CN111405474A - Indoor fingerprint map self-adaptive updating method based on communication investigation - Google Patents

Abstract

The invention requests to protect an indoor fingerprint map self-adaptive updating method based on communication investigation. Firstly, collecting static and dynamic wireless fingerprint data by using a mobile terminal, and then estimating the position of a reference point for the collected wireless fingerprint data based on a path matching mode; secondly, modeling the relation between the reference point and the RSS fingerprints of other non-reference points; and finally triggering and updating the fingerprint map according to the modeling model. The invention realizes automatic updating on the basis of the wireless fingerprint map, thereby realizing the self-adaptive updating of the fingerprint map to the dynamic environment change without any additional hardware deployment or any explicit user participation, and avoiding the loss caused by the fingerprint map offset in the real environment.

Description

An adaptive update method of indoor fingerprint map based on communication survey

technical field

The invention belongs to the field of indoor fingerprint map of communication survey, in particular to an adaptive update method of indoor fingerprint map based on communication survey.

Background technique

In recent years, with various demands and the maturity of outdoor positioning technology, indoor positioning technology is in full swing at this stage. Common indoor positioning technologies include UWB (Ultra Wideband) indoor positioning technology, RFID (Radio Frequency Identification) positioning technology , ZigBee indoor positioning technology, ultrasonic positioning, Wi-Fi positioning, etc. Among them, Wi-Fi positioning has the advantages of easy expansion, automatic data update, and low cost, so it is the first to achieve scale. However, with the development of Wi-Fi positioning technology to this day, many problems faced by Wi-Fi positioning technology have been deeply and comprehensively studied and discussed, including problems such as high cost of site survey and low positioning accuracy. These works have made the positioning of this technology mature. However, before it is actually used on a large scale, there is still a key problem that has not been solved, that is, the problem of adaptive update of fingerprint maps.

The indoor environment is not static, and environmental changes will have a strong impact on wireless signal propagation. The dynamics of the environment include not only short-term disturbances such as door openings and user movements, but also long-term changes such as changes in light, temperature, humidity, and other weather conditions. RSS is very sensitive to environmental changes, and may also have significant amplitude fluctuations under small environmental changes. The dense multipath propagation of wireless signals in a complex indoor environment exacerbates this degree of fluctuation. Therefore, the RSS fingerprints measured in real-time during the localization run phase may deviate from the initial fingerprint map constructed during the training phase. In other words, in the long-term operation process, the initially constructed static fingerprint map will gradually deviate from the real-time fingerprint or even eventually fail because it cannot adapt to dynamic environmental changes. As a result, the positioning accuracy of the positioning system will be greatly reduced over time. Therefore, in view of the above problems, this paper considers the use of mobile devices to automatically and continuously update the wireless signal fingerprint map, without relying on any additional hardware deployment or deliberate human participation, and designs a communication survey-based indoor fingerprint map adaptive update method. method.

SUMMARY OF THE INVENTION

The present invention aims to solve the problems in the prior art. An adaptive updating method of indoor fingerprint map based on communication survey is proposed. The technical scheme of the present invention is as follows:

An adaptive updating method for indoor fingerprint map based on communication survey, which comprises the following steps:

Step 1: use the mobile terminal to collect static and dynamic wireless fingerprint data; wherein the wireless fingerprint data collection is the real-time data automatically collected by the user's mobile device in the process of their daily work and life;

Step 2: Estimate the position of the reference point based on the path matching method for the collected wireless fingerprint data;

Step 3: Model the relationship between the reference point and the RSS fingerprints of other non-reference points;

Step 4: According to the modeling model, adaptively update the fingerprint map.

Further, the static wireless signal fingerprint of the step 1 is collected and recorded when the mobile device remains in a stationary state for a period of time at a certain position; the dynamic wireless signal fingerprint is collected and recorded simultaneously when the user moves. data to monitor the user's movement path.

Further, in the second step, the collected wireless fingerprint data is estimated based on the path matching method, and the specific steps are as follows:

(1) Estimate the feasible area according to the RSS fingerprint, use the RSS fingerprint measurement value on the moving path as the initial value of the overall path, search for candidate positions in a part of its subspace, and match the feasible area. The rough location estimate corresponding to the fingerprint always falls within a limited area (which is of course no larger than the entire location space). Therefore, we can outline a feasible region covering all fingerprint positions on the moving path, and only find the best match for the entire path within this feasible region;

的两侧考虑对称的方向区间

(2) Lock the feasible azimuth, and set the possible maximum directional error to be Δφ, then only the center direction

Consider the symmetric direction interval on both sides of

(3) Joint position estimation, the translation increment Δα and rotation increment Δβ are used to embed the moving path on the fingerprint map, and the matching algorithm finds the path J={s ₁ , s ₂ which can minimize the geometric constraints of the path under the premise of considering the path. ,s ₃ ,...s _w } The sequence position of the mean square error of all fingerprints is taken as the target position, and s _w represents the wth fingerprint measurement value in the path;

where d _j ′=||l _c(j+1) -l _c(j) || represents the distance between two candidate positions, f _c(j) represents the fingerprint data of the candidate positions, and d _j represents the phase in the path. The distance between adjacent positions, and c _j is the candidate position corresponding to s _j , Δd is a minimum distance constraint value, which can be set according to the specific environment and needs, after obtaining the candidate position corresponding to the entire path, the first position l _c(1) is selected as the reference position, and all fingerprint data at time t _k are aggregated together to obtain a set of reference points R _k ={l _r1 ,l _r2 ,...,l _rm }, where l _rm represents the first RSS values on m reference points, each reference point corresponds to a position estimate l _ri =(x _i , y _i ), i=1,2,...,m, x _i , y _j represent the horizontal direction of the position respectively Y-axis.

Further, in the step 3, the relationship between the reference point and the RSS fingerprints of other non-reference points is modeled, and the specific steps are as follows:

(1) For a series of reference points R _k obtained at time t _k , where the position of the jth reference point is l _rj , the relationship between the position contained in R _k and the RSS at other positions in the fingerprint map should be learned. Prediction model θ, the relationship model that needs to be learned for the jth AP (1≤j≤p) at position l _i , 1≤i≤n RSS is as follows θ _ij :

f _ij (t ₀ )=θ _ij (f _r1j (t ₀ ),f _r2j (t ₀ ),...,f _rmj (t ₀ ),) (2)

Here f _ij (t ₀ ) and f _rmj (t ₀ ) represent the RSS values of the jth AP at position l _i and position l _rm in the initial fingerprint map, respectively;

(2) Use partial least squares regression to establish a functional regression model PLSR, which specifically includes: first, decompose X and Y at the same time by searching a set of latent vectors to maximize the correlation coefficient between X and Y; Decomposition is used to predict Y, PLSR still has the form of multiple regression Y=XB+E, where B=X ^T U(T ^T XX ^T U) ^-1 T ^T Y, where T and U are the latent variable matrix and E is the residual difference matrix;

也就是来自m个参考点的RSS观测值，

是位置l_i上的RSS测量值，由于Y是一个一维向量，这里记为y,针对于单因变量的PLSR问题，采用PLS1方法求解。In the fingerprint map,

That is, the RSS observations from m reference points,

is the RSS measurement value at position _li . Since Y is a one-dimensional vector, it is denoted as y here. For the PLSR problem of a single variable, the PLS1 method is used to solve it.

的t_j＝X_jw_j：Further, the PLSR problem for a single variable is solved by using the PLS1 method, which specifically includes: for the jth latent variable, find the maximum covariance cov(X _j w _j , y _j ) according to the following rules and satisfy the conditions

t _j =X _j w _j :

t _j =X _j w _j (4)

When finding the first latent variable, let X ₁ =X and y ₁ =y. When finding the next latent variable t _j+1 , remove their respective regression estimates based on t _j from X _j and y _j , and then use the degraded residual values to repeat the above steps to solve the latent variable;

构成的列向量

由此可以得到PLSR的预测模型：Thus, after repeating the h rounds, two m×h matrices W and P, and an n×h matrix T can be obtained. The three matrices take w _j , p _j and t _j as column vectors respectively, and at the same time, One more by h

composed of column vectors

From this, the prediction model of PLSR can be obtained:

是预测值，

in

is the predicted value,

Further, in the step 4, according to the modeling model, the fingerprint map is adaptively updated, which specifically includes: when a sufficient number of reference point positions and their real-time measurement data are obtained, the update process can be triggered to update the current fingerprint map to the latest. Since each time the fingerprint map update is started, the number of reference points and their corresponding positions are different, so before each update, the targeted regression function must be re-learned from the initial fingerprint map.

The advantages and beneficial effects of the present invention are as follows:

1. The present invention uses the mobile device used by ordinary users daily as a movable reference point to collect real-time reference data, so that no additional hardware deployment or any explicit user participation is required, and the fingerprint map can be used to change the dynamic environment. Adaptive update avoids the loss caused by the offset of the fingerprint map in the real environment, and overcomes the change of indoor environment (including short-term interference such as door opening, user movement, etc., as well as light, temperature, humidity and other weather conditions). The long-term change caused by the change of conditions) brings about the problem of RSS fluctuation, which causes the static fingerprint map to be initially constructed to gradually deviate from the real-time fingerprint or even fail at last.

2. The present invention proposes a path matching method for the reference data in step 2, thereby estimating the positions of these moving reference points, making the wireless fingerprint data measured by them meaningful, and facilitating the accurate update of the subsequent fingerprints.

3. The present invention uses the partial least squares method in step 3 to obtain the correlation model between the RSS fingerprints of adjacent positions in the indoor environment, which is helpful to obtain a stable, correct and highly predictable model, which is convenient for positioning The accurate estimation of the position overcomes the problem that the time-invariant relationship model between the fingerprint of the reference point and the fingerprints at other positions is difficult to accurately model under the complex and changeable indoor signal propagation environment.

Description of drawings

FIG. 1 is a schematic flowchart of a preferred embodiment provided by the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be described clearly and in detail below with reference to the accompanying drawings in the embodiments of the present invention. The described embodiments are only some of the embodiments of the invention.

The technical scheme that the present invention solves the above-mentioned technical problems is:

In this embodiment, an adaptive updating method for indoor fingerprint map based on communication survey is performed as follows.

Step 1: Collect static and dynamic wireless fingerprint data using mobile terminals

The static and dynamic wireless fingerprint data is collected by using the mobile terminal, wherein the wireless fingerprint data collection is the automatic collection of real-time data through the user's mobile device in the process of their daily work and life. In particular, wireless signal fingerprints are collected and recorded when a mobile device remains stationary at a certain location for a period of time. When the user moves, wireless fingerprint data and movement data are simultaneously collected to monitor the user's movement path.

Step 2: For the collected wireless fingerprint data, based on the path matching method, estimate the position of the reference point

For the wireless fingerprint data collected by the mobile terminal, and then estimate the reference position according to the path matching method. The specific steps are as follows:

(1) Estimate the feasible area according to the RSS fingerprint, use the RSS fingerprint measurement value on the moving path as the initial value of the overall path, search for candidate positions in a part of its subspace, and match the feasible area. The rough location estimate corresponding to the fingerprint always falls within a limited area (which is of course no larger than the entire location space). Therefore, we can outline a feasible region covering all fingerprint positions on the moving path, and find the best match for the entire path only within this feasible region.

的两侧考虑对称的方向区间

(2) Lock the feasible azimuth, and set the possible maximum directional error to be Δφ, then only the center direction

Consider the symmetric direction interval on both sides of

(3) Joint position estimation, using translation increment Δα and rotation increment Δβ to embed the moving path on the fingerprint map, where Δα and Δβ can be set to 0.5 m and 2° respectively according to empirical values and environmental settings). Under the premise of considering the geometric constraints of the path, the matching algorithm finds the sequence position that can minimize the mean square error of all fingerprints on the path J={s ₁ , s ₂ , s ₃ ,...s _w } as the target position.

where d _j ′=||l _c(j+1) -l _c(j) || represents the distance between two candidate positions, f _c(j) represents the fingerprint data of the candidate positions, and d _j represents the phase in the path. The distance between adjacent positions, and c _j is the candidate position corresponding to s _j , Δd is a minimum distance constraint value, which can be set according to the specific environment and needs, after obtaining the candidate position corresponding to the entire path, the first position l _c(1) is selected as the reference position, and all fingerprint data at time t _k are aggregated together to obtain a set of reference points R _k ={l _r1 ,l _r2 ,...,l _rm }, where l _rm represents the first RSS values on m reference points, each reference point corresponds to a position estimate l _ri =(x _i , y _i ), i=1,2,...,m, x _i , y _j represent the horizontal direction of the position respectively Y-axis.

Step 3: Model the relationship of the reference point to the RSS fingerprints of other non-reference points

Learning and modeling the fingerprint relationship between reference points and non-reference points, the specific steps are as follows:

(1) For a series of reference points R _k obtained at time t _k , where the position of the jth reference point is l _rj , the relationship between the position contained in R _k and the RSS at other positions in the fingerprint map should be learned. For the prediction model θ, taking the RSS of the jth AP (1≤j≤p) at position l _i and 1≤i≤n as an example, the relational model to be learned is as follows θ _ij :

f _ij (t ₀ )=θ _ij (f _r1j (t ₀ ),f _r2j (t ₀ ),...,f _rmj (t ₀ ),) (9)

Here f _ij (t ₀ ) and f _rmj (t ₀ ) represent the RSS values of the jth AP at position l _i and position l _rm in the initial fingerprint map, respectively.

(2) Partial least square regression (PLSR) is used to establish a function regression model. In short, PLSR looks for the part of the component decomposition of the independent variable X that is also related to the dependent variable Y. Firstly, X and Y are simultaneously decomposed by searching a set of latent vectors to maximize the correlation coefficient between X and Y, and then the decomposition of X is used to predict Y. In general, PLSR still has the form of multiple regression Y=XB+E, where B=X ^T U(T ^T XX ^T U) ^-1 T ^T Y, where T and U are the latent variable matrices and E is the residual matrix.

也就是来自m个参考点的RSS观测值，

是位置l_i上的RSS测量值，由于Y是一个一维向量，这里记为y,针对于单因变量的PLSR问题，可以采用PLS1方法求解，具体的，对于第j个潜在变量，按如下规则求最大化协方差cov(X_jw_j,y_j)并满足条件

的t_j＝X_jw_j：In the fingerprint map,

That is, the RSS observations from m reference points,

is the RSS measurement value at position l _i . Since Y is a one-dimensional vector, it is denoted as y here. For the PLSR problem of a single variable, the PLS1 method can be used to solve it. Specifically, for the jth latent variable, as follows The rule seeks to maximize the covariance cov(X _j w _j , y _j ) and satisfy the condition

t _j =X _j w _j :

t _j =X _j w _j (11)

When finding the first latent variable, let X ₁ =X and y ₁ =y. To find the next latent variable _{tj +1} , remove their respective _tj -based regression estimates from _Xj and yj, and then repeat the above steps to solve for the latent variable with the degraded residual values.

构成的列向量

由此可以得到PLSR的预测模型：Thus, after repeating the h rounds, two m×h matrices W and P, and an n×h matrix T can be obtained. The three matrices take w _j , p _j and t _j as column vectors respectively. At the same time, there must be a set of h

composed of column vectors

From this, the prediction model of PLSR can be obtained:

是预测值，

in

is the predicted value,

Step 4: According to the modeling model, adaptively update the fingerprint map.

Once a sufficient number of reference point locations and their real-time measurement data are obtained, an update process can be triggered to update the current fingerprint map to the latest state. Since the number of reference points and their corresponding positions are different each time the fingerprint map update is started, the targeted regression function must be re-learned from the initial fingerprint map before each update.

The above embodiments should be understood as only for illustrating the present invention and not for limiting the protection scope of the present invention. After reading the contents of the description of the present invention, the skilled person can make various changes or modifications to the present invention, and these equivalent changes and modifications also fall within the scope defined by the claims of the present invention.

Claims (6)

1. an indoor fingerprint map adaptive updating method based on communication survey, is characterized in that, comprises the following steps: Step 1: use the mobile terminal to collect static and dynamic wireless fingerprint data; wherein the wireless fingerprint data collection is the real-time data automatically collected by the user's mobile device in the process of their daily work and life; Step 2: Estimate the position of the reference point based on the path matching method for the collected wireless fingerprint data; Step 3: Model the relationship between the reference point and the RSS fingerprints of other non-reference points; Step 4: According to the modeling model, adaptively update the fingerprint map. 2. a kind of indoor fingerprint map adaptive updating method based on communication survey according to claim 1, is characterized in that, the static wireless signal fingerprint of described step 1 is when the mobile device is kept at a certain position for a period of time static When the user moves, the wireless fingerprint data and the movement data are collected at the same time to monitor the user's moving path. 3. a kind of indoor fingerprint map adaptive update method based on communication survey according to claim 1, is characterized in that, in described step 2, to the wireless fingerprint data collected, based on the mode of path matching, to the reference point position estimation ,Specific steps are as follows: (1) Estimate the feasible area according to the RSS fingerprint, use the RSS fingerprint measurement value on the moving path as the initial value of the overall path, search for candidate positions in a part of its subspace, and match the feasible area, usually the fingerprint on a path corresponds to The rough position estimate always falls within a limited area, which is no larger than the entire position space, outlines a feasible area covering all fingerprint positions on the moving path, and only finds the best match for the entire path within this feasible area; (2) Lock the feasible azimuth, and set the possible maximum directional error to be Δφ, then only the center direction

Consider the symmetric direction interval on both sides of

(3) Joint position estimation, the translation increment Δα and rotation increment Δβ are used to embed the moving path on the fingerprint map, and the matching algorithm finds the path J={s ₁ , s ₂ which can minimize the geometric constraints of the path under the premise of considering the path. ,s ₃ ,...s _w } The sequence position of the mean square error of all fingerprints is taken as the target position, and s _w represents the wth fingerprint measurement value in the path;

where d' _j =||l _c(j+1) -l _c(j) || represents the distance between two candidate positions, f _c(j) represents the fingerprint data of the candidate positions, and d _j represents the phase in the path. The distance between adjacent positions, and c _j is the candidate position corresponding to s _j , Δd is a minimum distance constraint value, which can be set according to the specific environment and needs, after obtaining the candidate position corresponding to the entire path, the first position l _c(1) is selected as the reference position, and all fingerprint data at time t _k are aggregated together to obtain a set of reference points R _k ={l _r1 ,l _r2 ,...,l _rm }, where l _rm represents the first RSS values on m reference points, each reference point corresponds to a position estimate l _ri =(x _i , y _i ), i=1,2,...,m, x _i , y _j represent the horizontal direction of the position respectively Y-axis. 4. a kind of indoor fingerprint map adaptive update method based on communication survey according to claim 3, is characterized in that, in described step 3, the relation of the RSS fingerprint of reference point and other non-reference points is modeled, its Specific steps are as follows: (1) For a series of reference points R _k obtained at time t _k , where the position of the jth reference point is l _rj , the relationship between the position contained in R _k and the RSS at other positions in the fingerprint map should be learned. Prediction model θ, the relationship model that needs to be learned for the jth AP (1≤j≤p) at position l _i , 1≤i≤n RSS is as follows θ _ij : f _ij (t ₀ )=θ _ij (f _r1j (t ₀ ),f _r2j (t ₀ ),...,f _rmj (t ₀ ),) (2) Here f _ij (t ₀ ) and f _rmj (t ₀ ) represent the RSS values of the jth AP at position l _i and position l _rm in the initial fingerprint map, respectively; (2) Use partial least squares regression to establish a functional regression model PLSR, which specifically includes: first, decompose X and Y at the same time by searching a set of latent vectors to maximize the correlation coefficient between X and Y; Decomposition is used to predict Y, PLSR still has the form of multiple regression Y=XB+E, where B=X ^T U(T ^T XX ^T U) ^-1 T ^T Y, where T and U are the latent variable matrix and E is the residual difference matrix; In the fingerprint map,

That is, the RSS observations from m reference points,

is the RSS measurement value at position _li . Since Y is a one-dimensional vector, it is denoted as y here. For the PLSR problem of a single variable, the PLS1 method is used to solve it. 5. a kind of indoor fingerprint map adaptive updating method based on communication survey according to claim 4, is characterized in that, described for the PLSR problem of single dependent variable, adopts PLS1 method to solve, specifically comprises: for the jth Latent variables, maximize the covariance cov(X _j w _j ,y _j ) according to the following rules and satisfy the conditions

t _j =X _j w _j :

t _j =X _j w _j (4)

When finding the first latent variable, let X ₁ =X and y ₁ =y. When finding the next latent variable t _j+1 , remove their respective regression estimates based on t _j from X _j and y _j , and then use the degraded residual values to repeat the above steps to solve the latent variable; Thus, after repeating the h rounds, two m×h matrices W and P, and an n×h matrix T can be obtained. The three matrices take w _j , p _j and t _j as column vectors respectively, and at the same time, One more by h

composed of column vectors

From this, the prediction model of PLSR can be obtained:

in

is the predicted value,

6. a kind of indoor fingerprint map adaptive update method based on communication survey according to claim 5, is characterized in that, in described step 4, according to modeling model, adaptively update fingerprint map, specifically comprises: when sufficient quantity is obtained The location of the reference point and its real-time measurement data, the update process can be triggered to update the current fingerprint map to the latest state. Since each time the fingerprint map update is started, the number of reference points and their corresponding positions are different, so Before each update, the targeted regression function is relearned from the initial fingerprint map.

Images