CN105910601A - Indoor geomagnetic 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. The method includes an off-line stage and an on-line stage. The off-line stage includes: dividing a to-be-positioned area into grids according to a map; using a built-in magnetometer of a smartphone to measure geomagnetic field strength data at each grid center RP; constructing an off-line fingerprint database, which is composed of N fingerprint, wherein each fingerprint data includes a fingerprint position lw=[xw, yx] and a fingerprint vector Xiw=[Muw, Sigmaw], and the on-line positioning stage includes: determining a pace length Di and a movement direction angle Phi i according to pace length estimation and the magnetometer, and predicting a pedestrian position; calculating a state transition probability; and estimating the position of a pedestrian walking i step. The method provided by the invention can achieve high indoor positioning precision only by a smartphone.

Description

A kind of indoor ground based on HMM magnetic positioning method

Technical field

The invention belongs to the field utilizing Geomagnetism Information to carry out pedestrian's indoor positioning, especially under indoor complex environment Orientation problem.

Background technology

High accuracy and blanket indoor positioning have seemed more and more important in every field.To this end, a lot of sections The worker of grinding has been proposed for a lot of location technology, as based on the time of advent (TOA), based on angle of arrival (AOA), based on arrival Phase contrast (PDOA), based on the fusion etc. receiving signal energy (RSS), inertial navigation and above several method.Additionally, use Signal type in indoor positioning also gets more and more, and such as WiFi, UWB, Zigbee etc., these methods have been achieved for extraordinary Positioning result.But, existing localization method major part is required for extra hardware supported, and owing to wireless signal can be by people Body absorbs, and the crowd is dense when, wireless signal the faintest can even not receive and causes alignment system in actual applications Locating effect is the best.

Nowadays, Many researchers has turned to earth's magnetic field sight.Earth's magnetic field, as the intrinsic resource of the earth, is a vector , there is round-the-clock, the round-the-clock and feature of full region.Earth's magnetic field is not affected by human body and being distributed of indoor earth's magnetic field Mainly being determined by fabric structure, and earth's magnetic field is highly stable after fabric structure determines, therefore earth's magnetic field is potential should For high accuracy and blanket indoor positioning.Owing to being affected by indoor complex environment, especially armored concrete Impact, indoor earth's magnetic field is complicated and changeable, and the GEOMAGNETIC FIELD of this change just can be as a kind of finger corresponding with position Stricture of vagina information carries out coupling location.In fact, earth's magnetic field has been widely used in indoor positioning.A kind of method is in inertial navigation system System utilize magnetic field distinguish the direction of motion.Another method is to utilize fingerprinting to carry out as a kind of fingerprint geomagnetic field intensity Location.Some researcheres, with the help of inertial navigation system, use dynamic time programming (DTW) algorithm ground to the continuous moment Magnetic field intensity information sequence carries out coupling location.This method can reach the highest positioning precision, but this method is only fitted The region the longest and the narrowest for corridor positions.Some researcheres are also had to utilize particle filter absolute force information and inertia Navigation is merged, and this method needs substantial amounts of calculating to reach higher positioning precision during actual location.

Although earth's magnetic field has been widely used in indoor positioning, but still problematic not process is fine, first earth magnetism Field intensity the faintest (only about tens uT), secondly for fingerprinting, utilize solely magnetic field intensity as fingerprint The resolution distinguishing diverse location is the lowest.Although three axle magnetometers can obtain the earth's magnetic field data of three-dimensional, thinking naturally To utilize the geomagnetic field intensity of whole three axles to improve the resolution of earth magnetism fingerprint, but it practice, magnetometer gather three numbers According to changing along with the change of sensor coordinate system, therefore, the most total utilizable magnetic field intensity in indoor.

Summary of the invention

Should for relatively low being difficult to of the earth's magnetic field weak output signal being currently based in the indoor positioning technologies of earth's magnetic field and resolution For the problem of fingerprinting location, the present invention provides a kind of resolution increasing earth magnetism finger print information, it is thus achieved that preferably positioning accurate The indoor ground magnetic positioning method of degree.Technical scheme is as follows:

A kind of indoor ground based on HMM magnetic positioning method, including off-line phase and on-line stage,

The off-line data collecting stage comprises the following steps:

1) according to the map area to be targeted is divided into grid, B_wIt is the w grid；

2) magnetometer using smart mobile phone built-in measures geomagnetic field intensity data in each grid element center RP；

3) building off-line fingerprint base, off-line fingerprint base is made up of N number of fingerprint, and each finger print data includes fingerprint positions l_w= [x_w,y_x] and fingerprint vector ξ_w=[μ_w,σ_w], wherein μ_wAnd σ_wPutting down of the earth's magnetic field data respectively gathered in the w grid Average and variance, L represents the set that all grid element center RP form, L={l_w|1≤w≤N}；

The tuning on-line stage comprises the following steps:

1) result of order last time location isPositionRepresent the position after people walking i step, when people's walking being detected One step, determines paces length D according to step-size estimation and magnetometer_iWith movement direction angle Φ_i, it is believed that D_iWith Φ_iSeparate also Gaussian distributed, calculates D respectively_iWith Φ_iProbability distribution；Utilize bayesian criterion, it was predicted that the probability distribution of pedestrian positionSeek the probability distribution of pedestrian positionMore than p_TSet H:

$H=\left\{l\right|p(l_i=l|{\hat{l}}_{i-1},D_i,{\mathrm{\Φ}}_i)\≥p_T\},$

Wherein p_TThreshold probability l for arranging is the position that personnel may currently exist；

2) state transition probability is calculated: being located at the absolute force value sequence stored after pedestrian walking i step is O_i:Wherein o_i-_k+1It it is the Geomagnetism Information recorded during the i-th-k+1 step；

A. the common factor calculating H and L is H', wherein l_i,jRepresent pedestrian position that may be present after walking i walks, may The total number of positions existed is N_P:

$H^{\′}=\left\{l_{i,j}\right|l_{i,j}\∈H,l_{i,j}\∈L\}=\{l_{i,1},l_{i,2},...,l_{i,N_P}\};$

B. according to movable information for each position l_i,j=(x_i,j,y_i,j) N before prediction_sIndividual position, abscissa is x_i,j, vertical coordinate is y_i,j, N_sFor the length of sequence, l_i,j,k=(x_i,j,k,y_i,j,k) represent the l predicted by PDR_i,jBefore Kth position:

$x_{i,j,k}=x_{i,j}-\underset{t=i-k+1}{\overset{i}{\&Sigma;}}{d}_t\&CenterDot;{\mathrm{cos\&phi;}}_t,0\&le;k<N_s,$

$y_{i,j,k}=y_{i,j}-\underset{t=i-k+1}{\overset{i}{\&Sigma;}}{d}_t\&CenterDot;{\mathrm{sin\&phi;}}_t,0\&le;k<N_s;$

C. in fingerprint positions l of off-line fingerprint base_wMiddle searching distance l_i,j,kNearest point, and store respectively in this grid The average of heart RP and variance are μ_i,j,kAnd σ_i,j,k；

D. for each l_i,jBuild two backward sequences:With

E. assert that geomagnetic field intensity observation meets the Gauss distribution centered by actual value, calculates at position l_i,j,kOccur Geomagnetic field intensity o_i-k+1Probability:

F. for each possible position l_i,j, calculating observation value probability b_i,j

G. to each position l that may be present_i,jCalculate state transition probability a_i,j；

3) by a_i,jThe position after pedestrian walking i step is estimated as weight.

The present invention is a kind of backward sequences match localization method based on HMM, and the method uses pedestrian's step Cut down detection to obtain movable information (PDR), utilize backward sequences match location technology to increase the resolution of earth magnetism finger print information, Based on the basis of HMM, pedestrian is positioned.And different user'ss (height, body weight) is had very well Robustness, it is hereby achieved that preferably positioning precision and amount of calculation own are less.The localization method of the present invention exists MatLab uses the emulation experiment that monte carlo method has carried out more than 2000 time.In emulation, test scene is 20 × 20 × 5 meters The ranged space in, pedestrian's walking direction is random, sets paces length and the walking direction of pedestrian's walking, obtains true Position, compensation and movement direction angle that positioning stage obtains add Gaussian noise, meanwhile, for test environment factor pair The impact of locating effect, considers the noise jamming of 0.1uT to 1uT in receiving geomagnetic field intensity signal.Simulation result shows, Under conditions of different noises, average positioning precision is about 1.2 meters.The present invention also adopts at smart mobile phone (Meizu MX3 end) Collection data carry out actual experiment, and experimental site is positioned at 26th floors Building D of University Of Tianjin 5th floors, the volunteer of 5 different heights take respectively Same portion mobile phone and carry out data acquisition, and position on computers.Test result indicate that, the people for different heights comes Saying, positioning precision has all reached less than 1.4 meters.It is indicated above that the present invention not only positioning precision is higher, and there is good Shandong Rod.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of the present invention, and wherein, (a) (b) (c) represents position prediction stage, sequences match rank respectively Section and location estimation stage.

Detailed description of the invention

In order to make technical scheme become apparent from, below in conjunction with accompanying drawing and example, the present invention is carried out further Detailed description.Should be appreciated that instantiation described herein is only used for explaining invention, be not intended to limit the present invention.As The present invention shown in Fig. 1 includes three key steps: position prediction, backward sequences match and location estimation.

The off-line data collecting stage comprises the following steps:

1) according to the map area to be targeted is divided into grid, B_wBeing the w grid, its grid element center (RP) is l_w。

2) magnetic field measuring device (magnetometer, the mobile phone etc. of band magnetic field sensor) is used to measure earth magnetism in each grid element center Field intensity data.

3) building off-line fingerprint base, fingerprint base is made up of N number of fingerprint.Each finger print data includes fingerprint positions l_w=[x_w, y_x] and fingerprint vector ξ_w=[μ_w,σ_w], wherein μ_wAnd σ_wIt is respectively the meansigma methods of the earth's magnetic field data gathered in the w grid And variance.L represents the set that all RP form: L={l_w|1≤w≤N}。

The tuning on-line stage comprises the following steps:

3) result of order last time location isPositionRepresent the position after people's walking i step.When paces testing machine System detects people's walking one step, determines paces length D according to step-size estimation and magnetometer_iWith movement direction angle Φ_i。

A) D is assumed_iWith Φ_iSeparate and Gaussian distributed, respectively calculating probability:

$p\left(D_i\right|l_i,{\hat{l}}_{i-1})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\&CenterDot;{\mathrm{\&sigma;}}_d}\&CenterDot;\exp (-\frac{1}{2{\mathrm{\&sigma;}}_d^2}{\left(d_i-d\left(l_i,\widehat{l_{i-1}}\right)\right)}^2),$

$p\left({\mathrm{\&Phi;}}_i\right|l_i,{\hat{l}}_{i-1})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\&CenterDot;{\mathrm{\&sigma;}}_{\mathrm{\&phi;}}}\&CenterDot;\exp (-\frac{1}{2{\mathrm{\&sigma;}}_{\mathrm{\&phi;}}^2}{({\mathrm{\&phi;}}_i-\mathrm{\&phi;}(l_i,\widehat{l_{i-1}}\left)\right)}^2),$

Wherein,WithIt is respectively D_iWith Φ_iProbability distribution, σ_dAnd σ_ΦBe respectively motion away from The variance of walk-off angle degree, l_iThe position positioned for this,Estimation position for last time location.

B) bayesian criterion is utilized, it was predicted that the probability distribution of pedestrian position

$p\left(l_i\right|{\hat{l}}_{i-1},D_i,{\mathrm{\&Phi;}}_i)=p\left(D_i\right|l_i,{\hat{l}}_{i-1})\&CenterDot;p\left({\mathrm{\&Phi;}}_i\right|l_i,{\hat{l}}_{i-1}).$

C) probability is soughtMore than p_TSet H:

$H=\left\{l\right|p(l_i=l|{\hat{l}}_{i-1},D_i,{\mathrm{\&Phi;}}_i)\&GreaterEqual;p_T\},$

Wherein p_TThreshold probability l for arranging is the position that personnel may currently exist.

4) the geomagnetic field intensity information utilizing inertial navigation information and online acquisition is mated, and calculates state transition probability.? After pedestrian walking i step, the absolute force value sequence of storage is O_i:Wherein o_i-k+1It is the i-th-k+ The Geomagnetism Information recorded during 1 step.

A) common factor calculating H and L is H', wherein l_i,jRepresent pedestrian position that may be present after walking i walks, may The total number of positions existed is N_P:

$H^{\&prime;}=\left\{l_{i,j}\right|l_{i,j}\&Element;H,l_{i,j}\&Element;L\}=\{l_{i,1},l_{i,2},...,l_{i,N_P}\}.$

B) according to movable information ∑ Δ l_iFor each position l_i,j=(x_i,j,y_i,j) N before prediction_sIndividual position (horizontal seat It is designated as x_i,j, vertical coordinate is y_i,j), N_sFor the length of sequence, in other words, l_i,j,k=(x_i,j,k,y_i,j,k) represent to be come by PDR The l of prediction_i,j(abscissa is x in kth position before_i,j,k, vertical coordinate is y_i,j,k):

$x_{i,j,k}=x_{i,j}-\underset{t=i-k+1}{\overset{i}{\&Sigma;}}{d}_t\&CenterDot;{\mathrm{cos\&phi;}}_t,0\&le;k<N_s,$

$y_{i,j,k}=y_{i,j}-\underset{t=i-k+1}{\overset{i}{\&Sigma;}}{d}_t\&CenterDot;{\mathrm{sin\&phi;}}_t,0\&le;k<N_s.$

C) at fingerprint base l_wMiddle searching distance l_i,j,kNearest point, and to store the average of this RP respectively with variance be μ_i,j,k And σ_i,j,k:

$\{\begin{array}{c}{\mathrm{\&mu;}}_{i,j,k}=\left\{{\mathrm{\&mu;}}_w\right|\underset{{\mathrm{RP}}_w}{\arg \min }{||l_{i,j,k}-p_w||}_2\}\\ {\mathrm{\&sigma;}}_{i,j,k}=\left\{{\mathrm{\&sigma;}}_w\right|\underset{{\mathrm{RP}}_w}{\arg \min }{||l_{i,j,k}-p_w||}_2\}\end{array},1\&le;k\&le;N_s,$

Wherein p_wPosition for RP.

D) for each l_i,jBuild two backward sequences:With

E) magnetic field intensity observation meets the Gauss distribution centered by actual value potentially, calculates at position l_i,j,kOccur Geomagnetic field intensity o_i-k+1Probability:

$p\left(o_{i-k+1}\right|l_{i-k+1}=l_{i,j,k})=\frac{1}{\sqrt{2\mathrm{\&pi;}}\&CenterDot;{\mathrm{\&sigma;}}_{i,j,k}}\&CenterDot;\exp \left(\frac{1}{2{\mathrm{\&sigma;}}_{i,j,k}^2}{\left(o_{i-k+1}-{\mathrm{\&mu;}}_{i,j,k}\right)}^2\right),$

Wherein μ_i,j,kThe σ for average_i,j,kFor corresponding variance, l_i-k+1It it is the position of personnel after the i-th-k+1 paces.

F) for each possible position l_i,j, compare U_i,jAnd O_i, calculating observation value probability b_i,j(i.e. p (o_i|l_i= l_i,j)):

$b_{i,j}=p\left(o_i\right|l_i=l_{i,j})=p\left(O_i\right|l_i=l_{i,j})=\underset{k=1}{\overset{N_s}{\&Pi;}}p\left(o_{i-k+1}\right|l_{i-k+1}=l_{i,j,k}).$

G) to each position l that may be present_i,jCalculate state transition probability a_i,j, C is normaliztion constant:

5) by a_i,jPosition after estimating pedestrian walking i step as weight is:

${\hat{l}}_i=\underset{a_{i,j}\&Element;A_i,l_{i,j}\&Element;H^{\&prime;}}{\&Sigma;}{a}_{i,j}\&CenterDot;l_{i,j}$

6) pedestrian movement one step again detected when system, repeat (1) to (3) step, estimation customer location.

The present embodiment mainly includes off-line data collecting and On-line matching two stages of location.

Off-line phase mainly includes the following steps that, first area to be targeted is divided into the grid of 0.6m × 0.6m, with often Individual net center of a lattice is as reference mode RP.Then, the instrument that can measure and record geomagnetic field intensity sensor is carried, such as intelligence Energy mobile phone, IMU etc. are different each grid element center (RP) surveying record absolute force data about 100 groups according to sensor rate. Data are averaged and calculates variance.Finally, the position of RP and obtained meansigma methods and variance to should be used as a finger Stricture of vagina data, build fingerprint base.

In on-line stage, as a example by smart mobile phone, user to be positioned walks with hands mobile phone in tow, and screen is placed, i.e. upward The Z axis of the mobile phone direction of Y-axis orientation movements upward, such people in the process of walking, can be according to the acceleration of mobile phone Sensor obtains movable information.The detection of paces can be carried out according to Z axis acceleration information, first Z axis can be accelerated the number of degrees According to being integrated, then utilize peakvalue's checking it is determined that paces.Interference due to noise etc., it is possible to transport at paces Twice paces is detected, to this end, a threshold time T can be set in Dong_MIN, within this time, several paces no matter detected, All judge the first step as paces, according to experimental result, it is proposed that T_MINIt is set as 0.3 second.Permissible according to Y-axis acceleration information Estimate paces length.Utilize magnetometer and acceleration transducer to obtain the direction of motion, it is also possible to utilize gyroscope to obtain Angular velocity information obtains the direction of motion.

According to emulation and experimental result, N_sThe least will affect positioning precision, N_sCross conference and produce too much amount of calculation, therefore Suggestion N_sIt is set as between 7 to 15.And P_TSetting too conference causes system to be equivalent to inertial navigation, the least, can cause mating model Enclose for full map, it is therefore proposed that P_TIt is set as between 0.4 to 0.65.

Adopting in this way, we have carried out the positioning experiment between many groups different people to the present invention.We have invited 5 The volunteer of different heights tests, and experimental result display mean accuracy can reach about 1.2 meters.

Claims (1)

1. indoor ground based on a HMM magnetic positioning method, including off-line phase and on-line stage,

The off-line data collecting stage comprises the following steps:

1) according to the map area to be targeted is divided into grid, B_wIt is the w grid；

2) magnetometer using smart mobile phone built-in measures geomagnetic field intensity data in each grid element center RP；

3) building off-line fingerprint base, off-line fingerprint base is made up of N number of fingerprint, and each finger print data includes fingerprint positions l_w=[x_w, y_x] and fingerprint vector ξ_w=[μ_w,σ_w], wherein μ_wAnd σ_wIt is respectively the meansigma methods of the earth's magnetic field data gathered in the w grid And variance, L represents the set that all grid element center RP form, L={l_w|1≤w≤N}；

The tuning on-line stage comprises the following steps:

1) result of order last time location isPositionRepresent the position after people walking i step, when people's walking one step being detected, Paces length D is determined according to step-size estimation and magnetometer_iWith movement direction angle Φ_i, it is believed that D_iWith Φ_iSeparate and obey Gauss distribution, calculates D respectively_iWith Φ_iProbability distribution；Utilize bayesian criterion, it was predicted that the probability distribution of pedestrian positionSeek the probability distribution of pedestrian positionMore than p_TSet H:

$H=\left\{l\right|p(l_i=l|{\hat{l}}_{i-1},D_i,{\mathrm{\&Phi;}}_i)\&GreaterEqual;p_T\},$

Wherein p_TThreshold probability l for arranging is the position that personnel may currently exist；

2) state transition probability is calculated: being located at the absolute force value sequence stored after pedestrian walking i step is O_i:Wherein o_i-k+1It it is the Geomagnetism Information recorded during the i-th-k+1 step；

A. the common factor calculating H and L is H', wherein l_i,jRepresent pedestrian position that may be present after walking i walks, it is understood that there may be Total number of positions be N_P:

$H^{\&prime;}=\left\{l_{i,j}\right|l_{i,j}\&Element;H,l_{i,j}\&Element;L\}=\{l_{i,1},l_{i,2},...,l_{i,N_P}\};$

B. according to movable information for each position l_i,j=(x_i,j,y_i,j) N before prediction_sIndividual position, abscissa is x_i,j, vertical Coordinate is y_i,j, N_sFor the length of sequence, l_i,j,k=(x_i,j,k,y_i,j,k) represent the l predicted by PDR_i,jKth before Position:

$x_{i,j,k}=x_{i,j}-\underset{t=i-k+1}{\overset{i}{\&Sigma;}}{d}_t\&CenterDot;{\mathrm{cos\&phi;}}_t,0\&le;k<N_s,$

$y_{i,j,k}=y_{i,j}-\underset{t=i-k+1}{\overset{i}{\&Sigma;}}{d}_t\&CenterDot;{\mathrm{sin\&phi;}}_t,0\&le;k<N_s;$

C. in fingerprint positions l of off-line fingerprint base_wMiddle searching distance l_i,j,kNearest point, and store this grid element center RP respectively Average and variance be μ_i,j,kAnd σ_i,j,k；

D. for each l_i,jBuild two backward sequences:With

E. assert that geomagnetic field intensity observation meets the Gauss distribution centered by actual value, calculates at position l_i,j,kEarth magnetism occurs Field intensity o_i-k+1Probability:

F. for each possible position l_i,j, calculating observation value probability b_i,j

G. to each position l that may be present_i,jCalculate state transition probability a_i,j；

3) by a_i,jThe position after pedestrian walking i step is estimated as weight.