阅读说明：本技术 一种兼顾非视距识别的无线信号室内定位 (Wireless signal indoor positioning considering non-line-of-sight identification ) 是由 王天保 林城誉 车洪峰 王大维 于 2021-08-26 设计创作，主要内容包括：本发明公开了一种兼顾非视距识别的无线信号室内定位,其具体定位方法具体包括以下步骤：S1、在待定位的室内环境中,网格化室内空间环境,部署AP后利用全站仪标定不同采样点以及AP的位置坐标；训练数据模块和测试数据模块的输出端分别与第一预处理模块和第二预处理模块的输入端连接,并且第一预处理模块和第二预处理模块的输出端均与堆栈降噪编码器模块的输入端连接,并且堆栈降噪编码器模块的输出端与二分类器模块的输入端连接,本发明涉及室内定位、非视距识别WiFi技术领域。该兼顾非视距识别的无线信号室内定位,解决了基于wifi的NLOS环境识别问题,提出了结合非视距识别的粒子滤波融合策略,提高了定位结果的可靠。(The invention discloses a wireless signal indoor positioning method considering non-line-of-sight recognition. The specific positioning method specifically comprises the following steps: S1, meshing an indoor space environment in an indoor environment to be positioned, and calibrating different sampling points and position coordinates of an AP by using a total station after the AP is deployed; wherein the output ends of the training data module and the test data module are respectively connected with the input ends of the first preprocessing module and the second preprocessing module; wherein the output ends of the first preprocessing module and the second preprocessing module are both connected with the input end of the stack noise reduction encoder module, and the output end of the stack noise reduction encoder module is connected with the input end of the binary classifier module. The invention relates to the technical field of indoor positioning and non-line-of-sight recognition WiFi. According to the wireless signal indoor positioning considering non-line-of-sight recognition, the problem of NLOS environment recognition based on WiFi is solved, a particle filtering fusion strategy combining non-line-of-sight recognition is provided, and the reliability of a positioning result is improved.)

1. A wireless signal indoor positioning method considering non-line-of-sight identification specifically comprises the following steps:

s1, gridding an indoor space environment in an indoor environment to be positioned, and calibrating different sampling points and position coordinates of the AP by using a total station after the AP is deployed;

s2, collecting wifi data and MEMS magnetometer data at different sampling points in the indoor environment by using the smart phone, and recording corresponding position coordinates and corresponding non-line-of-sight identification tags, wherein the specific format is as follows:

{(x,y),RSSI₁,RSSI₂,RSSI₃…,RSSI_n,M_x,M_y,M_zlabel N ∈ (1, N), storing the data in the database, where (x, y) is the position coordinate of the sampling point, RSSI is the wifi signal strength, M_x,M_y,M_zThe method comprises the steps that three-axis data of geomagnetism at a sampling point are obtained, n is the number of APs in an indoor environment, label is a non-line-of-sight identification label, 1 is line-of-sight, and 0 is non-line-of-sight;

s3, normalizing the constructed fingerprint data to obtain a normalized fingerprint database, aiming at the problem of unstable positioning result caused by fingerprint data fluctuation in an indoor environment, the SDAE is a stack noise reduction self-encoder network SDAE, the input and output of the SDAE are data with the same dimensionality, the input of the SDAE is fingerprint data with random noise added, the output of the SDAE is accurate fingerprint data, and the SDAE model is formed by stacking a plurality of noise reduction self-encoder networks DAE, wherein the network structure is shown in FIG. 1, the input data with noise added can also prevent overfitting of the model;

the noise of the input data in the model is increased through qDPost-vocalisation to obtain new input data x_nCarrying out unsupervised training of the first DAE encoder to obtain implicit characteristics, using the output of the first layer DAE as the input of the next layer DAE, repeating the training for multiple DAEs, finally fine-tuning the whole network by a back propagation method, and finally outputting y_rIs a deep characteristic;

s4, after the deep features are obtained, connecting the network model with a classifier II, namely realizing the identification of the non-line-of-sight and line-of-sight environments;

s5, under the indoor environment, calculating corresponding parameters for the line-of-sight environment and the non-line-of-sight environment by using a wireless signal free space attenuation model, wherein the free space attenuation model formula is as follows:

(1) its logarithmic form is:

(2) neglecting the antenna gain, let P_d＝P_r(d)[dB]，The above equation becomes:

(3) in the formula:

G_t、G_rrespectively, the gain of the receiving and transmitting antenna;

λ is the radio wavelength;

d is the distance between the transmitting and receiving antennas;

is the path loss (usually d) of the reference distance₀＝1m)

S6, after training to obtain a sight distance and non-sight distance signal propagation model, initializing a particle filter algorithm, and firstly, initializing position coordinates, the number and the moving step length of particles; then, taking the data of the accelerometer and the magnetometer in the MEMS sensor as data support of a state transition equation; the position resolved by the AP ranging information is used as an observation result, and it is worth noting that the sight distance environment and the non-sight distance environment use different models to obtain corresponding observation results, and finally continuous position estimation of the user is completed.

2. The wireless signal indoor positioning method compatible with non-line-of-sight identification as claimed in claim 1, wherein: the non-line-of-sight recognition model specifically comprises the following modules: the system comprises a training data module, a test data module, a first preprocessing module, a second preprocessing module, a stack noise reduction encoder module, a classifier module and an MLOS identification model.

3. The wireless signal indoor positioning method compatible with non-line-of-sight identification as claimed in claim 2, wherein: the output ends of the training data module and the test data module are respectively connected with the input ends of the first preprocessing module and the second preprocessing module, the output ends of the first preprocessing module and the second preprocessing module are respectively connected with the input end of the stack noise reduction encoder module, the output end of the stack noise reduction encoder module is connected with the input end of the two classifier modules, and the output ends of the two classifier modules are connected with the input end of the MLOS recognition model.

Technical Field

The invention relates to the technical field of indoor positioning and non-line-of-sight identification WiFi, in particular to wireless signal indoor positioning considering non-line-of-sight identification.

Background

Low cost indoor positioning solutions using wireless signals (e.g., WiFi and bluetooth) that are easily distorted by the presence of dynamic objects, room temperature, dust, and even humidity have long been studied, and furthermore, shadow fading and multipath propagation severely hamper the reliability of ranging signal strength.

The current state of the art for radio-based positioning technology encompasses four different categories: (1) a Received Signal Strength Indicator (RSSI); (2) angle of arrival (AOA); (3) time of arrival (TOA); (4) physical layer information (PHY), except for the first item (1), requires specialized hardware to obtain range measurements from WiFi Access Points (APs), a requirement that limits the applicability of these methods to non-commercial applications, and once range measurements are obtained, the device location can be determined using positioning techniques such as spherical or hyperbolic positioning.

In order to improve the positioning accuracy, filtering methods such as kalman filtering and particle filtering are generally used to combine the measurement results of multiple sensors such as accelerometers, magnetometers and cameras. The method based on wireless signal ranging depends on the line-of-sight environment to a great extent, and high-precision distance measurement is difficult to realize under the non-line-of-sight environment.

The radio fingerprint identification and positioning technology usually needs to construct a fingerprint database of an environment in advance, collect fingerprint data after gridding an indoor environment, and realize the position determination of a user through a matching method in an online stage.

Disclosure of Invention

Technical problem to be solved

Aiming at the defects of the prior art, the invention provides wireless signal indoor positioning considering non-line-of-sight identification, and solves the problems in the background technology.

(II) technical scheme

The patent provides a non-line-of-sight identification method based on stack noise reduction self-encoder to the wifi data and MEMS self sensor data in the environment are obtained as the carrier to the smart mobile phone, combine the particle filter means to realize the intelligent switching and the integration of different range finding models, improve the reliability and the stability of location result.

A wireless signal indoor positioning method considering non-line-of-sight identification specifically comprises the following steps:

s1, gridding an indoor space environment in an indoor environment to be positioned, and calibrating different sampling points and position coordinates of the AP by using a total station after the AP is deployed;

s2, collecting wifi data and MEMS magnetometer data at different sampling points in the indoor environment by using the smart phone, and recording corresponding position coordinates and corresponding non-line-of-sight identification tags, wherein the specific format is as follows:

{(x,y),RSSI₁,RSSI₂,RSSI₃…,RSSI_n,M_x,M_y,M_zlabel N ∈ (1, N), storing the data in the database, where (x, y) is the position coordinate of the sampling point, RSSI is the wifi signal strength, M_x,M_y,M_zThe method comprises the steps that three-axis data of geomagnetism at a sampling point are obtained, n is the number of APs in an indoor environment, label is a non-line-of-sight identification label, 1 is line-of-sight, and 0 is non-line-of-sight;

s3, normalizing the constructed fingerprint data to obtain a normalized fingerprint database, aiming at the problem of unstable positioning result caused by fingerprint data fluctuation in an indoor environment, the SDAE is a stack noise reduction self-encoder network SDAE, the input and output of the SDAE are data with the same dimensionality, the input of the SDAE is fingerprint data with random noise added, the output of the SDAE is accurate fingerprint data, and the SDAE model is formed by stacking a plurality of noise reduction self-encoder networks DAE, wherein the network structure is shown in FIG. 1, the input data with noise added can also prevent overfitting of the model;

in the model, new input data x is obtained after qD noise addition of the input data_nCarrying out unsupervised training of the first DAE encoder to obtain implicit characteristics, using the output of the first layer DAE as the input of the next layer DAE, repeating the training for multiple DAEs, finally fine-tuning the whole network by a back propagation method, and finally outputting y_rIs a deep characteristic;

s4, after the deep features are obtained, connecting the network model with a classifier II, namely realizing the identification of the non-line-of-sight and line-of-sight environments;

s5, under the indoor environment, calculating corresponding parameters for the line-of-sight environment and the non-line-of-sight environment by using a wireless signal free space attenuation model, wherein the free space attenuation model formula is as follows:

(1) its logarithmic form is:

(2) neglecting the antenna gain, let P_d＝P_r(d)[dB]，The above equation becomes:

(3) in the formula:

G_t、G_rrespectively, the gain of the receiving and transmitting antenna;

λ is the radio wavelength;

d is the distance between the transmitting and receiving antennas;

is the path loss (usually d) of the reference distance₀＝1m)

S6, after training to obtain a sight distance and non-sight distance signal propagation model, initializing a particle filter algorithm, and firstly, initializing position coordinates, the number and the moving step length of particles; then, taking the data of the accelerometer and the magnetometer in the MEMS sensor as data support of a state transition equation; the position resolved by the AP ranging information is used as an observation result, and it is worth noting that the sight distance environment and the non-sight distance environment use different models to obtain corresponding observation results, and finally continuous position estimation of the user is completed.

Preferably, the non-line-of-sight recognition model specifically includes the following modules: the system comprises a training data module, a test data module, a first preprocessing module, a second preprocessing module, a stack noise reduction encoder module, a classifier module and an MLOS identification model.

Preferably, the output ends of the training data module and the test data module are respectively connected with the input ends of the first preprocessing module and the second preprocessing module, the output ends of the first preprocessing module and the second preprocessing module are respectively connected with the input end of the stack noise reduction encoder module, the output end of the stack noise reduction encoder module is connected with the input end of the classifier module, and the output end of the classifier module is connected with the input end of the MLOS recognition model.

(III) advantageous effects

The invention provides a wireless signal indoor positioning method considering non-line-of-sight identification. The method has the following beneficial effects:

(1) the problem of NLOS environment identification based on wifi is solved;

(2) a particle filter fusion strategy combined with non-line-of-sight identification is provided, and the reliability of a positioning result is improved.

Drawings

FIG. 1 is a SDAE network model structure of the stacked noise reduction self-encoder of the present invention;

FIG. 2 is a non-line-of-sight recognition model of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

Referring to fig. 1-2, an embodiment of the present invention provides a technical solution: a wireless signal indoor positioning method considering non-line-of-sight identification specifically comprises the following steps:

s1, in the indoor environment to be positioned, gridding the indoor space environment, after AP is deployed, calibrating different sampling points and the position coordinates of the AP by using a Total Station, wherein the Total Station, namely a Total Station type Electronic distance meter (Electronic Total Station), is a high-tech measuring instrument integrating light collection, mechanics and electricity, and is a surveying instrument system integrating horizontal angle, vertical angle, distance (slant distance, horizontal distance) and height difference measurement functions. Compared with the optical theodolite, the electronic theodolite changes the optical scale into the photoelectric scanning scale, and replaces manual optical micrometer reading with automatic recording and displaying reading, so that the angle measurement operation is simplified, and the generation of reading errors can be avoided. The total station is called because the instrument can be arranged once to complete all measurement work on the station. The method is widely applied to the field of precision engineering measurement or deformation monitoring of aboveground large buildings, underground tunnel construction and the like;

s2, collecting wifi data and MEMS magnetometer data at different sampling points in an indoor environment by using a smart phone, and recording corresponding position coordinates and corresponding non-line-of-sight identification tags, wherein the smart phone is the same as a personal computer, has an independent operating system and an independent operating space, can be used by a user to install programs provided by third-party service providers such as software, games and navigation, and can realize a general name of a mobile phone type accessed by a wireless network through a mobile communication network, and the specific format is as follows:

{(x,y),RSSI₁,RSSI₂,RSSI₃…,RSSI_n,M_x,M_y,M_zlabel N ∈ (1, N), storing the data in the database, where (x, y) is the position coordinate of the sampling point, RSSI is the wifi signal strength, M_x,M_y,M_zThe method comprises the steps that three-axis data of geomagnetism at a sampling point are obtained, n is the number of APs in an indoor environment, label is a non-line-of-sight identification label, 1 is line-of-sight, and 0 is non-line-of-sight;

and S3, normalizing the constructed fingerprint data to obtain a normalized fingerprint database, wherein the database is a warehouse for organizing, storing and managing data according to a data structure. The network structure is shown in figure 1, the input data of the increased noise can also prevent overfitting of the model, and a plurality of DAEs are stacked to form an SDAE model;

in the model, new input data x is obtained after qD noise addition of the input data_nCarrying out unsupervised training of the first DAE encoder to obtain implicit characteristics, using the output of the first layer DAE as the input of the next layer DAE, repeating the training for multiple DAEs, finally fine-tuning the whole network by a back propagation method, and finally outputting y_rIs a deep characteristic;

s4, after the deep features are obtained, connecting the network model with a classifier II, namely realizing the identification of the non-line-of-sight and line-of-sight environments;

s5, under the indoor environment, calculating corresponding parameters for the line-of-sight environment and the non-line-of-sight environment by using a wireless signal free space attenuation model, wherein the free space attenuation model formula is as follows:

(1) its logarithmic form is:

(2) neglecting the antenna gain, let P_d＝P_r(d)[dB]，The above equation becomes:

(3) in the formula:

G_t、G_rrespectively, the gain of the receiving and transmitting antenna;

λ is the radio wavelength;

d is the distance between the transmitting and receiving antennas;

is the path loss (usually d) of the reference distance₀＝1m)

S6, after training to obtain a sight distance and non-sight distance signal propagation model, initializing a particle filter algorithm, and firstly, initializing position coordinates, the number and the moving step length of particles; then, taking the data of the accelerometer and the magnetometer in the MEMS sensor as data support of a state transition equation; the position resolved by the AP ranging information is used as an observation result, and the observation results are obtained by using different models in a line-of-sight environment and a non-line-of-sight environment, so that the continuous position estimation of a user is finally completed, the MEMS sensor, namely a micro-electro-mechanical system, is the advanced research field of multidisciplinary intersection developed on the basis of the microelectronic technology, relates to various disciplines and technologies such as electronics, machinery, materials, physics, chemistry, biology, medicine and the like, has wide application prospect, wherein the MEMS sensor accounts for a large proportion, the MEMS sensor is a novel sensor manufactured by adopting microelectronic and micro-machining technology, compared with the traditional sensor, the sensor has the characteristics of small volume, light weight, low cost, low power consumption, high reliability, suitability for batch production, easy integration and realization of intellectualization, and simultaneously, feature sizes on the order of microns make it possible to perform functions that some conventional mechanical sensors cannot achieve.

In the invention, the non-line-of-sight recognition model specifically comprises the following modules: the system comprises a training data module, a test data module, a first preprocessing module, a second preprocessing module, a stack noise reduction encoder module, a classifier module and an MLOS (multi-level operating system) identification model, wherein the stack is a data structure. A stack is a data structure in which data items are arranged in order, and data items can only be inserted and deleted at one end, called the top of the stack (top). In the application of a single chip microcomputer, a stack is a special storage area, the main function is to temporarily store data and addresses, the data and the addresses are usually used for protecting breakpoints and sites, an automatic encoder is a neural network which can reproduce input signals as far as possible, an output vector and an input vector of the automatic encoder are in the same dimension, and the representation of one data or the effective encoding of original data is usually learned through a hidden layer according to a certain form of the input vector. It is noted that the self-encoder is a non-linear feature extraction method that does not use class labels, and as far as the method is concerned, the feature extraction aims to preserve and obtain a better representation of the information, rather than to perform classification tasks, although sometimes these two goals are related, a typical simplest auto-encoder has an input layer representing the original data or input feature vectors, a hidden layer representing feature transformations, and an output layer matched to the input layer for information reconstruction, as shown in figure one. When the number of hidden layers is greater than 1, the automatic encoder is regarded as a deep structure, which is called a stacked automatic encoder, and the automatic encoder is generally called a structure in which the number of hidden layers is 1. The main objective of an auto-encoder is to make the input value and the output value equal, so the input is first encoded, passed through the activation function, and then decoded, and the weights of the encoding layer and the decoding layer are usually taken as transpose matrices of each other, i.e. this is the core idea of the auto-encoder: the coding is carried out firstly, then the decoding is carried out, and the front and the back are kept unchanged.

In the invention, the output ends of a training data module and a test data module are respectively connected with the input ends of a first preprocessing module and a second preprocessing module, the output ends of the first preprocessing module and the second preprocessing module are respectively connected with the input end of a stack noise reduction encoder module, the output end of the stack noise reduction encoder module is connected with the input end of a two-classifier module, and the output end of the two-classifier module is connected with the input end of an MLOS recognition model.

To sum up the above

The invention mainly solves the problems that: a non-line-of-sight identification model based on a stack denoising auto-encoder; a multi-information particle filter fusion framework based on a non-line-of-sight recognition model; lightweight indoor positioning solution idea based on AP and smart phone MEMS sensor.