CN104766427A - Detection method for illegal invasion of house based on Wi-Fi - Google Patents

Abstract

The invention relates to a method used for detecting illegal invasion of a house, in particular to a detection method for the illegal invasion of the house based on a Wi-Fi. The detection method comprises the steps that CSI signal sequences are collected, denoising treatment and a discrete wavelet transform treatment are carried out on the CSI signal sequences, then a motion module of an owner is established, after one CSI signal sequence is collected, the denoising treatment and the discrete wavelet transform treatment are carried out on the CSI signal sequences, the CSI signal sequences are decomposed into motion sequences, the motion sequences are matched with the motion module which is established before, and it is judged that whether the illegal invasion is conducted or not. The detection is carried out on the illegal invasion of the house by using the influence of human motions on the channel state information (CSI). According to the method, extra expenditure brought by buying special equipment is avoided, and the safety of the house can be improved.

Description

A Wi-Fi-based house illegal intrusion detection method

technical field

The invention relates to a method for detecting illegal intrusion of houses, in particular to a method for detecting illegal intrusion of houses based on Wi-Fi.

Background technique

In the field of wireless communication, CSI refers to the instant state of the channel, which describes the channel properties of the communication link, and can be regarded as the impulse response of the digital filter. The average CSI refers to the statistical characteristics of the channel over a period of time. Contains the distribution of ambient attenuation, average channel gain, line-of-sight component and spatial correlation. In an environment where wireless Wi-Fi is deployed, human activities will affect the wireless signal. The impact of a series of actions on the wireless signal is all reflected in the CSI. Therefore, by measuring CSI and learning and modeling human activities, it is possible to determine what actions a person is doing, and to detect and track people. At present, some scholars have used CSI information to track and detect the movements of people in the room, so that they can judge how many people are in the room through the wall and what actions they are doing, and then realize various applications, such as the elderly Judging by daily life, once a fall or other situation occurs, it can quickly and effectively rescue; for example, when the police are performing tasks, they can judge how many suspects are in the house from a long distance, and prepare for the next step of arrest. Most of the existing house illegal intrusion detection systems are based on camera or infrared monitoring. However, these two technologies are either limited by line-of-sight conditions, or by the brightness of the light, are easily affected by the external environment, and require additional expenses to purchase specific equipment. Therefore, there is an urgent need for a stable and low-cost house illegal intrusion detection system.

Contents of the invention

In order to realize low-cost and high-precision detection of illegal intrusion in a house, the present invention provides a method for detecting illegal intrusion in a house based on Wi-Fi.

The present invention is realized by adopting the following technical scheme: a method for detecting illegal house intrusion based on Wi-Fi, which includes the following steps:

Arrange wireless Wi-Fi transmitters and receivers in the house, combine the wireless communication links generated by smart devices in the house, and collect the CSI signal sequence formed by the influence of the owner's action behavior on the wireless signal;

Using discrete wavelet to firstly decompose the CSI signal sequence, secondly filter out the singular value and noise, and finally restore and reconstruct, and obtain the reconstructed signal as the input variable of the whole system;

Perform discrete wavelet transform on the reconstructed signal, and decompose it into approximate scale factor and detailed scale factor, so that the reconstructed signal can be transformed from time domain to frequency domain;

After the transformation, the reconstructed signal is decomposed according to time and frequency, and decomposed into a series of action sequences, which are used as the action model of the main behavior characteristics of the system training;

Classify the action sequences according to the classification method using the k-means clustering method, and set up important sequences in each category;

In the detection stage, after a certain CSI signal sequence is collected, the CSI signal sequence is denoised and processed by discrete wavelet transform, decomposed into an action sequence, and then matched with each category in the previously established action model. The matching process uses dynamic time Regular algorithm, if the match is successful, it will be identified as the owner, if the match is unsuccessful, it will be identified as illegal intrusion.

This method uses the influence of human actions on Channel State Information (CSI) to detect illegal house intrusion. Use machine learning to model the behavior characteristics of the owner, and then identify and alarm the actions of illegal intruders. This algorithm avoids the extra cost of purchasing special equipment and improves the security of the house.

Description of drawings

Figure 1 is the flow chart of this method.

Figure 2 is a schematic diagram of CSI information collection.

Detailed ways

A method for detecting illegal house intrusion based on Wi-Fi, comprising the following steps:

Arrange wireless Wi-Fi transmitters and receivers in the house, combine the wireless communication links generated by smart devices in the house, such as smart phones, tablets and computers, and collect the CSI signal sequence formed by the influence of the owner's actions on the wireless signal ;

Using discrete wavelet to denoise the CSI signal sequence, first decompose the CSI signal sequence, secondly filter out the noise such as singular value, and finally restore and reconstruct to obtain the reconstructed signal as the input variable of the whole system;

Then carry out discrete wavelet five-level transform on the reconstructed signal, and decompose it into approximate scale factor and detailed scale factor, so that the reconstructed signal can be transformed from time domain to frequency domain;

After the transformation, the reconstructed signal is decomposed according to time and frequency, and decomposed into a series of action sequences, namely action (start time, action, duration), which is used as the action model of the master's behavior characteristics for system training;

The action sequences are classified according to large-scale movements, movements, and micro-movements. The classification method adopts the k-means clustering method, and important sequences are set up in these three categories. The important sequences of large-scale movements are set as changing clothes and important sequences of moving Set it as walking, and set the important sequence of micro-movements as watching TV;

In the matching stage, after a certain CSI signal sequence is collected, the CSI signal sequence is denoised and processed by discrete wavelet transform, and decomposed into action sequences. Then match with the previously established action model. The matching process uses the DTW (Dynamic Time Warping) algorithm; in order to reduce the computational overhead of the system, if there are three main actions that are successfully matched, then it will be considered as the master; if two of the three match If it succeeds, it will continue to match two other actions. If the match is successful, it will be identified as the master; if one of the three matches is successful, then continue to match four other actions. If the match is successful, it will be identified as the master; If it does not match, it is considered illegal intrusion.

When denoising, the discrete wavelet filter is used to filter out the singular value and noise in the CSI signal sequence, and the human influence on the wireless signal is preserved to the greatest extent. The details are divided into three steps: signal decomposition, determination of threshold detail factor, and signal reconstruction. In the signal decomposition, the discrete wavelet transform divides the CSI signal sequence into high-frequency detailed factors and low-frequency approximate factors; then, based on the Stein unbiased estimation, the threshold is dynamically selected to remove the noise; finally, the denoised signal is reconstructed .

Extract semantics. Different actions of people will bring different amplitude, frequency and time changes to the CSI signal. Discrete wavelet transform (DWT) is used on the reconstructed signal to form a time-frequency window that changes with frequency. Discrete wavelet transform can transform signals in the time domain and frequency domain, so according to the difference in frequency and time, the reconstructed signal flow can be decomposed into different action sequences, namely action (time of occurrence, action, duration), Data fusion is performed under multiple links. Classify next. Classification standards are divided into two levels: coarse-grained and fine-grained, that is, first generalized and then subdivided. Coarse-grained can be divided into micro-movements, large-scale movements and movements; on top of this, it is further subdivided in each category: micro-movements include sleeping, watching TV, playing computer and making calls, and large-scale movements include wearing shoes, doing Meals, changing clothes and washing, mobility includes walking and related accompanying actions. In the actual matching, it is not necessary to distinguish all these fine-grained actions. It is only necessary to first determine which category it belongs to at a coarse-grained level, and then compare it with the action sequences contained in this category.

The clustering method can use the K-means method in machine learning, and the comparison can use DTW or EMD to calculate the distance deviation, and it can be judged as a legal action if it is lower than the threshold value.

When building the model, use a semi-supervised learning mode, such as moving, and record the pace of the owner, that is, the impact of walking speed on the frequency. When the action is collected, if it does not match all the previous models and the features do not match, then it will be judged as illegal and a warning will be issued. At this time, we set a predetermined action, such as waving the arm to the right first and then to the left, then the system can judge that it is the owner, and at the same time, incorporate the new action into the model library. If it still does not match, it can be judged as illegal intrusion. The system can be set to automatically send information to the owner's mobile phone through the network to prevent illegal actions.

Claims (1)

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

Arrange wireless Wi-Fi transmitter and receiver in doors, in conjunction with the wireless communication link that smart machine within doors produces, the CSI burst that the action behavior collecting owner is formed the impact of wireless signal;

Secondly utilize discrete wavelet first to decompose CSI burst, filtering singular value wherein and noise, finally carry out recovery and rebuilding, obtain the input variable of reconstruction signal as whole system;

Again wavelet transform is carried out to reconstruction signal, is decomposed into approximate size factor and detailed size factor, thus make reconstruction signal carry out time-frequency domain conversion;

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

To action sequence according to classifying, sorting technique adopts k-means clustering method, and sets up important sequence respectively in this each classification;

At detection-phase, after the collection of some CSI bursts, denoising and wavelet transform process are carried out to this CSI burst, be decomposed into action sequence, then mate with each classification in the action model set up before, matching process adopts dynamic time warping algorithm, if the match is successful, regards as owner, if mate unsuccessful, regards as illegal invasion.