CN105910601B - A kind of indoor ground magnetic positioning method based on Hidden Markov Model - Google Patents

Abstract

The invention relates to an indoor geomagnetic positioning method based on a hidden Markov model, which includes an offline stage and an online stage. The offline stage includes: dividing a to-be-located area into grids according to a map; using a magnetometer built in a smart phone to locate each grid The center RP measures the geomagnetic field strength data; builds an offline fingerprint database, which consists of N fingerprints, each fingerprint data includes the fingerprint position l _w =[x _w ,y _x ] and the fingerprint vector ξ _w =[μ _w ,σ _w ], the online positioning includes: determining the step length D _i and the motion direction angle Φ _i according to the step size estimation and the magnetometer, predicting the pedestrian position; calculating the state transition probability; estimating the position of the pedestrian after i steps. The present invention can achieve higher indoor positioning accuracy only through the smart phone.

Description

An Indoor Geomagnetic Localization Method Based on Hidden Markov Model

technical field

The invention belongs to the field of indoor positioning of pedestrians by using geomagnetic information, and is particularly aimed at positioning problems in complex indoor environments.

Background technique

High-precision and universally applicable indoor positioning has become more and more important in various fields. To this end, many researchers have proposed many positioning technologies, such as time-of-arrival (TOA), angle-of-arrival (AOA), phase-difference-of-arrival (PDOA), received signal energy (RSS), inertial navigation, and front Fusion of several methods, etc. In addition, more and more signal types are used for indoor positioning, such as WiFi, UWB, Zigbee, etc. These methods have achieved very good positioning results. However, most of the existing positioning methods require additional hardware support, and because the wireless signal will be absorbed by the human body, the wireless signal will be very weak or even not received when the crowd is dense, resulting in poor positioning effect of the positioning system in practical applications. .

Today, many researchers have turned their attention to the geomagnetic field. As an inherent resource of the earth, the geomagnetic field is a vector field with the characteristics of all-day, all-weather and all-region. The geomagnetic field is not affected by the human body and the distribution of the indoor geomagnetic field is mainly determined by the building structure, and the geomagnetic field is very stable after the building structure is determined, so the geomagnetic field has the potential to be applied to high-precision and universally applicable indoor positioning. Due to the influence of the complex indoor environment, especially the influence of reinforced concrete, the indoor geomagnetic field is complex and changeable, and this changing geomagnetic anomaly field can just be used as a kind of fingerprint information corresponding to the location for matching and positioning. In fact, the geomagnetic field has been widely used for indoor positioning. One approach is to use magnetic fields in inertial navigation systems to discern the direction of motion. Another method is to use the geomagnetic field strength as a fingerprint to locate using the fingerprint method. With the help of inertial navigation system, some researchers use dynamic time planning (DTW) algorithm to match and locate the sequence of geomagnetic field strength information at successive moments. This method can achieve high positioning accuracy, but this method is only suitable for positioning in narrow and long areas such as corridors. Some researchers use particle filtering to fuse geomagnetic intensity information with inertial navigation. This method requires a lot of calculations in the actual positioning process to achieve high positioning accuracy.

Although the geomagnetic field has been widely used in indoor positioning, there are still problems that have not been handled very well. First, the geomagnetic field strength is very weak (about tens of uT), and secondly, for the fingerprint method, a single geomagnetic field strength is used as a fingerprint to The resolution to distinguish different locations is too low. Although the three-axis magnetometer can obtain three-dimensional geomagnetic field data, it is natural to think of using all three-axis geomagnetic field strengths to improve the resolution of the geomagnetic fingerprint, but in fact, the three data collected by the magnetometer will follow the sensor coordinate system. changes, therefore, only the total magnetic field strength is available indoors.

SUMMARY OF THE INVENTION

Aiming at the problem that the weak geomagnetic field signal and low resolution in the current indoor positioning technology based on the geomagnetic field are difficult to apply to fingerprint positioning, the present invention provides an indoor positioning technology that increases the resolution of geomagnetic fingerprint information and obtains better positioning accuracy. Geomagnetic positioning method. The technical scheme of the present invention is as follows:

An indoor geomagnetic positioning method based on hidden Markov model, including offline stage and online stage,

The offline data collection phase includes the following steps:

1) Divide the to-be-located area into grids according to the map, and Bw is the _wth grid;

2) Use the built-in magnetometer of the smartphone to measure the geomagnetic field strength data at each grid center RP;

3) Build an offline fingerprint database. The offline fingerprint database consists of N fingerprints. Each fingerprint data includes a fingerprint position l _w =[x _w ,y _x ] and a fingerprint vector ξ _w =[μ _w ,σ _w ], where μ _w and σw are the mean and variance of the geomagnetic field data collected in the _wth grid, respectively, L represents the set composed of all grid centers RP, L={l _w |1≤w≤N};

The online positioning phase includes the following steps:

1) Let the result of the last positioning be Location Represents the position after a person walks i steps. When a person walks one step, the step length D _i and the movement direction angle Φ _i are determined according to the step size estimation and the magnetometer. It is considered that D _i and Φ _i are independent of each other and obey the Gaussian distribution, and calculate separately Probability distribution of D _i and Φ _i ; using Bayesian criterion to predict the probability distribution of pedestrian locations Find the probability distribution of pedestrian locations Set H greater than p _T :

where p _T is the set threshold probability l is the current possible location of the person;

2) Calculate the state transition probability: set the sequence of geomagnetic intensity values stored after the pedestrian walks i steps as O _i : where o _i - _k+1 is the geomagnetic information measured at step i-k+1;

a. Calculate the intersection of H and L as H', where l _i,j represent the possible positions of pedestrians after walking i steps, and the total number of possible positions is N _P :

b. According to the motion information, for each position l _i,j =( _xi,j ,y _i,j ), predict the previous N _s positions, the abscissa is x _i,j , the ordinate is y _i,j , N _s is the length of the sequence, and l _i,j,k =( _xi,j,k ,y _i,j,k ) represents the kth position before l _i,j predicted by PDR:

c. Find the points closest to l _{i, j, k} in the fingerprint position l _w of the offline fingerprint database, and store the mean and variance of the grid center RP as μ _{i, j, k} and σ _{i, j, k respectively} ;

d. For each l _i,j construct two backward sequences: and

e. Determine that the observed value of the geomagnetic field strength conforms to the Gaussian distribution centered on the true value, and calculate the probability that the geomagnetic field strength o _i-k+1 appears at the positions l _{i, j, k} :

f. For each possible position l _i,j , calculate the observation probability b _i,j

g. Calculate the state transition probability a _i,j for each possible position l _i ,j;

3) Use a _i,j as the weight to estimate the pedestrian's position after walking i steps.

The invention is a backward sequence matching positioning method based on hidden Markov model. The method adopts pedestrian pace detection to obtain motion information (PDR), and uses backward sequence matching positioning technology to increase the resolution of geomagnetic fingerprint information. Pedestrian location based on Hidden Markov Model. And it has good robustness for different users (height, weight), so that better positioning accuracy can be obtained and the amount of calculation itself is less. The positioning method of the present invention has been used for more than 2000 simulation experiments in MatLab by using the Monte Carlo method. In the simulation, the test scene is within the range of 20×20×5 meters. The walking direction of pedestrians is random. Set a pedestrian’s walking pace length and walking direction to obtain the real position. The compensation and motion direction angles obtained in the positioning stage are added. Gaussian noise, at the same time, in order to test the influence of environmental factors on the positioning effect, the noise interference of 0.1uT to 1uT is considered in the received geomagnetic field strength signal. The simulation results show that under different noise conditions, the average positioning accuracy is about 1.2 meters. The present invention also collects data on a smartphone (Meizu MX3 terminal) to conduct an actual experiment. The experimental site is located on the 5th floor, Building D, 26th Floor, Tianjin University. Five volunteers with different heights hold the same mobile phone to collect data, and perform data collection on the computer. position on. The experimental results show that for people of different heights, the positioning accuracy is below 1.4 meters. This shows that the present invention not only has high positioning accuracy, but also has good robustness.

Description of drawings

FIG. 1 is a schematic flowchart of the present invention, wherein (a) (b) and (c) respectively represent the position prediction stage, the sequence matching stage and the position estimation stage.

Detailed ways

In order to make the technical solutions of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific examples described herein are only used to explain the invention, but not to limit the invention. As shown in Figure 1, the present invention includes three main steps: position prediction, backward sequence matching and position estimation.

The offline data collection phase includes the following steps:

1) Divide the to-be-located area into grids according to the map, B _w is the wth grid, and its grid center (RP) is l _w .

2) Use a magnetic field measuring device (magnetometer, mobile phone with a magnetic field sensor, etc.) to measure the geomagnetic field strength data at the center of each grid.

3) Build an offline fingerprint database, which consists of N fingerprints. Each fingerprint data includes the fingerprint position l _w =[x _w ,y _x ] and the fingerprint vector ξ _w =[μ _w ,σ _w ], where μ _w and σ _w are the geomagnetic field collected in the wth grid respectively The mean and variance of the data. L represents the set composed of all RPs: L={l _w |1≤w≤N}.

The online positioning phase includes the following steps:

3) Let the result of the last positioning be Location Represents the position after the person has walked i steps. When the step detection mechanism detects that the person has taken a step, the step length D _i and the movement direction angle Φ _i are determined according to the step length estimation and the magnetometer.

a) Assuming that D _i and Φ _i are independent of each other and obey a Gaussian distribution, calculate the probabilities separately:

in, and are the probability distributions of D _i and Φ _i , respectively, σ _d and σ _Φ are the variance of the movement distance and angle, respectively, l _i is the position of this positioning, is the estimated position of the last fix.

b) Using the Bayesian criterion, predict the probability distribution of pedestrian locations

c) Find the probability Set H greater than p _T :

Where p _T is the set threshold probability l is the current possible location of the person.

4) Use the inertial navigation information to match the geomagnetic field strength information collected online to calculate the state transition probability. The sequence of geomagnetic intensity values stored after a pedestrian walks i steps is O _i : where o _i-k+1 is the geomagnetic information measured at the i-k+1th step.

a) Calculate the intersection of H and L as H', where l _i,j represent the possible positions of pedestrians after walking i steps, and the total number of possible positions is N _P :

b) According to the motion information ΣΔl _i , for each position l _i,j =( _xi,j ,y _i,j ), predict the previous N _s positions (the abscissa is x _i,j , the ordinate is y _{i, j} ), N _s is the length of the sequence, in other words, l _i,j,k =( _xi,j,k ,y _i,j,k ) represents the kth before l _i,j predicted by PDR positions (the abscissa is x _i,j,k , the ordinate is y _i,j,k ):

c) Find the points closest to l _i,j,k in the fingerprint library _lw , and store the mean and variance of the RP as μ _i,j,k and σ _{i,j,k respectively} :

where _pw is the location of the RP.

d) For each _li,j construct two backward sequences: and

e) Assuming that the observed value of the geomagnetic field strength conforms to a Gaussian distribution centered on the true value, calculate the probability that the geomagnetic field strength o _i-k+1 appears at the positions l _{i, j, k} :

where μ _i,j,k is the mean and σ _i,j,k is the corresponding variance, and l _i-k+1 is the position of the person after the i-k+1th step.

f) For each possible position l _i,j , compare U _i,j with O _i , calculate the observation probability b _i,j (ie p(o _i |l _i =l _i,j )):

g) Calculate the state transition probability a _i,j for each possible position l _i, j , C is a normalization constant:

5) Use a _{i, j} as weights to estimate the pedestrian's position after walking i steps as:

6) When the system detects pedestrian movement one step again, repeat steps (1) to (3) to estimate the user's position.

This embodiment mainly includes two stages of offline data collection and online matching and positioning.

The offline stage mainly includes the following steps. First, the area to be located is divided into grids of 0.6m×0.6m, and the center of each grid is used as the reference node RP. Then, carry instruments that can measure and record geomagnetic field intensity sensors, such as smartphones, IMUs, etc., and measure and record about 100 sets of geomagnetic intensity data at each grid center (RP) according to the sensor rate. Average the data and calculate the variance. Finally, the position of the RP corresponds to the obtained mean and variance as a fingerprint data, and a fingerprint library is constructed.

In the online stage, taking a smartphone as an example, the user to be located drags the phone while walking with the screen facing up, that is, the Z-axis of the phone is facing up and the Y-axis is facing the direction of movement. Built-in accelerometer to obtain motion information. Steps can be detected according to the Z-axis acceleration data. The Z-axis acceleration data can be integrated first, and then the steps can be determined by using the peak detection. Due to the interference of noise, it is possible to detect two steps in one step movement. Therefore, a threshold time T _MIN can be set. No matter how many steps are detected within this time, the first step is determined as a step. According to the experimental results, it is recommended to set T _MIN to 0.3 seconds. Step length can be estimated from Y-axis acceleration data. The direction of movement can be obtained by using the magnetometer and the acceleration sensor, and the direction of movement can also be obtained by using the angular velocity information obtained by the gyroscope.

According to the simulation and experimental results, if N _s is too small, the positioning accuracy will be affected, and if N _s is too large, too much computation will be generated. Therefore, it is recommended to set N _s between 7 and 15. If P _T is set too large, the system will be equivalent to inertial navigation. If it is too small, the matching range will be the whole map. Therefore, it is recommended to set P _T between 0.4 and 0.65.

Using this method, we performed localization experiments for the present invention among groups of different people. We invited 5 volunteers of different heights to conduct experiments, and the experimental results show that the average accuracy can reach about 1.2 meters.

Claims (1)

1. An indoor geomagnetic positioning method based on hidden Markov model, including offline stage and online stage, The offline data collection phase includes the following steps: 1) Divide the to-be-located area into grids according to the map, and Bw is the _wth grid; 2) Use the built-in magnetometer of the smartphone to measure the geomagnetic field strength data at each grid center RP; 3) Build an offline fingerprint database. The offline fingerprint database consists of N fingerprints. Each fingerprint data includes a fingerprint position l _w =[x _w ,y _x ] and a fingerprint vector ξ _w =[μ _w ,σ _w ], where μ _w and σw are the mean and variance of the geomagnetic field data collected in the _wth grid, respectively, represents the set of all grid centers RP, The online positioning phase includes the following steps: 1) Let the result of the last positioning be Location Represents the position after a person walks i steps. When a person walks one step, the step length D _i and the movement direction angle Φ _i are determined according to the step size estimation and the magnetometer. It is considered that D _i and Φ _i are independent of each other and obey the Gaussian distribution, and calculate separately Probability distribution of D _i and Φ _i ; using Bayesian criterion to predict the probability distribution of pedestrian locations Find the probability distribution of pedestrian locations sets greater than p _T where p _T is the set threshold probability, and l is the current possible location of the person; 2) Calculate the state transition probability: set the sequence of geomagnetic intensity values stored after the pedestrian walks i steps as O _i : where o _i-k+1 is the geomagnetic information measured at step i-k+1; a. Calculate and The intersection of where l _i,j represents the possible positions of pedestrians after walking i steps, and the total number of possible positions is N _P : b. According to the motion information, for each position l _i,j =( _xi,j ,y _i,j ), predict the previous N _s positions, the abscissa is x _i,j , the ordinate is y _i,j , N _s is the length of the sequence, and l _i,j,k =( _xi,j,k ,y _i,j,k ) represents the kth position before l _i,j predicted by PDR: c. Find the points closest to l _{i, j, k} in the fingerprint position l _w of the offline fingerprint database, and store the mean and variance of the grid center RP as μ _{i, j, k} and σ _{i, j, k respectively} ; d. For each l _i,j construct two backward sequences: and e. Determine that the observed value of the geomagnetic field strength conforms to the Gaussian distribution centered on the true value, and calculate the probability that the geomagnetic field strength o _i-k+1 appears at the positions l _{i, j, k} : f. For each possible position l _i,j , calculate the observation probability b _i,j g. Calculate the state transition probability a _i,j for each possible position l _i ,j; 3) Use a _i,j as the weight to estimate the pedestrian's position after walking i steps.