学 Goal behavior recognition methods after a kind of wall based on Wi-Fi

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10104195B2 (en) * 2015-03-20 2018-10-16 The Trustees Of The Stevens Institute Of Technology Device-free activity identification using fine-grained WiFi signatures
CN109670434B (en) * 2018-12-13 2022-09-27 南京理工大学 Sitting posture duration detection method based on wireless channel state information

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107154088A (en) * 2017-03-29 2017-09-12 西安电子科技大学 Activity personnel amount method of estimation based on channel condition information
CN107154088B (en) * 2017-03-29 2019-03-26 西安电子科技大学 Activity staff quantity survey method based on channel state information
CN107947874A (en) * 2017-11-21 2018-04-20 太原理工大学 Indoor map semanteme identification method based on WiFi channel condition informations
CN109587645A (en) * 2018-11-12 2019-04-05 南京邮电大学 Personnel's recognition methods under indoor environment based on channel state information
CN109657572A (en) * 2018-12-04 2019-04-19 重庆邮电大学 Goal behavior recognition methods after a kind of wall based on Wi-Fi
CN109474890A (en) * 2018-12-19 2019-03-15 重庆邮电大学 A kind of indoor objects passive tracking method based on WiFi Doppler frequency shift

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
FreeSense: Indoor Human Identification with Wi-Fi Signals;Tong Xin等;《 2016 IEEE Global Communications Conference (GLOBECOM)》;20161208;全文 *
Leveraging Topic Model for CSI Based Human Activity Recognition;Kun Zhao等;《 2016 12th International Conference on Mobile Ad-Hoc and Sensor Networks (MSN)》;20161218;全文 *
基于WiFi信号的人体行为感知技术研究综述;鲁勇等;《计算机学报》;20190228;全文 *

Also Published As

Publication number Publication date
CN110337066A (en) 2019-10-15

Similar Documents

Publication Publication Date Title
CN110337066B (en) Indoor personnel activity identification method based on channel state information and man-machine interaction system
US11902857B2 (en) Handling concept drift in Wi-Fi-based localization
Wang et al. Device-free wireless localization and activity recognition: A deep learning approach
US20210250420A1 (en) Device-free activity identification using fine-grained wifi signatures
Arshad et al. Wi-chase: A WiFi based human activity recognition system for sensorless environments
CN109670434B (en) Sitting posture duration detection method based on wireless channel state information
KR102397311B1 (en) Method of Recognizing Daily Activities of Living-alone Person Using Channel State Information between WiFi Transmitter and WiFi Receiver
Yang et al. Wi-Count: Passing people counting with COTS WiFi devices
Zhang et al. WiFiMap+: High-level indoor semantic inference with WiFi human activity and environment
CN110072186B (en) Weighted naive Bayes indoor positioning method based on independent attributes
Zhu et al. NotiFi: A ubiquitous WiFi-based abnormal activity detection system
CN111698258B (en) WiFi-based environmental intrusion detection method and system
WO2021043126A1 (en) System and method for event recognition
CN114469076B (en) Identity-feature-fused fall identification method and system for solitary old people
CN111901028B (en) Human body behavior identification method based on CSI (channel State information) on multiple antennas
Andrade et al. Outlier detection using k-means clustering and lightweight methods for Wireless Sensor Networks
Fahad et al. Anomalies detection in smart-home activities
CN114781463A (en) Cross-scene robust indoor tumble wireless detection method and related equipment
Khalil et al. Sonicdoor: scaling person identification with ultrasonic sensors by novel modeling of shape, behavior and walking patterns
CN104536996A (en) Computational node anomaly detection method in isomorphic environments
CN111465093A (en) Fingerprint positioning method and device, electronic equipment and computer storage medium
CN113453180B (en) Intelligent detection method and system for human body tumble and information data processing terminal
CN116304604A (en) Multivariate time series data anomaly detection and model training method and system
CN108834043B (en) Priori knowledge-based compressed sensing multi-target passive positioning method
Raghuvanshi et al. Machine learning approach for anomaly detection in wireless sensor data

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant