CN106714110B - Wi-Fi position fingerprint map automatic construction method and system - Google Patents

Wi-Fi position fingerprint map automatic construction method and system Download PDF

Info

Publication number
CN106714110B
CN106714110B CN201710039771.5A CN201710039771A CN106714110B CN 106714110 B CN106714110 B CN 106714110B CN 201710039771 A CN201710039771 A CN 201710039771A CN 106714110 B CN106714110 B CN 106714110B
Authority
CN
China
Prior art keywords
map
data
sample
indoor
distance
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201710039771.5A
Other languages
Chinese (zh)
Other versions
CN106714110A (en
Inventor
李清泉
周宝定
朱家松
涂伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen University
Original Assignee
Shenzhen University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen University filed Critical Shenzhen University
Priority to CN201710039771.5A priority Critical patent/CN106714110B/en
Publication of CN106714110A publication Critical patent/CN106714110A/en
Application granted granted Critical
Publication of CN106714110B publication Critical patent/CN106714110B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/023—Services making use of location information using mutual or relative location information between multiple location based services [LBS] targets or of distance thresholds
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/029—Location-based management or tracking services

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)
  • Navigation (AREA)

Abstract

The invention discloses a method and a system for automatically constructing a Wi-Fi position fingerprint map, wherein the method comprises the following steps: acquiring crowdsourcing data, constructing a Wi-Fi position fingerprint map by using a pedestrian dead reckoning method and a machine learning method, and screening Wi-Fi intensity information with position labels according to an indoor map; and taking the Wi-Fi intensity information with the position label as a training sample of the Gaussian process to obtain a functional relation between the signal intensity and the position information, solving the hyperparameter in the Gaussian process, and predicting to obtain the Wi-Fi intensity information of the area with poor indoor map constraint according to the hyperparameter. The method adopts Gaussian process regression, and predicts the Wi-Fi position fingerprint of an open area based on the Wi-Fi position fingerprint information of an area with better map constraint, so that the automatic construction of the Wi-Fi position fingerprint map of the whole indoor area is realized.

Description

Wi-Fi position fingerprint map automatic construction method and system
Technical Field
The invention relates to the technical field of location services, in particular to a method and a system for automatically constructing a Wi-Fi location fingerprint map.
Background
The Wi-Fi position fingerprint method is a common indoor positioning method, can utilize the existing wireless local area network infrastructure, can realize positioning through a smart phone, does not need additional equipment added by a user, and is most widely applied.
The existing Wi-Fi position fingerprint method comprises two steps of off-line fingerprint acquisition and on-line positioning. The purpose of offline fingerprint acquisition is to build a Wi-Fi position fingerprint database of an indoor area. In the off-line acquisition stage, data needs to be acquired at each acquisition point for a period of time to improve the quality of the location fingerprint database. And in the on-line positioning stage, the Wi-Fi signal intensity information acquired by the user in real time is matched and compared with the information in the Wi-Fi position fingerprint database by using a positioning algorithm, so that the position of the user is estimated.
According to the Wi-Fi position fingerprint positioning principle, whether the Wi-Fi position fingerprint database is accurately constructed is very important. In the off-line acquisition stage, a large amount of acquisition point samples are needed, the workload of data acquisition is large, especially for large-scale indoor areas, the acquisition work of position fingerprint samples needs to consume a large amount of manpower and material resources, and the large-scale popularization and application of the Wi-Fi position fingerprint method are severely limited.
The existing Wi-Fi position fingerprint automatic construction and updating method uses behavior map matching to automatically construct a Wi-Fi position fingerprint map through crowdsourcing track data. However, the method can only obtain the position fingerprint of the area with better map constraint, and for the wide area, the pedestrian dead reckoning method of the smart phone has poor precision and cannot obtain the position label at the moment of collecting the position fingerprint, so the application of the method is greatly limited.
Accordingly, the prior art is yet to be improved and developed.
Disclosure of Invention
In view of the defects of the prior art, the invention aims to provide a method and a system for automatically constructing a Wi-Fi position fingerprint map, and aims to solve the problems that in the prior art, only position fingerprints of a region with better map constraint can be obtained, and for an open region, a position label at the time of acquiring the position fingerprints cannot be obtained due to the poor accuracy of a pedestrian dead reckoning method of a smart phone.
The technical scheme of the invention is as follows:
a Wi-Fi position fingerprint map automatic construction method, wherein the method comprises the following steps:
A. acquiring crowdsourcing data, constructing a Wi-Fi position fingerprint map by using a pedestrian dead reckoning method and a machine learning method, and screening Wi-Fi intensity information with position labels according to an indoor map;
B. the Wi-Fi intensity information with the position labels is used as a training sample of a Gaussian process to obtain a functional relation between signal intensity and position information, a hyper-parameter in the Gaussian process is solved, and Wi-Fi intensity information of an indoor map area with poor constraint is obtained according to the hyper-parameter prediction; when the distance between the indoor map obstacles is larger than a preset distance threshold value or the distance between the indoor map wall and the obstacles is larger than a preset distance threshold value, the area is an area with poor indoor map constraint.
The Wi-Fi position fingerprint map automatic construction method comprises the following specific steps:
a1, acquiring related crowdsourcing data;
a2, calculating and obtaining a relative motion track of the user by using a pedestrian dead reckoning method based on the crowdsourcing data;
a3, identifying the specific behaviors of a user comprising a plurality of behavior types through a machine learning method;
a4, constructing a behavior sequence model formed by the type of the specific behaviors in the relative motion trail and the relative spatial relationship among the specific behaviors;
a5, generating a dotted line model of the indoor map, wherein the point is the position where the specific action occurs, and the line is the edge of the connection point;
a6, matching the dotted line model with the behavior sequence model through a hidden Markov model to obtain indoor position coordinate information of a relative motion track;
a7, constructing a Wi-Fi position fingerprint map of an area with better map constraint based on the indoor position coordinate information and crowdsourcing data; when the distance between the obstacles of the indoor map is smaller than or equal to the distance threshold value, or the distance between the wall of the indoor map and the obstacles is smaller than or equal to the distance threshold value, the area is a better indoor map constraint area;
and A8, screening Wi-Fi strength information with position labels according to the indoor map, and taking the Wi-Fi strength information as a sample.
The Wi-Fi position fingerprint map automatic construction method comprises the following steps:
b1, marking the sample points included in the sample as p1,...,pi,...pnH, the Wi-Fi signal strength of the sample point is noted as f, an
Figure GDA0002200253550000031
Modeling Wi-Fi signal strength and position information of a sample point according to a Gaussian process to obtain a functional relation of modeling of the Wi-Fi signal strength and the position information; wherein, when the functional relationship is denoted as F, the distribution of F is:
Figure GDA0002200253550000032
wherein m (-) is a mean function of the Gaussian process, and k (-) is a covariance function of the Gaussian process;
b2, modeling the relationship between Wi-Fi signal strength and distance according to the signal propagation model to obtain the strength f (p) of the received signali)＝γ-10αlog10(||pi-pS||)+Ψ(pi,ps) Wherein gamma is constant, α is signal path attenuation coefficient, psi (p)i,ps) Is to describe the Wi-Fi access point position psAnd Wi-Fi access point location piA location-dependent parameter of signal attenuation therebetween;
b3, according to the training sample Sj＝{(p1,f(p1)),(p2,f(p2)),...,(pn,f(pn) ) } and two-dimensional coordinates comprising training samples
Figure GDA0002200253550000033
Obtaining Wi-Fi signal strength contained in training samples
Figure GDA0002200253550000034
And includes Wi-Fi signal strength in training samples
Figure GDA0002200253550000035
F in (1)jObey a multidimensional joint Gaussian distribution and is noted
Figure GDA0002200253550000041
Wherein，
Figure GDA0002200253550000042
C is an n × n matrix, each element
Figure GDA0002200253550000043
When i is j, deltaij1, in other cases deltaij＝0；dcIs the distance of the correlation, and,
Figure GDA0002200253550000044
for the attenuation variance,. epsilon.for the measurement noise,. sigmaεIs Gaussian noise, pi、pjTwo-dimensional coordinates representing a training sample;
b4, position p to be predicted*The signal strength of (A) is expressed as
Figure GDA0002200253550000045
And combining y and f into a multidimensional Gaussian variable [ f with (n +1) multiplied by 1 columns; y is]：
Figure GDA0002200253550000046
Derived from conditional distribution of Gaussian random variables
Figure GDA0002200253550000047
Wherein
Figure GDA0002200253550000048
Figure GDA0002200253550000049
The Wi-Fi intensity average value of the Wi-Fi position fingerprint map without position labels,
Figure GDA00022002535500000410
and the Wi-Fi intensity variance of the Wi-Fi position fingerprint map without position labels is obtained, k represents the covariance, and n represents the number of training samples.
The Wi-Fi position fingerprint map automatic construction method, wherein the crowdsourcing data comprises: crowd-sourced user acceleration data, gyroscope data, magnetometer data, barometer data, and Wi-Fi data.
The Wi-Fi position fingerprint map automatic construction method comprises the following specific steps of:
calculating the number of the steps of the user through a peak detection algorithm based on the acceleration data;
estimating the walking step length of the user through a step frequency step length model;
obtaining the advancing direction of the user according to the magnetometer data;
calculating to obtain a forward distance according to the number of the forward steps and the walking step length; and generating a relative motion track of the user according to the advancing distance and the advancing direction.
The Wi-Fi position fingerprint map automatic construction method comprises the following specific steps of:
collecting sample time sequence data containing a plurality of specific behaviors;
segmenting the sample timing data through a sliding window of a predetermined length to obtain a specific behavior sample;
extracting features of the specific behavior sample;
training a classifier for classifying a specific behavior based on the features of the specific behavior sample;
the crowd-sourced data is segmented using the same sliding window and classified using a trained classifier, generating a type of a particular behavior of the crowd-sourced data.
A Wi-Fi position fingerprint map automatic construction system, wherein, include:
the training sample acquisition module is used for acquiring crowdsourcing data, constructing a Wi-Fi position fingerprint map by using a pedestrian dead reckoning method and a machine learning method, and screening Wi-Fi intensity information with position labels according to an indoor map;
the prediction module is used for taking the Wi-Fi intensity information with the position label as a training sample of a Gaussian process to obtain a functional relation between signal intensity and position information, solving a hyper-parameter in the Gaussian process, and predicting to obtain the Wi-Fi intensity information of an area with poor indoor map constraint according to the hyper-parameter; when the distance between the indoor map obstacles is larger than a preset distance threshold value or the distance between the indoor map wall and the obstacles is larger than a preset distance threshold value, the area is an area with poor indoor map constraint.
The Wi-Fi position fingerprint map automatic construction system comprises a training sample acquisition module, a training sample acquisition module and a matching module, wherein the training sample acquisition module specifically comprises:
a data acquisition unit for acquiring relevant crowdsourcing data;
a relative motion trajectory generation unit, configured to calculate and obtain a relative motion trajectory of the user by using a pedestrian dead reckoning method based on the crowdsourcing data;
a specific behavior recognizing unit for recognizing a specific behavior of a user including several behavior types by a machine learning method;
the behavior sequence generation unit is used for constructing a behavior sequence model formed by the type of the specific behaviors in the relative motion trail and the relative spatial relationship among the specific behaviors;
a point-line model generating unit for generating a point-line model of the indoor map, wherein the point is a position where the specific behavior occurs, and a line is an edge of a connection point;
the matching unit is used for matching the point line model with the behavior sequence model through a hidden Markov model so as to obtain indoor position coordinate information of a relative motion track;
the fingerprint map generating unit is used for constructing a Wi-Fi position fingerprint map of an area with better map constraint based on the indoor position coordinate information and crowdsourcing data; when the distance between the obstacles of the indoor map is smaller than or equal to the distance threshold value, or the distance between the wall of the indoor map and the obstacles is smaller than or equal to the distance threshold value, the area is a better indoor map constraint area;
and the sample acquisition unit is used for screening the Wi-Fi intensity information with the position label according to the indoor map and taking the Wi-Fi intensity information as a sample.
The Wi-Fi position fingerprint map automatic construction system, wherein the prediction module specifically comprises:
a functional relationship obtaining unit for recording sample points included in the samples as { p1,...,pi,...pnH, the Wi-Fi signal strength of the sample point is noted as f, an
Figure GDA0002200253550000061
Modeling Wi-Fi signal strength and position information of a sample point according to a Gaussian process to obtain a functional relation of modeling of the Wi-Fi signal strength and the position information; wherein, when the functional relationship is denoted as F, the distribution of F is:
Figure GDA0002200253550000062
wherein m (-) is a mean function of the Gaussian process, and k (-) is a covariance function of the Gaussian process;
a sample point signal strength acquisition unit for modeling the relationship between Wi-Fi signal strength and distance according to the signal propagation model to obtain the strength f (p) of the received signali)＝γ-10αlog10(||pi-pS||)+Ψ(pi,ps) Wherein gamma is constant, α is signal path attenuation coefficient, psi (p)i,ps) Is to describe the Wi-Fi access point position psAnd Wi-Fi access point location piA location-dependent parameter of signal attenuation therebetween;
a training unit for training the sample Sj＝{(p1,f(p1)),(p2,f(p2)),...,(pn,f(pn) ) } and two-dimensional coordinates comprising training samples
Figure GDA0002200253550000063
Obtaining Wi-Fi signal strength contained in training samples
Figure GDA0002200253550000064
And includes Wi-Fi signal strength in training samples
Figure GDA0002200253550000065
F in (1)jObey a multidimensional joint Gaussian distribution and is noted
Figure GDA0002200253550000066
Wherein the content of the first and second substances,
Figure GDA0002200253550000067
c is an n × n matrix, each element
Figure GDA0002200253550000071
When i is j, deltaij1, in other cases deltaij＝0；dcIs the distance of the correlation, and,
Figure GDA0002200253550000072
for the attenuation variance,. epsilon.for the measurement noise,. sigmaεIs Gaussian noise, pi、pjTwo-dimensional coordinates representing a training sample;
a mean and variance obtaining unit for obtaining the position p to be predicted*The signal strength of (A) is expressed as
Figure GDA0002200253550000073
And combining y and f into a multidimensional Gaussian variable [ f with (n +1) multiplied by 1 columns; y is]：
Figure GDA0002200253550000074
Derived from conditional distribution of Gaussian random variables
Figure GDA0002200253550000075
Wherein
Figure GDA0002200253550000076
Figure GDA0002200253550000077
Figure GDA0002200253550000078
The Wi-Fi intensity average value of the Wi-Fi position fingerprint map without position labels,
Figure GDA0002200253550000079
and the Wi-Fi intensity variance of the Wi-Fi position fingerprint map without position labels is obtained, k represents the covariance, and n represents the number of training samples.
The Wi-Fi location fingerprint map automatic construction system, wherein the crowdsourcing data comprises: crowd-sourced user acceleration data, gyroscope data, magnetometer data, barometer data, and Wi-Fi data.
The invention provides a Wi-Fi position fingerprint map automatic construction method and a system, wherein the method comprises the following steps: acquiring crowdsourcing data, constructing a Wi-Fi position fingerprint map by using a pedestrian dead reckoning method and a machine learning method, and screening Wi-Fi intensity information with position labels according to an indoor map; and taking the Wi-Fi intensity information with the position label as a training sample of the Gaussian process to obtain a functional relation between the signal intensity and the position information, solving the hyperparameter in the Gaussian process, and predicting to obtain the Wi-Fi intensity information of the area with poor indoor map constraint according to the hyperparameter. The method adopts Gaussian process regression, and predicts the Wi-Fi position fingerprint of an open area based on the Wi-Fi position fingerprint information of an area with better map constraint, so that the automatic construction of the Wi-Fi position fingerprint map of the whole indoor area is realized.
Drawings
FIG. 1 is a flowchart illustrating an automatic Wi-Fi location fingerprint map construction method according to a preferred embodiment of the present invention.
FIG. 2 is a schematic diagram of an indoor area in the Wi-Fi location fingerprint map automatic construction method of the present invention.
FIG. 3 is a functional block diagram of an automatic Wi-Fi location fingerprint map construction system according to a preferred embodiment of the present invention.
Detailed Description
The invention provides a method and a system for automatically constructing a Wi-Fi position fingerprint map, which are further described in detail below in order to make the purpose, the technical scheme and the effect of the invention clearer and clearer. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, which is a flowchart of a preferred embodiment of the method for automatically constructing a Wi-Fi location fingerprint map according to the present invention, the method includes the following steps:
s100, acquiring crowdsourcing data, constructing a Wi-Fi position fingerprint map by using a pedestrian dead reckoning method and a machine learning method, and screening Wi-Fi intensity information with position labels according to an indoor map;
step S200, using Wi-Fi intensity information with position labels as training samples of a Gaussian process to obtain a functional relation between signal intensity and position information, solving hyper-parameters in the Gaussian process, and predicting Wi-Fi intensity information of an indoor map area with poor constraint according to the hyper-parameters; when the distance between the indoor map obstacles is larger than a preset distance threshold value or the distance between the indoor map wall and the obstacles is larger than a preset distance threshold value, the area is an area with poor indoor map constraint.
By adopting the existing Wi-Fi position fingerprint map construction method, the Wi-Fi position fingerprint information with better map constraint area, namely the Wi-Fi signal intensity corresponding to each coordinate, can be obtained. The location fingerprint of the elongated corridor area (located around the sky well in fig. 2) in fig. 2 can be obtained, and for an open area (such as the rectangular area in fig. 2), the location fingerprint strength of the area cannot be obtained because the map constraint is poor and the whole area cannot be described by using a one-dimensional wire model. In step S200, the preset optimal value of the distance threshold is 2m, and the distance threshold can be adjusted in size and related to the accuracy of the indoor map.
In the embodiment of the invention, firstly, the Wi-Fi position fingerprint map obtained in the area with better map constraint is used as a training sample to obtain the hyperparameter of the Gaussian process regression model, and on the basis, the Wi-Fi position fingerprint of the open area is predicted.
Preferably, the step S100 specifically includes:
and step S101, acquiring related crowdsourcing data.
The crowdsourcing data refers to a plurality of different types of sensor data obtained from terminals of a plurality of crowdsourcing users. Crowdsourcing refers to a non-selective pointing mode, i.e., involving a variety of different types and ranges of users.
The user's terminal may be any suitable terminal device having at least one sensor, such as a smart phone, a smart watch, various wearable devices, and the like.
Specifically, the crowdsourcing data includes: crowd-sourced user acceleration data, gyroscope data, magnetometer data, barometer data, and Wi-Fi data. Wi-Fi data can include MAC address, SSID, strength RSSI, among others.
And S102, calculating and obtaining the relative motion track of the user by using a pedestrian dead reckoning method based on the crowdsourcing data.
Specifically, the pedestrian dead reckoning method specifically includes:
calculating the number of the steps of the user through a peak detection algorithm based on the acceleration data;
estimating the walking step length of the user through a step frequency step length model;
obtaining the advancing direction of the user according to the magnetometer data;
calculating to obtain a forward distance according to the number of the forward steps and the walking step length; and generating a relative motion track of the user according to the advancing distance and the advancing direction.
The step frequency step size model may be specifically defined by the following equation:
sl＝a·f+b
wherein sl is a step length, f is a step frequency, and a and b are constant parameters.
And multiplying the number of the advancing steps and the walking step length to calculate the advancing distance. Finally, a relative motion trajectory of the user is generated based on the forward distance and the forward direction (S24).
The relative movement locus may be represented by corresponding coordinates on the xy coordinate axis. Specifically, it can be calculated by the following formula:
Figure GDA0002200253550000101
wherein x istAnd ytRespectively the position of the user at time t, xt-1And yt-1Respectively, the position of the user at time t-1, deltad is the advance distance,
Figure GDA0002200253550000102
is the forward direction.
Step S103, identifying specific behaviors of the user comprising a plurality of behavior types through a machine learning method.
The specific behavior specifically refers to the behavior of a user (pedestrian) at a special indoor position (i.e. some behavior types other than normal walking), for example, the turning behavior of the pedestrian at a corner, the overweight and weightlessness behavior of the pedestrian when riding an elevator, and the like.
From crowdsourced data (e.g. accelerometer, gyroscope, magnetometer and barometer data) the different types of behaviour described above can be identified using suitable machine learning methods.
The machine learning method specifically comprises the following steps:
collecting sample time sequence data containing a plurality of specific behaviors;
segmenting the sample timing data through a sliding window of a predetermined length to obtain a specific behavior sample;
extracting features of the specific behavior sample;
training a classifier for classifying a specific behavior based on the features of the specific behavior sample;
the crowd-sourced data is segmented using the same sliding window and classified using a trained classifier, generating a type of a particular behavior of the crowd-sourced data.
That is, first, sample timing data is collected that contains several specific behaviors. Generally, sensor data (i.e., crowdsourcing data) of a mobile terminal (e.g., a smart phone) is time-series data.
Then, the sample timing data is divided by a sliding window of a predetermined length to obtain each specific behavior sample. The time window may be set to 2 seconds.
Then, features of the specific behavior sample are extracted. The characteristics used can be selected and determined according to actual conditions. For example, based on the data type of crowdsourcing data of a specific embodiment of the present invention, the mean and standard deviation of the three-axis acceleration, the mean and standard deviation of the three-axis angular velocity, and the variation value of the air pressure may be used as the characteristics of the sample.
The specific calculation method is represented by the following formula:
Figure GDA0002200253550000111
Figure GDA0002200253550000112
Figure GDA0002200253550000113
Figure GDA0002200253550000114
Figure GDA0002200253550000115
wherein
Figure GDA0002200253550000116
And σaThe mean and standard deviation of the three-axis acceleration,
Figure GDA0002200253550000117
and σgIs the mean value and standard deviation of the angular velocity of the three axes, Δ p is the variation value of the air pressure, n is the number of sensor data in the behavior sample, npCalculating the data number of the air pressure mean value for the user as a constant np≤n
Then entering a training stage: training a classifier for classifying a particular behavior based on features of the particular behavior sample. The specific classifier selected and used and the corresponding classifier parameters can be determined according to actual conditions.
Finally, the crowdsourced data is segmented using the same sliding window and classified using a trained classifier, generating a type of a particular behavior of the crowdsourced data.
And step S104, constructing a behavior sequence model formed by the type of the specific behaviors in the relative motion trail and the relative spatial relationship among the specific behaviors.
Modeling the relative motion trajectory obtained in the above steps and the recognition result of the specific behavior, so as to obtain a model including the relative spatial relationship between the specific behaviors and the specific behavior type, where the term "behavior sequence model" is used to represent the model.
And step S105, generating a point-line model of the indoor map, wherein the point is the position where the specific behavior occurs, and the line is the edge of the connecting point.
The dotted line model can be realized by a manual method or obtained by an automatic method. For example, a voronoi diagram method may be used to generate a point line model of the indoor map, and other suitable automatic generation methods may be used to obtain a point line model that meets the usage requirements.
And S106, matching the dotted line model with the behavior sequence model through a hidden Markov model so as to obtain indoor position coordinate information of the relative motion track.
Step S106 specifically includes:
first, a hidden markov model is used to match a point where a specific behavior occurrence location corresponds to the dotted line model.
The specific matching method is as follows:
1. hidden state: nodes in the graph structure.
2. And (3) observation value: the relative displacement between the occurrence moments of the specific behaviors obtained by the dead reckoning of the pedestrian.
3. Probability of state transition: when a particular behavior is identified, a transition probability between hidden states is generated. Through the topological structure of the indoor map, a transition probability matrix between the hidden states can be obtained. Since a pedestrian (user) can move only between adjacent points, the transition probability of each point to its adjacent point can be assumed to be uniformly distributed, whereby the state transition matrix of the entire indoor map can be obtained. For example, as shown in fig. 3, a schematic diagram of the state transition probability calculation according to the present invention is shown.
4. And (3) outputting the probability: the output probability describes the probability distribution of the observed value in each hidden state. In the hidden markov model employed in the present invention, the observed value is a relative displacement between specific behaviors derived from dead reckoning of a pedestrian.
According to the principle of pedestrian dead reckoning, a relative displacement error consists of a distance estimation error and an angle error. Thus, the observation probability distribution consists of two parts, a distance observation probability distribution and an angle observation probability distribution. Since the distance and angle observations are independent of each other, the observation probability distribution is:
Figure GDA0002200253550000131
wherein σdIs the standard deviation of the distance measurements and,
Figure GDA0002200253550000132
is the standard deviation of the angle measurement.
5. Initial probability distribution: the initial probability distribution is assumed to be a uniform distribution, and when the first specific behavior is identified, the probability of the position at that time at each corresponding node is considered to be equal.
6. The Viterbi algorithm: the viterbi algorithm is used to find a hidden state sequence with the maximum matching probability with the relative motion trajectory of the user, that is, the relative motion trajectory of the user is matched with a point in the point-line model through a specific behavior contained in the relative motion trajectory of the user and a relative displacement between the specific behaviors, so that each step in the relative motion displacement is positioned. Using a mathematical model to represent O ═ (O)1,O2,...,OT) For a particular behavior contained in a track, the Viterbi variable is defined by:
Figure GDA0002200253550000133
wherein, deltat(i) Is the probability that time t is in state i, aijIs the state transition probability of states i to j, bj(Ot+1) A probability is output for the observed value for state j. To obtain the most probable state, ρt+1(j) The definition is as follows:
Figure GDA0002200253550000134
according to a plurality of continuous specific behaviors in the track obtained by behavior recognition, the absolute coordinates (coordinate positions of the indoor map) of the track are obtained by matching with a plurality of points in the point-line model, and the points matched with the specific behaviors in the track are called node chains. The probability of each chain of selected nodes is calculated using the following equation:
pt+1(j)＝pt(i)·aij·bj(Ot+1),1≤t≤T
wherein p ist(i) Is the probability of the candidate node chain at time t. And when the ratio of the probability value of the maximum candidate node chain to the node chain probability value with the next highest probability value is greater than the threshold value C, the candidate node chain with the highest probability value is the matching result.
And then, calculating according to the number of steps between adjacent specific behaviors in the behavior sequence model and the distance between two corresponding points in the point-line model to obtain corresponding coordinate information of each step in the indoor map. That is, according to the number of steps between specific behaviors in the track, the track is interpolated, and according to the distance between nodes in the indoor road network, the absolute coordinate information of each step is obtained.
S107, constructing a Wi-Fi position fingerprint map of an area with better map constraint based on the indoor position coordinate information and crowdsourcing data; when the distance between the obstacles of the indoor map is smaller than or equal to the distance threshold value or the distance between the wall of the indoor map and the obstacles is smaller than or equal to the distance threshold value, the area is a better indoor map constraint area.
Step S107 specifically includes:
first, the indoor map is gridded. Namely, the indoor map is divided into grids with the same size by using equally spaced vertical and horizontal lines.
And then, according to the indoor position coordinate information of the relative motion trail, selecting Wi-Fi information which is closest to the center of the grid as first position fingerprint information of the grid.
Namely, for each relative motion track, according to the position information of each step of the user in the track, the Wi-Fi grid center coordinate closest to the Euclidean distance is selected as the coordinate information of the Wi-Fi fingerprint acquired at the step detection time.
And repeatedly executing the steps on a plurality of relative motion tracks in the crowdsourcing data to obtain a plurality of first position fingerprint information.
And finally, averaging a plurality of first position fingerprint information of the grids corresponding to the relative motion tracks to form second position fingerprint information of the grids. In particular, the location fingerprint may be an average of the signal strength of each AP (i.e., hotspot, identifiable from MAC address).
The crowd-sourced data contains multiple relative motion trajectories (as opposed to different users). Therefore, a final Wi-Fi position fingerprint map or fingerprint map database can be constructed by integrating the crowdsourcing data and the calculated absolute position (i.e., indoor position coordinate information).
And S108, screening the Wi-Fi strength information with the position label according to the indoor map, and taking the Wi-Fi strength information as a sample.
By screening (selecting out the fingerprints of the map with better constraint, such as corridor areas), the Wi-Fi strength information with the position label can be obtained and used as a sample. Assume that n samples are obtained, denoted as:
Figure GDA0002200253550000151
wherein (x)i，yi) Is the coordinate information of the ith sample,
Figure GDA0002200253550000152
the signal strength of the jth Wi-Fi access point in the ith sample is 1 < i < n, 1 < j < m, n is the number of the samples, and m is the number of the Wi-Fi access points contained in the samples.
Preferably, the step S200 specifically includes:
step S201, recording sample points included in the sample as { p1,...,pi,...pnH, the Wi-Fi signal strength of the sample point is noted as f, an
Figure GDA0002200253550000153
Modeling Wi-Fi signal strength and position information of a sample point according to a Gaussian process to obtain a functional relation of modeling of the Wi-Fi signal strength and the position information; wherein, when the functional relationship is denoted as F, the distribution of F is:
Figure GDA0002200253550000154
wherein m (-) is a mean function of the Gaussian process, and k (-) is a covariance function of the Gaussian process.
The entire indoor area may be divided into R grids, where n are samples obtained by behavioral map matching, and n < R. According to the method, the Wi-Fi signal intensity of the remaining grids in the whole indoor area is predicted through samples obtained through crowdsourcing tracks, and the Wi-Fi signal intensity are recorded as follows: the contents of S and A are shown in the specification,
Figure GDA0002200253550000155
suppose there are m Wi-Fi access points, location p, throughout an indoor areaiThe received signal strength of the jth Wi-Fi access point is expressed as a function fj(pi)，pi＝(xi,yi) (without loss of generality, the subscript j in the formula is then removed) the signal strength observation for this point is then:
y(pi)＝f(pi)+ε (1)
wherein ε is the measurement noise and is a normal Gaussian distribution
Figure GDA0002200253550000156
Using Gaussian process to pair f (p)i) And modeling is carried out, the Gaussian process is trained through a sample set S, the returning hyper-parameter of the Gaussian process is obtained through training, finally, the grid signal strength of an unknown area is predicted through the trained hyper-parameter, and therefore the Wi-Fi position fingerprint map of the whole indoor area is automatically constructed and updated.
Step S202, modeling is carried out on the relation between the Wi-Fi signal strength and the distance according to the signal propagation model, and the strength f (p) of the received signal is obtainedi)＝γ-10αlog10(||pi-pS||)+Ψ(pi,ps) Wherein gamma is constant, α is signal path attenuation coefficient, psi (p)i，ps) Is to describe the Wi-Fi access point position psAnd Wi-Fi access point location piA location-dependent parameter of signal attenuation therebetween.
S＝((p1,rss1),(p2,rss2),...,(pn,rssn) For the obtained position-labeled sample, the sample point is { p }1,...,pi,...pnThe signal intensities of these sample points are expressed as
Figure GDA0002200253550000161
And modeling the signal intensity and the position information of the sample point by using a Gaussian process to obtain a functional relation F of the signal intensity and the position information, wherein the mean function of the Gaussian process is m (-) and the covariance function is k (-,). Then
Figure GDA0002200253550000162
Having the following distribution:
Figure GDA0002200253550000163
the gaussian process is generally represented using the following form:
Figure GDA0002200253550000164
m(p)＝E[f(p)](4)
k(p,p')＝E[(f(p)-m(p))(f(p')-m(p'))](5)
the object of the invention is to predict the signal strength, f, at the coordinate points of the respective input positions by means of a Gaussian processj(p*) Is in position p*The signal strength of Wi-Fi access point j.
Will f isj(pi) Modeling as a gaussian process (equations (3) - (5)):
Figure GDA0002200253550000165
the relationship between Wi-Fi signal strength and distance is modeled herein using, without limitation, an open-space signal propagation model, and the received signal strength (in dB) can be described as:
f(pi)＝γ-10αlog10(||pi-pS||)+Ψ(pi,ps) (7)
where γ is a constant and α is the signal path attenuation coefficient, Ψ (p)i，ps) For location-related parameters, the Wi-Fi access point location p is describedsAnd position piThe signal attenuation in between. Let Ψ (p)i,ps) Obey a normal distribution, i.e.:
Figure GDA0002200253550000171
wherein the content of the first and second substances,
Figure GDA0002200253550000172
is the attenuation variance.
Position piAnd pjThe covariance of (a) is given by:
Figure GDA0002200253550000173
wherein d iscIs the correlation distance.
Step S203, according to the training sample Sj＝{(p1,f(p1)),(p2,f(p2)),...,(pn,f(pn) ) } and two-dimensional coordinates comprising training samples
Figure GDA0002200253550000174
Obtaining Wi-Fi signal strength contained in training samples
Figure GDA0002200253550000175
And includes Wi-Fi signal strength in training samples
Figure GDA0002200253550000176
F in (1)jObey a multidimensional joint Gaussian distribution and is noted
Figure GDA0002200253550000177
Wherein the content of the first and second substances,
Figure GDA0002200253550000178
c is an n × n matrix, each element
Figure GDA0002200253550000179
When i is j, deltaij1, in other cases deltaij＝0；dcIs the distance of the correlation, and,
Figure GDA00022002535500001710
for the attenuation variance,. epsilon.for the measurement noise,. sigmaεIs Gaussian noise, pi、pjRepresenting two-dimensional coordinates of the training sample.
Step S204, the position p to be predicted*The signal strength of (A) is expressed as
Figure GDA0002200253550000181
And combining y and f into a multidimensional Gaussian variable [ f with (n +1) multiplied by 1 columns; y is]：
Figure GDA0002200253550000182
By GaussConditional distribution of machine variables
Figure GDA0002200253550000183
Wherein
Figure GDA0002200253550000184
Figure GDA0002200253550000185
The Wi-Fi intensity average value of the Wi-Fi position fingerprint map without position labels,
Figure GDA0002200253550000186
and the Wi-Fi intensity variance of the Wi-Fi position fingerprint map without position labels is obtained, k represents the covariance, and n represents the number of training samples.
Sj＝{(p1,f(p1)),(p2,f(p2)),...,(pn,f(pn) ) } are training samples, where f (-) is given by equation (1) and n is the number of training samples.
Figure GDA0002200253550000187
Is an n × 2 matrix, and contains the two-dimensional coordinates of the training samples.
Figure GDA0002200253550000188
Is an n-dimensional column vector that contains the received signal strength in the training samples.
Since ε in formula (1) is a normal Gaussian distribution, fj(pi) Is the sum of two gaussian random variables. Thus, the noise observation vector fjObeying a multidimensional joint gaussian distribution:
Figure GDA0002200253550000189
wherein the content of the first and second substances,
Figure GDA00022002535500001810
c is an n × n matrix, each element
Figure GDA00022002535500001811
When i is j, δijOther cases, δij＝0。
By using
Figure GDA00022002535500001812
Indicating the position to be predicted is p*The signal strength of (c). y and f constitute a multidimensional gaussian variable [ f; y is]：
Figure GDA00022002535500001813
Wherein the content of the first and second substances,
Figure GDA0002200253550000191
the conditional distribution of gaussian random variables gives:
Figure GDA0002200253550000192
Figure GDA0002200253550000193
Figure GDA0002200253550000194
by equations (12) to (14), the mean (equation (13)) and the variance (equation (14)) of the signal intensities of the sample regions not collected can be predicted by using gaussian process regression.
Based on the embodiment of the method, the invention also provides a Wi-Fi position fingerprint map automatic construction system. As shown in fig. 3, the Wi-Fi location fingerprint map automatic construction system includes:
the training sample acquisition module 100 is used for acquiring crowdsourcing data, constructing a Wi-Fi position fingerprint map by using a pedestrian dead reckoning method and a machine learning method, and screening Wi-Fi intensity information with position labels according to an indoor map;
the prediction module 200 is used for taking the Wi-Fi intensity information with the position label as a training sample of a Gaussian process to obtain a functional relation between signal intensity and position information, solving a hyper-parameter in the Gaussian process, and predicting to obtain the Wi-Fi intensity information of an area with poor indoor map constraint according to the hyper-parameter; when the distance between the indoor map obstacles is larger than a preset distance threshold value or the distance between the indoor map wall and the obstacles is larger than a preset distance threshold value, the area is an area with poor indoor map constraint.
Preferably, in the system for automatically constructing a Wi-Fi location fingerprint map, the training sample acquisition module specifically includes:
a data acquisition unit for acquiring relevant crowdsourcing data;
a relative motion trajectory generation unit, configured to calculate and obtain a relative motion trajectory of the user by using a pedestrian dead reckoning method based on the crowdsourcing data;
a specific behavior recognizing unit for recognizing a specific behavior of a user including several behavior types by a machine learning method;
the behavior sequence generation unit is used for constructing a behavior sequence model formed by the type of the specific behaviors in the relative motion trail and the relative spatial relationship among the specific behaviors;
a point-line model generating unit for generating a point-line model of the indoor map, wherein the point is a position where the specific behavior occurs, and a line is an edge of a connection point;
the matching unit is used for matching the point line model with the behavior sequence model through a hidden Markov model so as to obtain indoor position coordinate information of a relative motion track;
the fingerprint map generating unit is used for constructing a Wi-Fi position fingerprint map of an area with better map constraint based on the indoor position coordinate information and crowdsourcing data; when the distance between the obstacles of the indoor map is smaller than or equal to the distance threshold value, or the distance between the wall of the indoor map and the obstacles is smaller than or equal to the distance threshold value, the area is a better indoor map constraint area;
and the sample acquisition unit is used for screening the Wi-Fi intensity information with the position label according to the indoor map and taking the Wi-Fi intensity information as a sample.
Preferably, in the automatic Wi-Fi location fingerprint map building system, the prediction module specifically includes:
a functional relationship obtaining unit for recording sample points included in the samples as { p1,...,pi,...pnH, the Wi-Fi signal strength of the sample point is noted as f, an
Figure GDA0002200253550000201
Modeling Wi-Fi signal strength and position information of a sample point according to a Gaussian process to obtain a functional relation of modeling of the Wi-Fi signal strength and the position information; wherein, when the functional relationship is denoted as F, the distribution of F is:
Figure GDA0002200253550000202
wherein m (-) is a mean function of the Gaussian process, and k (-) is a covariance function of the Gaussian process;
a sample point signal strength acquisition unit for modeling the relationship between Wi-Fi signal strength and distance according to the signal propagation model to obtain the strength f (p) of the received signali)＝γ-10αlog10(||pi-pS||)+Ψ(pi,ps) Wherein gamma is constant, α is signal path attenuation coefficient, psi (p)i,ps) Is to describe the Wi-Fi access point position psAnd Wi-Fi access point location piA location-dependent parameter of signal attenuation therebetween;
a training unit for training the sample Sj＝{(p1,f(p1)),(p2,f(p2)),...,(pn,f(pn) ) } and two-dimensional coordinates comprising training samples
Figure GDA0002200253550000211
Obtaining Wi-Fi signal strength contained in training samples
Figure GDA0002200253550000212
And includes Wi-Fi signal strength in training samples
Figure GDA0002200253550000213
F in (1)jObey a multidimensional joint Gaussian distribution and is noted
Figure GDA0002200253550000214
Wherein the content of the first and second substances,
Figure GDA0002200253550000215
c is an n × n matrix, each element
Figure GDA0002200253550000216
When i is j, deltaij1, in other cases deltaij＝0；dcIs the distance of the correlation, and,
Figure GDA0002200253550000217
for the attenuation variance,. epsilon.for the measurement noise,. sigmaεIs Gaussian noise, pi、pjTwo-dimensional coordinates representing a training sample;
a mean and variance obtaining unit for representing the signal intensity at the position p needing prediction as
Figure GDA0002200253550000218
And combining y and f into a multidimensional Gaussian variable [ f with (n +1) multiplied by 1 columns; y is]：
Figure GDA0002200253550000219
Derived from conditional distribution of Gaussian random variables
Figure GDA00022002535500002110
Wherein
Figure GDA00022002535500002111
Figure GDA00022002535500002112
Figure GDA00022002535500002113
The Wi-Fi intensity average value of the Wi-Fi position fingerprint map without position labels,
Figure GDA00022002535500002114
and the Wi-Fi intensity variance of the Wi-Fi position fingerprint map without position labels is obtained, k represents the covariance, and n represents the number of training samples.
Preferably, in the Wi-Fi location fingerprint map automatic construction system, the crowdsourcing data includes: crowd-sourced user acceleration data, gyroscope data, magnetometer data, barometer data, and Wi-Fi data.
In summary, the method and system for automatically constructing the Wi-Fi location fingerprint map provided by the present invention includes: acquiring crowdsourcing data, constructing a Wi-Fi position fingerprint map by using a pedestrian dead reckoning method and a machine learning method, and screening Wi-Fi intensity information with position labels according to an indoor map; and taking the Wi-Fi intensity information with the position label as a training sample of the Gaussian process to obtain a functional relation between the signal intensity and the position information, solving the hyperparameter in the Gaussian process, and predicting to obtain the Wi-Fi intensity information of the area with poor indoor map constraint according to the hyperparameter. The method adopts Gaussian process regression, and predicts the Wi-Fi position fingerprint of an open area based on the Wi-Fi position fingerprint information of an area with better map constraint, so that the automatic construction of the Wi-Fi position fingerprint map of the whole indoor area is realized.
It is to be understood that the invention is not limited to the examples described above, but that modifications and variations may be effected thereto by those of ordinary skill in the art in light of the foregoing description, and that all such modifications and variations are intended to be within the scope of the invention as defined by the appended claims.

Claims (8)

1. A Wi-Fi position fingerprint map automatic construction method is characterized by comprising the following steps:
A. acquiring crowdsourcing data, constructing a Wi-Fi position fingerprint map by using a pedestrian dead reckoning method and a machine learning method, and screening Wi-Fi intensity information with position labels according to an indoor map;
B. the Wi-Fi intensity information with the position labels is used as a training sample of a Gaussian process to obtain a functional relation between signal intensity and position information, a hyper-parameter in the Gaussian process is solved, and Wi-Fi intensity information of an indoor map area with poor constraint is obtained according to the hyper-parameter prediction; when the distance between the obstacles of the indoor map is greater than a preset distance threshold value or the distance between the wall of the indoor map and the obstacles is greater than a preset distance threshold value, the area is an area with poor indoor map constraint;
the step B specifically comprises the following steps:
b1, marking the sample points included in the sample as p1,...,pi,...pnH, the Wi-Fi signal strength of the sample point is noted as f, an
Figure FDA0002200253540000011
Modeling Wi-Fi signal strength and position information of a sample point according to a Gaussian process to obtain a functional relation of modeling of the Wi-Fi signal strength and the position information; wherein, when the functional relationship is denoted as F, the distribution of F is:
Figure FDA0002200253540000012
wherein m (-) is a mean function of the Gaussian process, and k (-) is a covariance function of the Gaussian process;
b2, modeling the relationship between Wi-Fi signal strength and distance according to the signal propagation model to obtain the strength f (p) of the received signali)＝γ-10αlog10(||pi-pS||)+Ψ(pi,ps) Wherein gamma is constant, α is signal path attenuation coefficient, psi (p)i,ps) Is to describe the Wi-Fi access point position psAnd Wi-Fi access point location piA location-dependent parameter of signal attenuation therebetween;
b3, according to the training sample Sj＝{(p1,f(p1)),(p2,f(p2)),...,(pn,f(pn) ) } and two-dimensional coordinates comprising training samples
Figure FDA0002200253540000013
Obtaining Wi-Fi signal strength contained in training samples
Figure FDA0002200253540000014
And includes Wi-Fi signal strength in training samples
Figure FDA0002200253540000021
F in (1)jObey a multidimensional joint Gaussian distribution and is noted
Figure FDA0002200253540000022
Wherein the content of the first and second substances,
Figure FDA0002200253540000023
c is an n × n matrix, each element
Figure FDA0002200253540000024
When i is j, deltaij1, in other cases deltaij＝0；dcIs the distance of the correlation, and,
Figure FDA0002200253540000025
for the attenuation variance,. epsilon.for the measurement noise,. sigmaεIs Gaussian noise, pi、pjTwo-dimensional coordinates representing a training sample;
b4, position p to be predicted*The signal strength of (A) is expressed as
Figure FDA0002200253540000026
And combining y and f into a multidimensional Gaussian variable [ f with (n +1) multiplied by 1 columns; y is]：
Figure FDA0002200253540000027
Derived from conditional distribution of Gaussian random variables
Figure FDA0002200253540000028
Wherein
Figure FDA0002200253540000029
1≤i≤n，
Figure FDA00022002535400000210
The Wi-Fi intensity average value of the Wi-Fi position fingerprint map without position labels,
Figure FDA00022002535400000211
and the Wi-Fi intensity variance of the Wi-Fi position fingerprint map without position labels is obtained, k represents the covariance, and n represents the number of training samples.
2. The method for automatically constructing the Wi-Fi location fingerprint map according to claim 1, wherein the step a specifically comprises:
a1, acquiring related crowdsourcing data;
a2, calculating and obtaining a relative motion track of the user by using a pedestrian dead reckoning method based on the crowdsourcing data;
a3, identifying the specific behaviors of a user comprising a plurality of behavior types through a machine learning method;
a4, constructing a behavior sequence model formed by the type of the specific behaviors in the relative motion trail and the relative spatial relationship among the specific behaviors;
a5, generating a dotted line model of the indoor map, wherein the point is the position where the specific action occurs, and the line is the edge of the connection point;
a6, matching the dotted line model with the behavior sequence model through a hidden Markov model to obtain indoor position coordinate information of a relative motion track;
a7, constructing a Wi-Fi position fingerprint map of an area with better map constraint based on the indoor position coordinate information and crowdsourcing data; when the distance between the obstacles of the indoor map is smaller than or equal to the distance threshold value, or the distance between the wall of the indoor map and the obstacles is smaller than or equal to the distance threshold value, the area is a better indoor map constraint area;
and A8, screening Wi-Fi strength information with position labels according to the indoor map, and taking the Wi-Fi strength information as a sample.
3. The Wi-Fi location fingerprint map automatic construction method of claim 1 or 2, wherein the crowdsourcing data comprises: crowd-sourced user acceleration data, gyroscope data, magnetometer data, barometer data, and Wi-Fi data.
4. The Wi-Fi location fingerprint map automatic construction method according to claim 1 or 2, wherein the pedestrian dead reckoning method specifically comprises:
calculating the number of the steps of the user through a peak detection algorithm based on the acceleration data;
estimating the walking step length of the user through a step frequency step length model;
obtaining the advancing direction of the user according to the magnetometer data;
calculating to obtain a forward distance according to the number of the forward steps and the walking step length; and generating a relative motion track of the user according to the advancing distance and the advancing direction.
5. The Wi-Fi location fingerprint map automatic construction method according to claim 1 or 2, wherein the machine learning method specifically comprises:
collecting sample time sequence data containing a plurality of specific behaviors;
segmenting the sample timing data through a sliding window of a predetermined length to obtain a specific behavior sample;
extracting features of the specific behavior sample;
training a classifier for classifying a specific behavior based on the features of the specific behavior sample;
the crowd-sourced data is segmented using the same sliding window and classified using a trained classifier, generating a type of a particular behavior of the crowd-sourced data.
6. A Wi-Fi position fingerprint map automatic construction system is characterized by comprising:
the training sample acquisition module is used for acquiring crowdsourcing data, constructing a Wi-Fi position fingerprint map by using a pedestrian dead reckoning method and a machine learning method, and screening Wi-Fi intensity information with position labels according to an indoor map;
the prediction module is used for taking the Wi-Fi intensity information with the position label as a training sample of a Gaussian process to obtain a functional relation between signal intensity and position information, solving a hyper-parameter in the Gaussian process, and predicting to obtain the Wi-Fi intensity information of an area with poor indoor map constraint according to the hyper-parameter; when the distance between the obstacles of the indoor map is greater than a preset distance threshold value or the distance between the wall of the indoor map and the obstacles is greater than a preset distance threshold value, the area is an area with poor indoor map constraint;
the prediction module specifically comprises:
a functional relationship obtaining unit for recording sample points included in the samples as { p1,...,pi,...pnH, the Wi-Fi signal strength of the sample point is noted as f, an
Figure FDA0002200253540000041
Modeling Wi-Fi signal strength and position information of a sample point according to a Gaussian process to obtain a functional relation of modeling of the Wi-Fi signal strength and the position information; wherein, when the functional relationship is denoted as F, the distribution of F is:
Figure FDA0002200253540000042
wherein m (-) is a mean function of the Gaussian process, and k (-) is a covariance function of the Gaussian process;
a sample point signal strength acquisition unit for modeling the relationship between Wi-Fi signal strength and distance according to the signal propagation model to obtain the strength of the received signalf(pi)＝γ-10αlog10(||pi-pS||)+Ψ(pi,ps) Wherein gamma is constant, α is signal path attenuation coefficient, psi (p)i,ps) Is to describe the Wi-Fi access point position psAnd Wi-Fi access point location piA location-dependent parameter of signal attenuation therebetween;
a training unit for training the sample Sj＝{(p1,f(p1)),(p2,f(p2)),...,(pn,f(pn) ) } and two-dimensional coordinates comprising training samples
Figure FDA0002200253540000043
Obtaining Wi-Fi signal strength contained in training samples
Figure FDA0002200253540000044
And includes Wi-Fi signal strength in training samples
Figure FDA0002200253540000045
F in (1)jObey a multidimensional joint Gaussian distribution and is noted
Figure FDA0002200253540000046
Wherein the content of the first and second substances,
Figure FDA0002200253540000047
c is an n × n matrix, each element
Figure FDA0002200253540000051
When i is j, deltaij1, in other cases deltaij＝0；dcIs the distance of the correlation, and,
Figure FDA0002200253540000052
for the attenuation variance,. epsilon.for the measurement noise,. sigmaεIs Gaussian noise, pi、pjTwo-dimensional coordinates representing a training sample;
mean and variance gainA fetch unit for fetching a position p to be predicted*The signal strength of (A) is expressed as
Figure FDA0002200253540000053
And combining y and f into a multidimensional Gaussian variable [ f with (n +1) multiplied by 1 columns; y is]：
Figure FDA0002200253540000054
Derived from conditional distribution of Gaussian random variables
Figure FDA0002200253540000055
Wherein
Figure FDA0002200253540000056
1≤i≤n，
Figure FDA0002200253540000057
The Wi-Fi intensity average value of the Wi-Fi position fingerprint map without position labels,
Figure FDA0002200253540000058
and the Wi-Fi intensity variance of the Wi-Fi position fingerprint map without position labels is obtained, k represents the covariance, and n represents the number of training samples.
7. The Wi-Fi location fingerprint map automatic construction system of claim 6, wherein the training sample acquisition module specifically comprises:
a data acquisition unit for acquiring relevant crowdsourcing data;
a relative motion trajectory generation unit, configured to calculate and obtain a relative motion trajectory of the user by using a pedestrian dead reckoning method based on the crowdsourcing data;
a specific behavior recognizing unit for recognizing a specific behavior of a user including several behavior types by a machine learning method;
the behavior sequence generation unit is used for constructing a behavior sequence model formed by the type of the specific behaviors in the relative motion trail and the relative spatial relationship among the specific behaviors;
a point-line model generating unit for generating a point-line model of the indoor map, wherein the point is a position where the specific behavior occurs, and a line is an edge of a connection point;
the matching unit is used for matching the point line model with the behavior sequence model through a hidden Markov model so as to obtain indoor position coordinate information of a relative motion track;
the fingerprint map generating unit is used for constructing a Wi-Fi position fingerprint map of an area with better map constraint based on the indoor position coordinate information and crowdsourcing data; when the distance between the obstacles of the indoor map is smaller than or equal to the distance threshold value, or the distance between the wall of the indoor map and the obstacles is smaller than or equal to the distance threshold value, the area is a better indoor map constraint area;
and the sample acquisition unit is used for screening the Wi-Fi intensity information with the position label according to the indoor map and taking the Wi-Fi intensity information as a sample.
8. The Wi-Fi location fingerprint map automatic construction system of claim 6 or 7, wherein the crowdsourcing data comprises: crowd-sourced user acceleration data, gyroscope data, magnetometer data, barometer data, and Wi-Fi data.
CN201710039771.5A 2017-01-19 2017-01-19 Wi-Fi position fingerprint map automatic construction method and system Active CN106714110B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710039771.5A CN106714110B (en) 2017-01-19 2017-01-19 Wi-Fi position fingerprint map automatic construction method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710039771.5A CN106714110B (en) 2017-01-19 2017-01-19 Wi-Fi position fingerprint map automatic construction method and system

Publications (2)

Publication Number Publication Date
CN106714110A CN106714110A (en) 2017-05-24
CN106714110B true CN106714110B (en) 2020-03-31

Family

ID=58908533

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710039771.5A Active CN106714110B (en) 2017-01-19 2017-01-19 Wi-Fi position fingerprint map automatic construction method and system

Country Status (1)

Country Link
CN (1) CN106714110B (en)

Families Citing this family (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107607122B (en) * 2017-10-10 2019-09-10 内蒙古大学 The building of location fingerprint library and dynamic updating method towards indoor positioning
CN109769296A (en) * 2017-11-09 2019-05-17 深圳市交投科技有限公司 A kind of indoor orientation method merging multi-source data
CN108200562B (en) * 2018-01-04 2020-12-22 Oppo广东移动通信有限公司 Network rate adjustment method and related product
CN108225332B (en) * 2018-01-10 2020-10-02 内蒙古大学 Indoor positioning fingerprint map dimension reduction method based on supervision
CN108462992B (en) * 2018-03-05 2021-03-19 中山大学 Indoor positioning method based on super-resolution reconstruction Wi-Fi fingerprint map
CN108449709B (en) * 2018-03-16 2019-06-18 华中科技大学 A kind of indoor radio map constructing method based on crowdsourcing sample
CN108571970A (en) * 2018-04-26 2018-09-25 杭州晶智能科技有限公司 A kind of fingerprint map constructing method of indoor mobile robot
CN108810817B (en) * 2018-05-18 2020-10-27 东南大学 Indoor map generation and track encryption method based on mobile crowdsourcing mode
CN211668521U (en) * 2018-05-31 2020-10-13 上海快仓智能科技有限公司 Automatic guide vehicle for image acquisition and processing system
CN109739830B (en) * 2019-02-28 2021-01-26 电子科技大学 Position fingerprint database rapid construction method based on crowdsourcing data
US11269851B2 (en) 2019-05-03 2022-03-08 International Business Machines Corporation Indoor survey data collection
CN110248309A (en) * 2019-05-14 2019-09-17 上海交通大学 Wireless signal fingerprint restoration methods based on dynamic sliding window
CN110264154B (en) * 2019-05-28 2023-06-09 南京航空航天大学 Crowd-sourced signal map construction method based on self-encoder
CN110300370B (en) * 2019-07-02 2021-04-02 广州纳斯威尔信息技术有限公司 Indoor positioning method for reconstructing wifi fingerprint map
CN110708674A (en) * 2019-11-08 2020-01-17 北京云迹科技有限公司 Multi-floor positioning method and system
CN111491262A (en) * 2020-04-26 2020-08-04 中国信息通信研究院 Method and device for determining signal strength of mobile broadband network
CN112070808A (en) * 2020-09-01 2020-12-11 三一专用汽车有限责任公司 Trajectory prediction method, apparatus and computer-readable storage medium
CN112399555B (en) * 2020-10-20 2023-07-14 北京嘀嘀无限科技发展有限公司 Position locating method and device, readable storage medium and electronic equipment
CN113810850B (en) * 2021-08-11 2022-12-13 荣耀终端有限公司 Method for creating positioning database and electronic equipment
CN113727273B (en) * 2021-08-19 2022-04-26 武汉大学 Personnel indoor semantic track reconstruction method based on wireless crowdsourcing data
CN116184312B (en) * 2022-12-22 2023-11-21 泰州雷德波达定位导航科技有限公司 Indoor crowdsourcing fingerprint library construction method based on semantic Wi-Fi

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102638889A (en) * 2012-03-21 2012-08-15 浙江大学 Indoor wireless terminal positioning method based on Bayes compression sensing
CN105263113A (en) * 2015-10-26 2016-01-20 深圳大学 Wi-Fi location fingerprint map building method and system based on crowd-sourcing

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102638889A (en) * 2012-03-21 2012-08-15 浙江大学 Indoor wireless terminal positioning method based on Bayes compression sensing
CN105263113A (en) * 2015-10-26 2016-01-20 深圳大学 Wi-Fi location fingerprint map building method and system based on crowd-sourcing

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
INDOOR NAVIGATION WITH FOOT-MOUNTED STRAPDOWN INERTIAL NAVIGATION AND MAGNETIC SENSORS;JEFF BIRD et al.;《IEEE Wireless Communications》;20110430;全文 *
基于智能手机的个体室内时空行为分析;李清泉等;《地理科学进展》;20150430;全文 *
用户行为感知辅助的室内行人定位;周宝定等;《武汉大学报.信息科学版》;20140630;全文 *

Also Published As

Publication number Publication date
CN106714110A (en) 2017-05-24

Similar Documents

Publication Publication Date Title
CN106714110B (en) Wi-Fi position fingerprint map automatic construction method and system
CN107396322B (en) Indoor positioning method based on path matching and coding-decoding cyclic neural network
CN105263113B (en) A kind of WiFi location fingerprints map constructing method and its system based on crowdsourcing
CN109525935B (en) Intelligent floor sensing method and control system for universal indoor environment
Liu et al. Fusion of magnetic and visual sensors for indoor localization: Infrastructure-free and more effective
CN107635204B (en) Indoor fusion positioning method and device assisted by exercise behaviors and storage medium
CN107907124B (en) Positioning method based on scene recognition, electronic equipment, storage medium and system
Fox et al. Bayesian techniques for location estimation
CN110401978B (en) Indoor positioning method based on neural network and particle filter multi-source fusion
CN110501010A (en) Determine position of the mobile device in geographic area
CN110388926B (en) Indoor positioning method based on mobile phone geomagnetism and scene image
Ye et al. Local HMM for indoor positioning based on fingerprinting and displacement ranging
Yang et al. Deepwipos: a deep learning-based wireless positioning framework to address fingerprint instability
Xingli et al. Indoor positioning technology based on deep neural networks
Antsfeld et al. Deep smartphone sensors-WiFi fusion for indoor positioning and tracking
CN109121081B (en) Indoor positioning method based on position candidate set and EM algorithm
Gong et al. An enhanced indoor positioning solution using dynamic radio fingerprinting spatial context recognition
CN111623797B (en) Step number measuring method based on deep learning
CA2894863A1 (en) Indoor localization using crowdsourced data
Wietrzykowski et al. Adopting the FAB-MAP algorithm for indoor localization with WiFi fingerprints
Yoo et al. Mapless indoor localization by trajectory learning from a crowd
Jiang et al. Real‐Time and Accurate Indoor Localization with Fusion Model of Wi‐Fi Fingerprint and Motion Particle Filter
CN114339595A (en) Ultra-wideband dynamic inversion positioning method based on multi-model prediction
CN110572772B (en) GRNN-AdaBoost-based multi-device fusion positioning method
Dhanjal et al. DeepLocNet: Deep Observation Classification and Ranging Bias Regression for Radio Positioning Systems

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant