CN104766427B

CN104766427B - A kind of house illegal invasion detection method based on Wi Fi

Info

Publication number: CN104766427B
Application number: CN201510203849.3A
Authority: CN
Inventors: 赵菊敏; 霍世敏; 李灯熬
Original assignee: Taiyuan University of Technology
Current assignee: Taiyuan University of Technology
Priority date: 2015-04-27
Filing date: 2015-04-27
Publication date: 2017-04-05
Anticipated expiration: 2035-04-27
Also published as: CN104766427A

Abstract

The present invention relates to it is used for the method for detecting illegal invasion house, specially a kind of house illegal invasion detection method based on Wi Fi.This method first gathers CSI signal sequences, then the denoising of CSI signal sequences and wavelet transform are processed, then set up the action model of owner, after some CSI signal sequences collection, denoising is carried out to the CSI signal sequences and wavelet transform is processed, action sequence is decomposed into, is then matched with the action model set up before, and then is judged whether illegal invasion.This method is using the action of people to channel condition information（Channel State Information, CSI）Impact, house illegal invasion is detected.The algorithm avoids the overhead for purchasing that special installation brings, and makes the safety in house get a promotion.

Description

A kind of house illegal invasion detection method based on Wi-Fi

Technical field

The present invention relates to be used for the illegal invasion detection method in house, specially a kind of house based on Wi-Fi illegally enters Invade detection method.

Background technology

In the field of wireless communication, CSI just refers to channel immediate status, describes the channel attribute of communication link, visually For the impulse response of digital filter.And average CSI refers to statistical property of the channel within a period of time.Contain environment weak Distribution, average channel gain, angle of sight component and spatial coherence.In the environment of wireless Wi-Fi is disposed with, physical activity Wireless signal can be had an impact.The impact that a series of action is brought to wireless signal, is all reflected in CSI.Therefore, pass through Measurement CSI, and learning model building is carried out to physical activity, it becomes possible to judge that people is embodied in and what action done, realize to people's Detection and tracking.At present, there is scholar using CSI information, action tracking and detection have been carried out to the people in room, so as to through walls Just can interpolate that in room, there are several individuals, what action they respectively doing, and then realizes various applications, and such as old people is risen Residence judged, fast and effeciently can be rescued when the situation such as falling down；For another example when police performs task, It is remote it may determine that there is several suspects within doors, it is ready to be that next step arrests work.Existing house illegal invasion Detecting system, is mostly based on photographic head or infrared monitor control.And both technologies or a line of sight conditions is limited to, or be limited to The bright-dark degree of light, is vulnerable to the impact of external environment condition, and needs to spend extra expense to buy specific equipment.Cause This, is badly in need of one kind and stablizes, and the house illegal invasion detecting system of low cost.

The content of the invention

The present invention is in order to realize the detection of illegal invasion in inexpensive, high-precision house, there is provided a kind of to be based on Wi-Fi House illegal invasion detection method.

The present invention adopts the following technical scheme that realization：A kind of house illegal invasion detection method based on Wi-Fi, Which comprises the following steps：

Wireless Wi-Fi transmitter and receivers are arranged in doors, with reference to the wireless communication link that smart machine within doors is produced Road, collects the CSI signal sequences that the action behavior of owner is formed to the impact of wireless signal；

First CSI signal sequences are decomposed using discrete wavelet, next filters singular value therein and noise, finally carries out Recovery and rebuilding, obtains reconstruction signal as the input variable of whole system；

Wavelet transform is carried out again to reconstruction signal, approximate size factor and detailed size factor is decomposed into, so that Reconstruction signal carries out time-frequency domain conversion；

After conversion, reconstruction signal is decomposed according to time and frequency, be decomposed into a series of action sequence, as The action model of owner's behavior characteristicss of systematic training；

To action sequence according to classifying, sorting technique adopts k-means clustering methods, and divides in this each classification Important sequence is not set up；

In detection-phase, to some CSI signal sequences collection after, denoising and discrete is carried out to the CSI signal sequences Wavelet transform process, is decomposed into action sequence, is then matched with each classification in the action model set up before, was matched Cheng Caiyong dynamic time warping algorithms, regard as owner if the match is successful, if matching is unsuccessful regard as illegal invasion.

This method is using the action of people to channel condition information（Channel State Information, CSI）Shadow Ring, house illegal invasion is detected.The behavior characteristicss of owner are modeled using machine learning, then to illegal invasion The action of person is identified and warning.The algorithm avoids the overhead for purchasing that special installation brings, and makes the safety in house Get a promotion.

Description of the drawings

Fig. 1 is this method flow chart.

Fig. 2 is CSI information schematic diagrams.

Specific embodiment

A kind of house illegal invasion detection method based on Wi-Fi, comprises the following steps：

Wireless Wi-Fi transmitter and receivers are arranged in doors, with reference to smart machine within doors, such as smart mobile phone, flat board With the wireless communication link of the generation such as computer, the CSI signals that the action behavior of owner is formed to the impact of wireless signal are collected Sequence；

Denoising is carried out to CSI signal sequences using discrete wavelet, first CSI signal sequences are decomposed, next filters wherein The noise such as singular value, finally carry out recovery and rebuilding, obtain reconstruction signal, as the input variable of whole system；

Discrete wavelet Pyatyi conversion is carried out again to reconstruction signal, approximate size factor and detailed size factor is decomposed into, from And make reconstruction signal carry out time-frequency domain conversion；

After conversion, reconstruction signal is decomposed according to time and frequency, be decomposed into a series of action sequence, i.e., action（Initial time, action, persistent period）, as the action model of owner's behavior characteristicss of systematic training；

Action sequence is classified according to significantly action, movement, fine motion, sorting technique is clustered using k-means Method, and set up important sequence in these three classifications respectively, significantly the important sequence of action is set to change one's clothes, the important sequence of movement Row are set to walk, and fine motion is made important sequence and is set to see TV；

In matching stage, to some CSI signal sequences collection after, denoising and discrete little is carried out to CSI signal sequences Wave conversion process, is decomposed into action sequence.Then matched with the action model set up before, matching process adopts DTW（It is dynamic State Time alignment）Algorithm；In order to reduce the computing cost of system, if three main actions, the match is successful, then it is assumed that To be owner；If three have two, the match is successful, continues two other actions of matching, if the match is successful, regards as Owner；If three have one, the match is successful, then continues four other actions of matching, if the match is successful, based on identification People；If three main actions are all mismatched, then it is assumed that being illegal invasion.

During denoising, the singular value and noise in CSI signal sequences is filtered using discrete wavelet wave filter, is protected to greatest extent Stay impact of the people to wireless signal.It is divided into three steps in detail：Signal decomposition, the determination of threshold value detail factor, the reconstruction of signal. In signal decomposition, wavelet transform is divided into CSI signal sequences the approximate factor of the detailed factor and low frequency of high frequency；Then Based on this smooth unbiased esti-mator, dynamically selected threshold is removing noise；Finally, the signal after denoising is rebuild.

Extract the meaning of one's words.As the different action of people can bring different amplitudes, frequency and time change to CSI signals.It is right Reconstruction signal adopts wavelet transform（DWT）, form a T/F window with frequency shift.Wavelet transform Signal can be made in time domain and the mutual phase in version of frequency domain, therefore according to frequency and the difference of time, it becomes possible to reconstruction signal Traffic Decomposition Into different action sequences, as action（Time of origin, action, persistent period）, data fusion is carried out under multiple links. Followed by classification.Criteria for classification is divided into two aspects：It is coarseness and fine-grained, i.e., first summarize and segment afterwards.Coarseness Ground can be divided into fine motion work, significantly action and movement；Segmented in each category on here again：Fine motion make include sleep, TV is seen, computer is played and is made a phone call, significantly action includes wear shoes, cooks, changes one's clothes and wash one's face and rinse one's mouth, it is mobile including on foot and related Adjoint action.During actual match, without distinguishing these all of fine-grained actions, it is only necessary to which coarseness ground first judges category Which kind of in, then compare with the action sequence contained by this apoplexy due to endogenous wind.

Clustering method can adopt the K-means methods in machine learning, compare can using DTW or EMD calculate away from From deviation, fair play can be judged to less than threshold value.

When model is set up, using semi-supervised learning pattern, such as mobile this class, and the leg speed of owner is recorded, at once Walk impact of the speed to frequency.When action is collected, if all mismatch with all of model before, and feature is not yet Match somebody with somebody, then be judged as illegally, giving a warning.Now we set a set action, are brandished after such as arm is first to the right to the left, So system can be judged for owner, meanwhile, new element is included in model library.If still mismatched, then just can be with It is judged as illegal invasion.Information can be sent from trend master cellular phone by network with initialization system, prevent illegal.

Claims

1. a kind of house illegal invasion detection method based on Wi-Fi, it is characterised in that comprise the following steps：

Wireless Wi-Fi transmitter and receivers are arranged in doors, with reference to the wireless communication link that smart machine within doors is produced, are received The CSI signal sequences that the action behavior of collection owner is formed to the impact of wireless signal；

First CSI signal sequences are decomposed using discrete wavelet, next filters singular value therein and noise, is finally recovered Reconstruct, obtains reconstruction signal as the input variable of whole system；

Wavelet transform is carried out again to reconstruction signal, approximate size factor and detailed size factor is decomposed into, so that reconstruct Signal carries out time-frequency domain conversion；

The reconstruction signal after conversion is decomposed according to time and frequency, be decomposed into a series of action sequence, as being The action model of owner's behavior characteristicss of system training；

Action sequence is classified according to significantly action, movement, fine motion, sorting technique adopts k-means clustering methods, And important sequence is set up respectively in this each classification；

In detection-phase, after gathering to some CSI signal sequence, denoising and discrete wavelet are carried out to the CSI signal sequences Conversion process, is decomposed into action sequence, is then matched with each classification in the action model set up before, and matching process is adopted Dynamic time warping algorithm is used, owner is regarded as if the match is successful, illegal invasion is regarded as if matching is unsuccessful.