CN108566675B - WiFi indoor positioning method based on multiple access point selection - Google Patents

Abstract

The invention discloses a WiFi indoor positioning method based on multiple access point selection, which mainly solves the problem of inaccurate positioning result of the access point in the prior art, and adopts the technical scheme that: 1) collecting access point data; 2) according to the access point data, multiple access point selections are carried out, and an access point set with stable signals and strong resolving power is selected to form a position fingerprint database; 3) grouping the positions according to the fingerprints of the positions; 4) reselecting an access point for each position cluster, and selecting a characteristic access point set capable of better expressing the position cluster; 5) establishing a decision tree model for each position group; 6) determining the position grouping of the target position of the positioning sample by the given positioning sample; 7) and determining the specific position of the positioning sample by using the positioning model of the position grouping of the positioning sample. The invention can effectively overcome the influence of unstable access points in the environment on the positioning precision, and can be used in various complex positioning environments.

Description

WiFi indoor positioning method based on multiple access point selection

technical field

The present invention relates to the field of communication technologies, in particular, to a WiFi indoor positioning method, which can be used for location-based services such as positioning and navigation for indoor users.

Background technique

With the vigorous development of mobile communication technology and mobile Internet technology, mobile smart terminals have emerged and become popular rapidly, and various services based on mobile Internet have shown explosive growth. Among them, the location-based service LBS has also experienced rapid development in the past ten years, and has been widely used in people's lives, which has greatly changed people's lifestyles and brought convenience to people's lives. The positioning technology is an essential underlying support technology in LBS applications. In the outdoor environment, the global positioning system GPS is the most mature positioning system. However, indoors, due to the wall barrier, the indoor environment is complex and changeable, and there are many interferences, so it is impossible to use GPS signals for indoor positioning.

With the continuous improvement of the IEEE 802.11 protocol, WiFi has become popular. Although WiFi is not born for positioning, but because of its fast transmission speed, high coverage, free use in public places, no need to install special equipment, and low deployment costs, it has become the first choice for indoor positioning. In the WiFi positioning technology, the current main research direction is the positioning based on the signal strength RSSI, and the WiFi indoor positioning method based on the location fingerprint identification is a very popular method.

The WiFi indoor positioning method based on location fingerprinting is to abstract and formally describe the scene features observed in the environment to be located, and use the correlation between the signal strength RSSI and the physical location to locate. In different physical locations, the expressiveness of the signal strength RSSI is different, that is to say, the signal strength of the AP received by each point is different. By detecting the signal strength of AP points arranged in the positioning environment at each sampling point in advance, extracting the signal strength as a positioning feature value, training it into a mapping relationship with the physical location, and constructing a corresponding location fingerprint database. Then, through a specific matching method, the signal strength RSSI fingerprint data measured in real time at the point to be located is matched with the fingerprint data in the location fingerprint database, and the coordinate positions corresponding to several fingerprints with relatively large similarity are used to estimate the user to be located. s position.

With the rapid spread of WiFi, wireless access points have become ubiquitous. In this way, during data collection, hundreds of access points can be detected in the positioning environment. Among these access points, there are some unsatisfactory access points that are far away from the positioning environment. Their signals are unstable, fluctuate greatly, and carry a lot of noise. They hardly provide valuable information for positioning, and even some Positioning accuracy may be reduced. In addition, when all access points are used, this will greatly increase the dimension of the fingerprint database and increase the complexity of positioning.

SUMMARY OF THE INVENTION

The purpose of the present invention is to propose a WiFi indoor positioning method based on multiple access point selection, so as to select a set of access points that are more stable and capable of distinguishing to represent the location fingerprint, reduce the computational complexity and reduce the dimension of the fingerprint database , Remove the influence of the access point with poor signal quality on the positioning result, and improve the positioning accuracy.

The technical idea of the present invention is as follows: through multiple access point selection, select a set of access points with stable signals and strong discrimination ability, establish a location fingerprint database with lower dimensions and more stable, and locate the location through the location fingerprint database pattern. The locations in the environment are grouped, and the locations with higher fingerprint similarity are grouped into the same location group. Through the re-selection of access points, a set of access points that can better represent the characteristics of each location grouping is selected, and a decision-making model with better positioning effect for location grouping is established.

According to the above thinking, the implementation steps of the present invention include the following:

1) Build the access point signal strength database:

Divide the positioning environment into multiple grids of the same size, and use the position of the center point of the grid to indicate the position of the grid; then collect data for each position, and record the signal strength value of each access point detected at each position , forming an access point signal strength database recording the sampling data of each location;

2) Build the location fingerprint library:

2a) According to the coverage time distribution of each access point in the positioning environment, set the time initial threshold m that the access point covers at least each location, and the time initial threshold n that the access point covers at least the entire positioning space, according to the access point in the database. For the coverage time of each location and the coverage time of the access point in the entire positioning environment, perform multiple selections on the access points, delete the access points whose coverage time is less than these two thresholds, filter out the preselected access points, and form a preselected access point. set of entry points;

2b) Calculate the information gain of each access point in the preselected access point set, sort the access points in descending order of the information gain, and select the first k access points to form the final fingerprint access point set, k is greater than or equal to 10;

2c) Screen the access point signal strength database obtained in step 1) according to the fingerprint access point set, retain only the access point data contained in the fingerprint access point set, obtain the fingerprints of each location, and form the final location fingerprint database ;

3) Use the k-means algorithm to group the environmental locations, and re-select the access points for each location group, and select a feature access point set that can better represent the location group itself;

4) Use the C4.5 decision tree method to establish a decision tree model for each grouping, and obtain a decision model with the fastest reduction of location unknown information;

5) Positioning stage

5a) Given the sample data that needs to be positioned, calculate the Euclidean distance between the positioning sample fingerprint and each position cluster, select the position cluster with the smallest Euclidean distance from it, and use the cluster as the cluster where its target position is located;

5b) The positioning sample moves down from the root node along the grouped decision tree model until it moves to the leaf node of the decision tree, and the leaf node is the position of the final positioning sample.

Compared with the prior art, the present invention has the following advantages:

The present invention can delete unsatisfactory access points and access points with unstable signals through multiple access point selection, and can select a set of access points that are more stable and have strong distinguishing ability, thereby reducing the complexity of positioning calculation and the fingerprint database. It can improve the positioning accuracy and can be applied to more complex positioning environments;

The present invention can select a set of characteristic access points that better represent the location group by re-selection of the location grouping access points, so that a more reasonable positioning model can be established for each grouping, and the positioning accuracy can be further improved.

Description of drawings

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

2 is a schematic diagram of a positioning scenario implemented by the present invention;

3 is a schematic diagram of the positioning results of the present invention under different positioning errors in a positioning scene;

Fig. 4 is the positioning result comparison diagram of the present invention and the positioning method based on the information gain method in the positioning scene within 2m of positioning error;

Figure 5 is a decision tree model for location clustering.

Detailed ways

1, the implementation steps of the present invention are as follows:

Step 1, build the access point signal strength database.

As shown in Figure 2, the location scene of this example is the corridor of Zone I, 4th floor, the main building of Xidian University, with an area of 340m ² ;

In this step, the positioning environment is firstly divided into 177 square grids with a side length of 0.8m, the position of the grid is represented by the position of the center point of the grid, and each position is numbered, and G _j is used to represent the jth each location; then collect data for each location, record the RSSI value of the signal strength of each access point detected at each location, and form an RSSI database that records the sampling data of each location.

Step 2, access point selection.

When selecting an access point in this step, it is divided into the following two steps:

2a) The method preliminarily screens the detected access points based on the coverage time of the access points, deletes those access points with unstable signals, and uses relatively stable access points to form a pre-access point set:

2a1) Calculate the number of access points detected at each location in the detection time period, and keep those access points that cover the location at least 80% of the time in the detection time period, and delete other access points , using the set of access points reserved at each location to form a set of primary pre-selected access points;

2a2) Calculate the number of times that each access point in the primary pre-selected access point set appears at all positions in the positioning environment, and retain those access points that cover all positions at least 20% of the time during the detection period to form the final pre-selected access point. set of entry points;

2b) Calculate the information gain of the access points in the pre-selected access point set based on the information gain method, sort the access points in descending order of the information gain, and select 15 access points with the largest information gain to form the access point set. The calculation process of the information gain is as follows:

2b1) Calculate the uncertainty H(G) of the position information in the positioning environment:

Among them, G represents the position in the positioning environment, G _i represents the ith position, P(G _i ) represents the probability of the occurrence of the position G _i , and 177 is the number of positions in the positioning environment;

2b2) Calculate the uncertainty H(G|AP _i ) of the location information in the positioning environment under the condition of known access points:

Among them, AP _i represents the i-th access point, v _j represents the value of the signal strength of AP _i , N represents the number of values of the signal strength of AP _i , and H(G|AP _i =v _j ) represents the value of the known AP i Under the condition that the signal strength of _i is v _j , the information entropy of the location in the positioning environment, its calculation formula is the same as H(G);

2b3) Calculate the information entropy Gain(AP _i ) of the access point according to the results of 2b1) and 2b2):

Gain(AP _i )=H(G)-H(G|AP _i ),

Among them, Gain(AP _i ) represents the reduction of the uncertainty of the location information in the positioning environment under the condition of known access point AP _i . The greater the information gain, the greater the reduction of the location uncertainty information, and the The point-to-position resolution is stronger.

Step 3, build a location fingerprint library.

The access point signal strength database obtained in step 1 is screened according to the fingerprint access point set, that is, only the access point data contained in the fingerprint access point set is retained, and the fingerprints of each location are obtained to form the final location fingerprint database.

Step 4, location grouping.

This step uses the classical k-means grouping method to group the positions according to the fingerprints of each position, and the process is as follows:

4a) determine the number of groups k, this example determines that the number of groups is 5 under the positioning environment shown in Figure 2, optionally 5 positions are used as the group center of these 5 groups, and the fingerprint of the position is used as the group fingerprint;

4b) Calculate the Euclidean distance from all positions to the 5 cluster centers, and assign it to the group with the smallest distance. After all positions are assigned, find the mean value of the fingerprints of all the group elements in the position cluster as the new cluster center;

4c) Repeat step 4b until the fingerprint of the cluster center no longer changes, that is, the location grouping ends.

Step 5, re-selection of the access point.

In this step, the information gain method is used to re-select the access points, and the discrimination capability of each access point in the pre-selected access point set for each group element is calculated respectively, and a set of access point sets is selected for each location grouping, and the When selecting a clustered access point set, only the access point's ability to distinguish the position within the group is considered, and the access point's ability to distinguish other grouping elements is not considered, and the optimal access point set for each group is obtained;

Re-screen the access point signal strength database obtained in step 1 through the optimal access point set for each group, and obtain new fingerprints for each location in the location group. It consists of the set of access points selected in step 2, and the set of access points selected in step 2 is optimal for the entire positioning environment;

The purpose of this step is to select the optimal access point for each location grouping, that is, the local optimum, because the optimal access point set for the entire positioning environment is not necessarily the optimal access point for each location grouping. Therefore, the new location fingerprint can better represent the attributes of each location than the location fingerprint before the access point re-selection, that is, the new location fingerprint can better represent the characteristics of each location.

Step 6, establishing a positioning decision model.

In this step, a decision tree model is established for each location grouping based on the C4.5 algorithm. The algorithm selects the access points of each node of the decision tree based on the information gain rate of the access point, and establishes a decision model with the fastest reduction of location unknown information. The process is as follows:

6a) Discretize the signal strength value of each access point, that is, divide the value range of the access point into several continuous value ranges. In this example, the value range of the access point is divided into but not limited to two sections , and calculate the corresponding information gain rate of each access point after discretization, and select the access point with the largest information gain rate as the root node;

6b) Corresponding each value range of the access point to a branch, and using the value range as the judgment condition of the branch;

6c) Repeat the above process to further determine the sub-nodes of each branch connection point, until finally the sub-nodes of each branch are all leaf nodes, that is, the establishment of the decision tree is completed.

Step 7: It is determined that the location of the sample is grouped.

According to the sample data to be located, calculate the Euclidean distance between the data and each location group. In this example, 5 Euclidean distances are obtained in the positioning environment shown in Figure 2, and the location group with the smallest Euclidean distance from the positioning sample is selected. , take this group as the group where its target position is located, and its Euclidean distance is calculated as follows:

Among them, D(T, C _j ) represents the Euclidean distance between the positioning sample T and the j-th location cluster, C _j is the cluster center of the j-th location cluster, and SS _j (T) represents the i-th access in the positioning sample The signal strength of the point, SS _j (C _j ) represents the signal strength of the ith access point of the jth location grouping.

Step 8: Determine the specific location of the positioning sample.

The positioning sample moves down from the root node along the decision tree model of the cluster where the positioning sample is located, and matches the decision condition of the decision tree branch according to the value of the characteristic access point of the positioning sample, that is, along the path that satisfies the judgment condition along the positioning sample. The branch moves down until it moves to the leaf node, which is the final position of the positioning sample;

For example, Table 1 is a positioning sample, and Figure 5 is the decision tree model of the cluster where the positioning sample is located. The positioning process of the sample using this decision tree model is as follows:

8a) The root node of the decision tree model is AP4, so first determine which branch’s judgment condition the value of AP4 in the positioning sample satisfies. Obviously, this sample satisfies the judgment condition of the rightmost branch of the AP4 node, so along the rightmost branch of AP4 Move Downward;

8b) Move to the node AP1, the value of the positioning sample AP1 is 63, and the judgment condition of the rightmost branch of the node is satisfied, then move down along the rightmost branch of AP1;

8c) When AP1 moves down to the node AP6, the value of the positioning sample AP6 is 57, which satisfies the judgment condition of the leftmost branch of the node, and then moves down along the leftmost branch of AP6;

8d) When AP6 moves to node G6, the node is a leaf node, that is, the target position of the positioning sample is G6.

Table 1

AP1 AP2 AP3 AP4 AP5 AP6 63 56 45 70 61 57

The advantages of the present invention can be further illustrated by the following simulation results:

Simulation 1, in the positioning scenario, the randomly collected positioning samples are positioned with different positioning errors. The results are shown in Figure 3. It can be seen from Figure 3 that the number of optimally selected access points is the same under different positioning error conditions. and the positioning error is within 2m, the positioning accuracy can reach 93% when the positioning error is optimal, and the positioning error can be more than 60% when the positioning error is within 0.8m, indicating that the present invention has a good positioning effect.

Simulation 2, in the positioning scene, the present invention and the positioning method based on the information gain method are used to locate the randomly collected positioning samples within a positioning error of 2m. The results are shown in Figure 4. It can be seen from Figure 4 that under the same conditions, the The positioning effect is obviously better than the positioning method based on the information gain method.

Claims (7)

1. WiFi indoor positioning method based on multiple access point selection, including: 1) Build the access point signal strength database: Divide the positioning environment into multiple grids of the same size, and use the position of the center point of the grid to indicate the position of the grid; then collect data for each position, and record the signal strength value of each access point detected at each position , forming an access point signal strength database recording the sampling data of each location; 2) Build the location fingerprint library: 2a) According to the coverage time distribution of each access point in the positioning environment, set the time initial threshold m that the access point covers at least each location, and the time initial threshold n that the access point covers at least the entire positioning space, according to the access point in the database. For the coverage time of each location and the coverage time of the access point in the entire positioning environment, perform multiple selections on the access points, delete the access points whose coverage time is less than these two thresholds, filter out the preselected access points, and form a preselected access point. set of entry points; 2b) Calculate the information gain of each access point in the preselected access point set, sort the access points in descending order of the information gain, and select the first k access points to form the final fingerprint access point set, k is greater than or equal to 10; 2c) Screen the access point signal strength database obtained in step 1) according to the fingerprint access point set, retain only the access point data contained in the fingerprint access point set, obtain the fingerprints of each location, and form the final location fingerprint database ; 3) Use the k-means algorithm to group the environmental locations, and re-select the access points for each location group, and select a feature access point set that can better represent the location group itself; 4) Use the C4.5 decision tree method to establish a decision tree model for each grouping, and obtain a decision model with the fastest reduction of location unknown information; 5) Positioning stage 5a) Given the sample data that needs to be positioned, calculate the Euclidean distance between the positioning sample fingerprint and each position cluster, select the position cluster with the smallest Euclidean distance from it, and use the cluster as the cluster where its target position is located; 5b) The positioning sample moves down from the root node along the grouped decision tree model until it moves to the leaf node of the decision tree, and the leaf node is the position of the final positioning sample. 2. The method according to claim 1, wherein in step 2b), calculating the information gain of each access point in the preselected access point set, is carried out according to the following steps: 2b1) Calculate the uncertainty H(G) of the position information in the positioning environment:

Among them, G represents the position in the positioning environment, G _i represents the ith position, P(G _i ) represents the probability of the occurrence of the position G _i , and m represents the number of positions in the positioning environment; 2b2) Calculate the uncertainty H(G|AP _i ) of the location information in the positioning environment under the condition of known access points:

Among them, AP _i represents the ith access point, v _j represents the value of the signal strength of AP _i , N represents the number of values of the signal strength of AP _i , and P(AP _i =v _j ) represents the detection at the reference position j the probability that the signal strength to the access point AP _i is v _j ; H(G|AP _i =v _j ) represents the information entropy of the location in the positioning environment under the condition that the signal strength of AP _i is known to be v _j , and its calculation formula is the same as H(G); 2b3) Calculate the information entropy Gain(AP _i ) of the access point according to the results of 2b1) and 2b2): Gain(AP _i )=H(G)-H(G|AP _i ), Among them, Gain(AP _i ) represents the reduction of the uncertainty of the location information in the positioning environment under the condition that the access point AP _i is known. The greater the information gain, the greater the reduction of the location uncertainty information. The stronger the point-to-position resolution is. 3. method according to claim 1, wherein in step 3), use k-means algorithm to carry out grouping to environmental position, carry out according to the following steps: 3a) Determine the number k of position groups, select k positions as the group centers of the k groups, and use the fingerprints of the positions as the group fingerprints, where k is greater than or equal to 2; 3b) Calculate the Euclidean distances from all positions to the k group centers, and assign them to the group with the smallest distance. After all positions are allocated, find the mean value of the fingerprints of all group elements in the position group as the new group center; 3c) Repeat step 3b) until the group center no longer changes, that is, the location grouping ends. 4. The method according to claim 1 , wherein in step 3), the re-selection of access points for each location grouping is to use an information gain method to calculate, respectively, each access point in the preselected access point set to each group. The discrimination capability of the element, select a set of access points for each location group, and when re-selecting the access point set of the group, only the discrimination ability of the access point for the location in the group is considered, and the pair of access points is not considered. The discriminating ability of other grouping elements is used to obtain the optimal access point set for each grouping, and these reselected access point sets can better represent the characteristics of each location grouping. 5. method according to claim 1, wherein in step 4), use C4.5 decision tree method to establish decision tree model for each grouping, carry out according to the following steps: 5a) Discretize the signal strength value of each access point, that is, divide the value range of the access point into several continuous value ranges, and calculate the corresponding information gain rate of each access point after discretization, Select the access point with the largest information gain rate as the root node; 5b) Corresponding each value range of the access point to a branch, and using the value range as the judgment condition of the branch; 5c) Repeat the above process to further determine the child nodes of each branch connection point until the final child nodes of each branch are leaf nodes, that is, the establishment of the decision tree is completed. 6. method according to claim 1, wherein in step 5a), calculate the Euclidean distance between positioning sample fingerprint to each position grouping, calculate according to the following formula:

where D(T,C _j ) represents the Euclidean distance between the positioning sample T and the jth location cluster, C _j is the cluster center of the jth location cluster, k represents the number of access points selected in step 2b), SS _j (T) represents the signal strength of the ith access point in the positioning sample, and SS _j (C _j ) represents the signal strength of the ith access point of the jth location grouping. 7. The method according to claim 1, wherein in step 5b), the positioning sample moves down from the root node along the decision tree model of this grouping, is to judge that it satisfies the decision tree according to the value of the feature access point of the positioning sample The decision condition of which branch determines its moving direction until it moves to a leaf node, which is the final position of the positioning sample.

Images