CN109413578B - An indoor positioning method based on the fusion of WIFI and PDR - Google Patents

Abstract

The invention discloses an indoor positioning method based on the fusion of WIFI and PDR, and relates to the technical field of indoor positioning of WIFI signals and pedestrian dead reckoning. The technical problem to be solved is to provide an indoor positioning method with high measurement accuracy, including the following steps: (1) Establish WIFI offline fingerprint database; (2) Cluster training samples to obtain clustered samples and corresponding categories; (3) Obtain positioning coordinates through weighted K-nearest neighbor algorithm; (4) Integrate PDR positioning to update status and position; ( 5) Use the fusion result as the correction source of the PDR; (6) Obtain the correction factor to correct the PDR result by evaluating the parameters. The invention shortens the WIFI positioning time, improves the positioning accuracy, and has the characteristics of high positioning accuracy and low software calculation amount, and realizes the real-time positioning requirement on the premise of ensuring the positioning accuracy.

Description

An indoor positioning method based on the fusion of WIFI and PDR

technical field

The invention relates to the technical field of indoor positioning of WIFI signals and pedestrian dead reckoning, in particular to an indoor positioning method based on the fusion of WIFI and PDR.

Background technique

According to statistics, people spend 80% of their lives indoors, but GPS cannot work indoors. Industries such as travel navigation, smart manufacturing, and smart services also urgently need people to re-examine the value of indoor locations. Indoor positioning technology, as the key to open the door of indoor location services, has been paid more and more attention in recent years. At present, the relatively mature indoor positioning includes ultrasonic positioning, UWB positioning, inertial navigation positioning, radio frequency identification (RFID) positioning, Bluetooth positioning, WIFI positioning and other technologies. Compared with other indoor wireless positioning technologies, WIFI has its unique advantages. WIFI hotspots are located in every corner and building area of the city. Due to its ubiquity, its deployment cost is low, the hardware is easy to install, and it is easy to combine with smart phones for positioning. And its wide coverage, high positioning accuracy, easy to implement, has quickly become a research hotspot of indoor positioning technology.

During the positioning process using WIFI fingerprints, due to the complex indoor environment, the WIFI signal is susceptible to interference, the signal strength is prone to large jumps, and there are areas that the WIFI signal cannot cover, which leads to large deviations in WIFI positioning. Therefore, only using WIFI technology for indoor positioning cannot meet people's needs.

In the process of indoor positioning using smart terminals, because mobile terminals are generally equipped with motion sensors such as gyroscopes, acceleration sensors, electronic compasses, etc., the inertial navigation technology of mobile terminals has better promotion, and is not easily affected by the environment. high advantage. However, since the electronic compass is easily disturbed by the environment, the heading angle will be deviated, and the gait judgment error and step length estimation error will lead to the walking distance error. Eliminating accumulated errors becomes the key to solving the problem.

Chinese invention CN106610292 A discloses an indoor positioning method of hybrid WIFI and dead reckoning (PDR). The initial position of the target to be measured is obtained by using the WIFI positioning subsystem. After the initial position of the target to be measured is obtained, two sets of The positioning subsystem WIFI positioning subsystem and PDR positioning subsystem will output the position of the target to be measured respectively. The system uses a time window utility function to effectively linearly weight the coordinates output by the WIFI positioning subsystem and the coordinates output by the PDR-based positioning subsystem. coordinate. The positioning method does not consider the PDR fusion calibration. Due to the instability of WIFI positioning and the cumulative error of PDR positioning, the obtained positioning coordinate accuracy is still inaccurate.

Chinese invention CN107302754 A discloses a simple indoor positioning method based on WIFI and PDR. By setting the initial position and step length, the motion state of pedestrians is determined by the acceleration measured by the acceleration sensor, and the WIFI positioning method is carried out by the RSSI value. RSSI values of multiple reference nodes, when it is detected that the RSSI value exceeds the threshold, the improved algorithm is used to calculate the current position, and the current position is used as the actual position value, and the PDR positioning is performed to estimate the position of indoor pedestrians. In this method, the initial position of PDR is self-positioning, and calibration is not considered. Due to the instability of WIFI positioning and the cumulative error of PDR positioning, the obtained positioning coordinate accuracy is also inaccurate.

SUMMARY OF THE INVENTION

Aiming at the deficiencies of the prior art, the problem solved by the present invention is to provide an indoor positioning method with high measurement accuracy.

In order to solve the above technical problems, the technical solution adopted in the present invention is an indoor positioning method based on the fusion of WIFI and PDR, including the following steps:

(1) Establish WIFI offline fingerprint database: arrange M AP nodes transmitting WIFI hotspots in the positioning area, divide the positioning area into N nodes, calculate the actual position of each node, and collect the signal strength from each AP node in the N nodes RSSI, obtain the RSSI training samples, correspond the RSSI training samples to their actual positions, and establish a fingerprint database I; the fingerprint data set I is expressed as:

I={(r ₁ ,o ₁ ),(r ₂ ,o ₂ ),...,(r _i ,o _i ),...,(r _N ,o _N )} (1)

Among them, the vector r _i =(r _i1 ,r _i2 ,...,r _iM )∈R ^M represents the RSSI from M APs, r _i1 is the RSSI from the first WIFI hotspot, and r _iM is the Mth AP RSSI, the position vector o _i =(x,y)∈R ² represents the position corresponding to the _ri vector, x is the x-axis coordinate of the position, y is the y-axis coordinate of the position, and the RSSI vector r _i of the RSSI training sample and the position vector o _i are known, i=1,2,...N;

(2) Clustering training samples to obtain clustered samples and corresponding categories: The fingerprint database I is clustered by K-means algorithm, and the fingerprint database I is divided into V categories {c ₁ , c ₂ , .c _v ..c _V }, each class center vector is defined as {C ₁ , C ₂ ,...C _V }, the RSSI training samples in the fingerprint database are divided into V classes respectively, and the clustering sample data set is obtained:

L={(r ₁ ,c _v1 ),(r ₂ ,c _v2 ),...(r _i ,c _vi ),...,(r _N ,c _vN )} (2),

表示其属于的V类别，c_v类由p个RSSI训练样本{r_v1,r_v2,...,r_vp}组成，对所有c_v类RSSI训练样本取平均值得到c_v类中心向量

Among them, r _i =(r _i1 ,r _i2 ,...,r _iM )∈R ^M represents the RSSI of M APs,

Indicates the V category it belongs to. The c _v category consists of p RSSI training samples {r _v1 ,r _v2 ,...,r _vp }, and the average of all the c _v RSSI training samples is obtained to obtain the c _v class center vector

(3) Obtain the positioning coordinates through the weighted K nearest neighbor algorithm: obtain the real-time RSSI samples online, find the most similar class to the real-time RSSI samples obtained, and predict the attribute values of the real-time RSSI samples obtained online through the K nearest neighbor samples. The positions corresponding to the RSSI training samples are weighted to obtain their position coordinates; the real-time RSSI samples obtained online are recorded as T=((r _j1 ,r _j2 ,...,r _jM ),(x,y)), where online The acquired real-time RSSI sample r is r _j =(r _j1 ,r _j2 ,...,r _jM )∈R ^M , and the position vector o=(x,y) represents the position information of the real-time RSSI sample acquired online, where online The obtained RSSI vector is known, and the position vector (x, y) is unknown;

的余弦相似度定义为：First, calculate the similarity between the r vector and the V class center RSSI vectors {C ₁ , C ₂ ,...C _V }, and determine the class most similar to the r vector. The r vector {r _j1 ,r _j2 ,..., r _jM } and the RSSI vector of the ith class center

The cosine similarity of is defined as:

In the same way, calculate the similarity between the r vector and the RSSI vector of other categories. After V calculations, V cosine similarities {Sim ₁ , Sim ₂ ,..., Sim _V } are obtained, where the category corresponding to the maximum similarity c _v is the most similar category to the r vector;

Determine the K samples r _K1 , r _K2 , .r _Ki .., r _KK that are most similar to the most similar category _cv and r vector. The similarity is Sim _K1 , Sim _K2 ,..., Sim _KK ,

(x _K1 , y _K1 ), (x _K2 , y _K2 ),...,(x _KK ,y _KK ) represent the corresponding coordinates, (r _i1 ,r ₁₂ ,...r _iM ) represent the selected most similar The RSSI sample corresponding to the sample r _Ki is normalized to process the similarity, and the weight that affects the positioning result is defined as {w _K1 ,w _K2 ,...,w _KK };

Weight the sample coordinates to get the position coordinates of the RSSI obtained online:

x=w _K1 x _K1 +w _K2 x _K2 +...+w _KK x _KK (6),

y=w _K1 y _K1 +w _K2 y _K2 +...+w _KK y _KK (7),

(4) Integrate PDR positioning to update the state and position: build a pedestrian walking system model X _k , the initial position coordinates of PDR positioning are obtained through WIFI positioning, the error threshold is obtained by integrating WIFI positioning and PDR positioning, and walking steps are obtained through acceleration sensor. long, obtain the heading angle change after walking through the built-in gyroscope of the smart terminal, and use the measurement equation Z _k to update the state information and position information;

The system model X _k of the pedestrian walking is as follows:

为步长平均值，采用步长模型匹配加速度传感器结果，设步长是60cm，

为朝向角变化量；量测方程Z_k如下所示：The number of walking steps is represented by k, the position information after walking is represented by x _k , y _k , θ _k represents the azimuth and orientation angle after k steps, W _k-1 represents noise,

is the average value of the step length, the step length model is used to match the acceleration sensor results, and the step length is set to 60cm,

is the heading angle change; the measurement equation Z _k is as follows:

Among them, x _k , y _k represent the positioning results obtained through WIFI positioning; s _k represents the average walking step length of pedestrians, obtained from the results of the acceleration sensor, Δθ _k represents the change in the orientation angle of the pedestrian after walking, which can be obtained through the built-in gyroscope of the smart terminal Obtain, θ _k represents the orientation angle of the pedestrian after walking; V _k represents the noise.

(5) use the fusion result as the correction source of PDR;

Set the PDR deviation threshold threshold ε _pdr , if the deviation exceeds the set threshold, correct the positioning result to obtain the fusion positioning result of WIFI and PDR, and use the fusion result of WIFI and PDR as the correction source of PDR, see formula (9);

(6) Obtain the correction factor to correct the PDR result by evaluating the parameters;

Use the data in the interval time period, that is, in the [0, T] time, to obtain the evaluation parameters. The evaluation parameters are shown in formula (13) and formula (14), and the correction factor is obtained through the evaluation parameters. The correction factor is shown in formula (15). The results are corrected, and the specific process is as follows:

表示通过融合定位后的定位结果，

表示PDR单独定位结果，PDR单独定位结果由下式获得

Indicates the positioning result after fusion positioning,

Represents the PDR single positioning result, and the PDR single positioning result is obtained by the following formula

y ₂ =y ₁ +S ₁₂ cos θ ₁ (10),

x ₂ =x ₁ +S ₁₂ sin θ ₁ (11),

(x ₁ , x ₂ ) is the initial position of the PDR, (x ₂ , y ₂ ) is the PDR positioning coordinate, the displacement S is obtained by matching the acceleration sensor with the step size model, and the pedestrian step size is 60cm, and the direction angle θ is passed through the sensor and gyroscope. get,

The positioning result difference is:

Set in [0,K] time, the cumulative error is expressed as:

Using this feature as a correction factor for PDR positioning, the correction principles are as follows:

If the absolute values of accumulated errors C _x and C _y ∈ [0,σ ₁ ], the positioning error can be ignored and no correction is required;

If the absolute values of cumulative errors C _x and C _y ∈ [σ ₁ ,σ ₂ ], the error evaluation requirements are met, and the positioning results are corrected;

If the absolute values of cumulative errors C _x and C _y ∈ [σ ₂ ,∞], the error evaluation requirements are not met, and the WIFI positioning error is large, and the fusion positioning is performed again.

Due to the cumulative error problem in the PDR positioning results, the correction information will not correspond, and the correction ability is too weak. The correction scale factor α is set to represent the proportional relationship between the error of the current time and the error of the previous period, [0, K] represents the time segment, C _N =αC _k represents the proportional relationship, and the correction process in the M time segment is smoothed by the following formula to solve the problem of sudden change in the positioning result:

In the formula, j∈[1,M].

Compared with the prior art, the beneficial effects of the present invention:

The K-means algorithm is combined with the K-nearest neighbor method, which shortens the WIFI positioning time and improves the positioning accuracy. Using the WIFI positioning result and PDR fusion positioning, use the correction factor to correct the positioning factor of the system. It has the characteristics of high positioning accuracy and low software calculation amount, and realizes real-time requirements under the premise of ensuring positioning accuracy.

Description of drawings

Fig. 1 is the method flow chart of the present invention;

Figure 2. The fusion structure diagram of WIFI positioning and PDR positioning.

Detailed ways

The specific embodiments of the present invention are further described below with reference to the accompanying drawings, but the present invention is not limited.

Figure 1 shows an indoor positioning method based on the fusion of WIFI and PDR, including the following steps:

(1) Establish a WIFI offline fingerprint database: arrange M AP nodes transmitting WIFI hotspots in the positioning area, divide the positioning area into N nodes, calculate the actual position of each node, and collect the signal strength from each AP node in the N nodes RSSI, obtain the RSSI training samples, correspond the RSSI training samples to their actual positions, and establish a fingerprint database I; the fingerprint data set I is expressed as:

I={(r ₁ ,o ₁ ),(r ₂ ,o ₂ ),...,(r _i ,o _i ),...,(r _N ,o _N )} (1),

Among them, the vector r _i =(r _i1 ,r _i2 ,...,r _iM )∈R ^M represents the RSSI from M APs, r _i1 is the RSSI from the first WIFI hotspot, and r _iM is the Mth AP RSSI, the position vector o _i =(x,y)∈R ² represents the position corresponding to the _ri vector, x is the x-axis coordinate of the position, y is the y-axis coordinate of the position, and the RSSI vector r _i of the RSSI training sample and the position vector o _i are known, i=1,2,...N;

(2) Clustering training samples to obtain clustered samples and corresponding categories: The fingerprint database I is clustered by K-means algorithm, and the fingerprint database I is divided into V categories {c ₁ , c ₂ , .c _v ..c _V }, each class center vector is defined as {C ₁ , C ₂ ,...C _V }, the RSSI training samples in the fingerprint database are divided into V classes respectively, and the clustering sample data set is obtained:

L={(r ₁ ,c _v1 ),(r ₂ ,c _v2 ),...(r _i ,c _vi ),...,(r _N ,c _vN )} (2),

表示其属于的V类别，c_v类由p个RSSI训练样本{r_v1,r_v2,...,r_vp}组成，对所有c_v类RSSI训练样本取平均值得到c_v类中心向量

Among them, r _i =(r _i1 ,r _i2 ,...,r _iM )∈R ^M represents the RSSI of M APs,

Indicates the V category it belongs to. The c _v category consists of p RSSI training samples {r _v1 ,r _v2 ,...,r _vp }, and the average of all the c _v RSSI training samples is obtained to obtain the c _v class center vector

(3) Obtain the positioning coordinates through the weighted K nearest neighbor algorithm: obtain the real-time RSSI samples online, find the most similar class to the real-time RSSI samples obtained, and predict the attribute values of the real-time RSSI samples obtained online through the K nearest neighbor samples. The positions corresponding to the nearest neighbor training samples are weighted to obtain their position coordinates; the real-time RSSI samples obtained by the line are recorded as T=((r _j1 ,r _j2 ,...,r _jM ),(x,y)), where The real-time RSSI sample r obtained online is represented by r _j =(r _j1 ,r _j2 ,...,r _jM )∈R ^M , which represents the RSSI of M APs received online, and the position vector o=(x,y) represents The location information of the real-time RSSI samples obtained online, where the RSSI vector obtained online is known and the position vector (x, y) is unknown,

的余弦相似度定义为：First, calculate the similarity between the r vector and the V class center RSSI vectors {C ₁ , C ₂ ,...C _V }, and determine the class most similar to the r vector. The r vector {r _j1 ,r _j2 ,..., r _jM } and the RSSI vector of the ith class center

The cosine similarity of is defined as:

In the same way, calculate the similarity between the r vector and the RSSI vector of other categories. After V calculations, V cosine similarities {Sim ₁ , Sim ₂ ,..., Sim _V } are obtained, where the category corresponding to the maximum similarity c _v is the most similar category to the r vector;

Determine the K samples r _K1 , r _K2 , .. r _Ki ., r _KK that are most similar to the most similar category _cv and r vector. The similarity is Sim _K1 , Sim _K2 ,..., Sim _KK ,

(x _K1 , y _K1 ), (x _K2 , y _K2 ),...,(x _KK ,y _KK ) represent the corresponding coordinates, (r _i1 ,r ₁₂ ,...r _iM ) represent the selected most similar The RSSI sample corresponding to the sample r _Ki is normalized to process the similarity, and the weight that affects the positioning result is defined as {w _K1 ,w _K2 ,...,w _KK };

Weight the sample coordinates to get the position coordinates of the RSSI obtained online:

x=w _K1 x _K1 +w _K2 x _K2 +...+w _KK x _KK (6),

y=w _K1 y _K1 +w _K2 y _K2 +...+w _KK y _KK (7),

(4) Integrate PDR positioning to update the state and position: build a pedestrian walking system model X _k , the initial position coordinates of PDR positioning are obtained through WIFI positioning, the error threshold is obtained by integrating WIFI positioning and PDR positioning, and walking steps are obtained through acceleration sensor. Long, through the built-in gyroscope of the intelligent terminal to obtain the change of the orientation angle after walking, and use the measurement equation Z _k to update the state information and position information, the system model of the pedestrian walking is as follows:

为步长平均值，通过加速度传感器获得，

为朝向角变化量，量测方程Z_k如下所示：The number of walking steps is represented by k, the position information after walking is represented by x _k , y _k , θ _k represents the azimuth and orientation angle after k steps, W _k-1 represents noise,

is the average value of the step size, obtained by the acceleration sensor,

is the heading angle change, the measurement equation Z _k is as follows:

Among them, x _k , y _k represent the positioning results obtained by WIFI positioning; s _k represents the average walking step length of pedestrians, obtained by the acceleration sensor, Δθ _k represents the change in the orientation angle of the pedestrian after walking, obtained by the built-in gyroscope of the smart terminal, θ _k represents the orientation angle of the pedestrian after walking; V _k represents the noise;

(5) Use the fusion result as the PDR correction source: set the PDR deviation threshold ε _pdr , if the deviation exceeds the set threshold, correct the positioning result to obtain the WIFI and PDR fusion positioning results, use the WIFI and PDR fusion results as Correction source of PDR, see formula (9);

(6) Obtain the correction factor through the evaluation parameters to correct the PDR results: use the data in the interval time period, that is, in the [0, T] time, to obtain the evaluation parameters. The evaluation parameters are shown in formulas (13) and (14). The factor correction factor is shown in formula (15) to correct the PDR results; the specific process is as follows:

表示通过融合定位后的定位结果，

表示PDR单独定位结果，PDR单独定位由下式获得，

Indicates the positioning result after fusion positioning,

Indicates the result of PDR single positioning, PDR single positioning is obtained by the following formula,

y ₂ =y ₁ +S ₁₂ cos θ ₁ (10),

x ₂ =x ₁ +S ₁₂ sin θ ₁ (11),

(x ₁ , x ₂ ) is the initial position of the PDR, (x ₂ , y ₂ ) is the PDR positioning coordinate, the displacement S is obtained by matching the acceleration sensor with the step size model, and the pedestrian step size is 60cm, and the direction angle θ is passed through the sensor and gyroscope. get,

The positioning result difference is:

Set in [0,K] time, the cumulative error is expressed as:

Using this feature as a correction factor for PDR positioning, the correction principles are as follows:

If the absolute values of accumulated errors C _x and C _y ∈ [0,σ ₁ ], the positioning error can be ignored and no correction is required;

If the absolute values of cumulative errors C _x and C _y ∈ [σ ₁ ,σ ₂ ], the error evaluation requirements are met, and the positioning results are corrected;

If the absolute values of the accumulated errors C _x and C _y ∈ [σ ₂ ,∞], the error evaluation requirements are not met, and the WIFI positioning error is large, and the fusion positioning is performed again;

Due to problems such as accumulated errors in the PDR positioning results, the correction information will not correspond, and the correction ability is too weak. By setting the correction scale factor α to represent the proportional relationship between the error of the current time and the error of the previous period, [0, K] means Time period, C _N = αC _k represents the proportional relationship, and the correction process in the M time period is smoothed by the following formula to solve the problem of sudden changes in the positioning results:

In the formula, j∈[1,M].

Figure 2 shows the fusion structure of WIFI positioning and PDR positioning. The initial position coordinates of PDR positioning are obtained through WIFI positioning, and the error threshold is obtained by merging WIFI positioning and PDR positioning. The fusion result of WIFI and PDR is used as the correction source of PDR. PDR deviation threshold threshold, if the deviation exceeds the set threshold, correct the positioning result; use the data of the interval time period to obtain the evaluation parameters, obtain the correction factor through the evaluation parameters, and correct the PDR results.

Compared with the prior art, the beneficial effects of the present invention:

The K-means algorithm is combined with the K-nearest neighbor method, which shortens the WIFI positioning time and improves the positioning accuracy. Using the WIFI positioning result and PDR fusion positioning, use the correction factor to correct the positioning factor of the system. It has the characteristics of high positioning accuracy and low software calculation amount, and realizes real-time requirements under the premise of ensuring positioning accuracy.

The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, without departing from the principle and spirit of the present invention, various changes, modifications, substitutions and alterations to these embodiments still fall within the protection scope of the present invention.

Claims (2)

1. An indoor positioning method based on WIFI and PDR fusion is characterized by comprising the following steps:

(1) establishing a WIFI offline fingerprint database: arranging M APs for transmitting WIFI hotspots in a positioning area, dividing the positioning area into N nodes, calculating the actual position of each node, acquiring the signal strength RSSI from each AP in the N nodes to obtain an RSSI training sample, corresponding the RSSI training sample to the actual position of the RSSI training sample, and establishing a fingerprint database I;

(2) clustering the training samples to obtain clustering samples and corresponding categories: clustering the fingerprint database I through a K-means algorithm, and dividing the fingerprint database I into V classes { c₁,c₂,.c_v..c_VEach class center vector is defined as { C }₁,C₂,...C_VDividing RSSI training samples in the fingerprint database into V classes respectively to obtain a clustering sample data set:

L＝{(r₁,c_v1),(r₂,c_v2),...(r_i,c_vi),...,(r_N,c_vN)} (2)，

wherein r is_i＝(r_i1,r_i2,...,r_iM)∈R^MRSSI of M APs, their corresponding categories

Indicates the V category to which it belongs, c_vClass is trained by p RSSI samples { r_v1,r_v2,...,r_vpComposition of, for all c_vC is obtained by averaging RSSI-like training samples_vClass center vector

(3) Obtaining positioning coordinates through a weighted K nearest neighbor algorithm: acquiring real-time RSSI samples on line, finding out a class which is most similar to the acquired real-time RSSI samples in a fingerprint database I, predicting attribute values of the real-time RSSI samples acquired on line through K nearest neighbor samples, and weighting positions corresponding to the K nearest neighbor samples to acquire position coordinates of the K nearest neighbor samples; the real-time RSSI sample obtained online is denoted as T ═ r ((r)_j1,r_j2,...,r_jM) (x, y)), where the real-time RSSI samples r taken online are defined by r_j＝(r_j1,r_j2,...,r_jM)∈R^MIndicating RSSI of M APs received online, and a location vector o ═ x, y indicating location information of real-time RSSI samples acquired online, where the RSSI vector acquired online is known and the location vector (x, y) is unknown;

first, r vector and V class center RSSI vectors { C are calculated₁,C₂,...C_VThe similarity of the r vector, r, is determined_j1,r_j2,...,r_jMRSSI vector with the ith class center

The cosine similarity of (a) is defined as:

similarly, calculating the similarity between the r vector and the RSSI vectors of other categories, and obtaining V cosine similarities { Sim after V times of calculation₁,Sim₂,...,Sim_VIn which the category c corresponding to the maximum similarity is_vIs the most similar class to the r-vector,

determining the most similar class c_vK samples r most similar to the r vector_K1,r_K2,.r_Ki..,r_KKSimilarity is respectively Sim_K1,Sim_K2,...,Sim_KK,

(x_K1,y_K1)，(x_K2,y_K2),...,(x_KK,y_KK) Represents the corresponding coordinates, (r)_i1,r₁₂,...r_iM) Representing the most similar sample r selected_KiCorresponding RSSI samples, normalization processing similarity, defining the weight affecting the positioning result as { w }_K1,w_K2,...,w_KK}，

Weighting the sample coordinates to obtain the position coordinates of the on-line acquired RSSI:

x＝w_K1x_K1+w_K2x_K2+...+w_KKx_KK(6)，

y＝w_K1y_K1+w_K2y_K2+...+w_KKy_KK(7)，

(4) updating the state and the position by fusing PDR positioning;

system model X for constructing pedestrian walking_kAcquiring initial position coordinates of PDR positioning through WIFI positioning, acquiring an error threshold by fusing the WIFI positioning and the PDR positioning, acquiring a walking step length through an acceleration sensor, acquiring the variation of an orientation angle after walking through a gyroscope arranged in an intelligent terminal, and updating state information and position information by using a measurement equation;

(5) using the fusion results as a correction source for PDR: setting PDR deviation threshold_pdrIf the deviation exceeds a set threshold value, correcting the positioning result to obtain a WIFI and PDR fusion positioning result, and using the WIFI and PDR fusion result as a PDR correction source, referring to a formula (9);

(6) obtaining correction factor corrected PDR results by evaluating parameters: obtaining an evaluation parameter by using data in an interval time period, namely [0, T ] time, wherein the evaluation parameter is shown in formula (13) and formula (14), obtaining a correction factor by the evaluation parameter, and the correction factor is shown in formula (15), and correcting the PDR result, wherein the specific process is as follows:

indicating the positioning result after positioning by fusion,

representing the PDR individual positioning result, which is obtained by the following formula,

y₂＝y₁+S₁₂cosθ₁(10)，

x₂＝x₁+S₁₂sinθ₁(11)，

(x₁,x₂) Is PDR initial position, (x)₂,y₂) The coordinates are located for the PDR,the displacement S is obtained by matching the step length model with an acceleration sensor, the step length of the pedestrian is selected to be 60cm, and the direction angle theta is obtained through the sensor and a gyroscope;

the positioning result difference is as follows:

given a [0, K ] time, the accumulated error is expressed as:

the characteristic is used as a correction factor for PDR positioning, and the correction principle is as follows:

if the error C is accumulated_xAnd C_yThe absolute value is E [0, sigma ]₁]The positioning error is negligible and does not need to be corrected;

if the error C is accumulated_xAnd C_yAbsolute value ∈ [ σ ]₁,σ₂]The error evaluation requirement is met, and the positioning result is corrected;

if the error C is accumulated_xAnd C_yAbsolute value ∈ [ σ ]₂,∞]If the error evaluation requirement is not met, the WIFI positioning error is large at the moment, fusion positioning is carried out again, a correction scale factor alpha is set to represent the proportional relation between the error of the current time and the error of the previous time, and [0, K ]]Represents a time period, C_N＝αC_kExpressing a proportional relation, smoothing a correction process in the M time period by the following formula, and solving the problem of mutation of the positioning result:

wherein j belongs to [1, M ], and the system model of pedestrian walking is as follows:

wherein the number of steps of walking is represented by k, and the position information after walking is represented by x_i,y_iIs represented by theta_kDenotes the azimuth angle after k steps, W_k-1The representation of the noise is represented by,

the step length is the average value, the result of the acceleration sensor is matched through a step length model, the step length is selected to be 60cm,

is the orientation angle variation; the measurement equation is as follows:

wherein x is_k,y_kRepresenting a positioning result obtained through WIFI positioning; s_kRepresenting the average walking step length of the pedestrian, obtained by an acceleration sensor, Delta theta_kRepresenting the variation of the orientation angle of the pedestrian after walking, and obtaining the variation through a built-in gyroscope of the intelligent terminal_kRepresenting the orientation angle of the pedestrian after walking; v_kRepresenting noise.

2. The WIFI and PDR fusion based indoor positioning method according to claim 1, wherein in step (1), the number set of training samples is represented as:

I＝{(r₁,o₁),(r₂,o₂),...,(r_i,o_i),...,(r_N,o_N)} (1)

wherein, the vector r_i＝(r_i1,r_i2,...,r_iM)∈R^MRepresents the RSSI vector, r, from M APs_i1Is RSSI, r from the 1 st WIFI hotspot_iMFor RSSI of Mth AP, position vector o_i＝(x,y)∈R²Is represented by r_iThe position corresponding to the vector, x isThe x-axis coordinate of the position, y is the y-axis coordinate of the position, and the RSSI vector r of the training sample_iAnd a position vector o_iAre known, i ═ 1, 2.. N.

Images