An indoor visible light positioning method based on improved condensed hierarchical clustering

Abstract

The application relates to an indoor visible light positioning method based on improved condensed hierarchical clustering, which comprises the following steps: dividing a grid in a positioning area and selecting a plurality of reference points; installing LED light sources capable of being used for optical communication on a roof and laying photoelectric sensors and wireless nodes for data acquisition and transmission indoors; measuring the average power of the LED light sources and establishing a fingerprint library; adopting condensed hierarchical clustering to combine all data points into one cluster; selecting a required cluster number to construct a new sub-database; performing secondary classification on the new cluster according to the power of the light sources, and dividing the new cluster into a high-power area and a low-power area; in the positioning stage, comparing the Euclidean distance between the light power of a mobile terminal and a new clustering center, and putting the mobile terminal into the cluster closest to the mobile terminal; putting the power of the mobile terminal into the corresponding power area, and adopting a KNN matching algorithm to traverse the fingerprints to lock the target coordinate position.

Description

Technical Field

[0001] This invention relates to an indoor visible light positioning method based on improved agglomerative hierarchical clustering, belonging to the fields of artificial intelligence positioning and indoor visible light positioning. Background Technology

[0002] In recent years, with the development of technology, indoor positioning has become a widely studied topic among scholars. The development of smart cities has further promoted social attention to indoor positioning technology. Currently, common indoor positioning technologies include Wi-Fi, ultrasound, Zigbee, RFID, infrared sensors, and visible light positioning. Compared with other technologies, visible light positioning technology has advantages such as high security, environmental friendliness, low cost, fast operation speed, flexible bandwidth, no electromagnetic interference, and high positioning accuracy. Visible light sources are readily available in daily life. With the popularization of LED lights, a large number of indoor users can obtain positioning services and data transmission through Wi-Fi. Therefore, visible light positioning technology is well-suited for applications in hospitals, smart homes, factories, mines, and other scenarios. The agglomerative hierarchical clustering visible light positioning method adopted in this invention is a deep integration of visible light communication and artificial intelligence, which is very much in line with the current needs of indoor positioning. Summary of the Invention

[0003] The purpose of this invention is to propose an indoor visible light localization method based on improved agglomerative hierarchical clustering. Agglomerative hierarchical clustering is not only an unsupervised machine learning method, but also an important manifestation of machine learning applied to artificial intelligence. The principle of agglomerative hierarchical clustering is as follows: Figure 2 As shown in the figure, this positioning method uses agglomerative hierarchical clustering for classification when establishing a traditional fingerprint database, and then performs secondary classification based on power levels, effectively improving the efficiency of indoor positioning.

[0004] This invention proposes an indoor visible light positioning method based on improved agglomerative hierarchical clustering. The method is used to measure a specific positioning area and includes the following steps:

[0005] (1): Divide a certain positioning area into N grids evenly, and select the center of the grid to form N reference points;

[0006] (2): Install M LED light sources for optical communication on the roof of the positioning area, and lay photoelectric sensors and wireless nodes for data acquisition and transmission in the indoor area of ​​the positioning area.

[0007] (3): At each reference point, the optical signal power emitted from M LED light sources is continuously measured, and the average power P is obtained by forming a vector from the M optical signal power values. i , represented as in: This represents the power of the optical signal emitted by the Mth LED at the i-th reference point;

[0008] (4): Step (3) is repeated V times, and the average value is used to represent the power of the light signal emitted by each LED light source received by the photodetector; the average power P is then used to calculate the average power. i and the coordinates of the i-th reference point I i =(x i ,y i The fingerprint is stored in the fingerprint database; this establishes a traditional fingerprint database P, where P = [[P1, I1]]. T [P2,I2] T ,[P i ,I i ] T ,...,[P N ,I N ] T ]; where: P i I represents the average light intensity power at the i-th coordinate point. i This represents the position coordinates of the i-th point;

[0009] (5): After establishing a traditional fingerprint database, agglomerative hierarchical clustering is adopted, and each fingerprint in the fingerprint database is regarded as a single cluster. Single linkage is selected as the distance metric between two clusters.

[0010] (6): Perform several iterations, merge the two clusters with the smallest distance into one cluster, repeat the process until all data points are merged into one cluster;

[0011] (7): Select the required number of clusters K, treat each cluster as a localization sub-region, and establish a new cluster; take the average value of all fingerprints in the new cluster as the cluster center of the new cluster; each new cluster and the corresponding location fingerprint constitute a new sub-fingerprint database;

[0012] (8): The new clusters are classified a second time based on the optical signal power, dividing the sub-fingerprint database into two regions: high power and low power; the highest power received in the new sub-fingerprint database is P. max The minimum power is P min If the received optical signal power is less than The fingerprint is then included in the low-power region if the received optical power is greater than [a certain value]. The fingerprint is then included in the high-power region;

[0013] (9): During the positioning phase, the optical power received by the mobile terminal is compared with the Euclidean distance of the K new cluster centers, and the mobile terminal is included in the cluster with the closest distance.

[0014] (10): Then combine the mobile terminal optical power with Compare and include them in the corresponding power range;

[0015] (11): In the corresponding power region, the KNN matching algorithm is used to traverse the fingerprint to find the target coordinate position.

[0016] The beneficial effects of this invention are as follows: This invention optimizes the indoor positioning method of visible light communication fingerprint database, and effectively reduces the number of traversals in the positioning process by improving the agglomerative hierarchical clustering method, thereby improving the positioning speed. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of an indoor positioning scenario according to the present invention.

[0018] Figure 2 This is a flowchart illustrating the process of the hierarchical clustering algorithm.

[0019] Figure 3 This is an indoor visible light positioning method based on improved agglomerative hierarchical clustering. Detailed Implementation

[0020] The specific implementation of the present invention will now be described with reference to the accompanying drawings.

[0021] Example 1: Figure 1 An application example based on this invention is given. For example... Figure 1 As shown, LED light sources for optical communication are installed on the ceiling of the indoor area, and photodetectors for detecting light intensity signals are installed on the ground mobile terminal. The average power received at each reference point is collected, and a traditional fingerprint database is established. A new sub-database is established using agglomerative hierarchical clustering and secondary classification methods. During the localization phase, rapid positioning can be achieved to find the target coordinates. The principle of agglomerative hierarchical clustering is as follows: Figure 2 As shown.

[0022] Figure 3 The flowchart shown is for an indoor visible light positioning method based on improved agglomerative hierarchical clustering. The specific steps are as follows:

[0023] Step 1: Divide the positioning area into N grids evenly, and select the center of each grid to form N reference points.

[0024] Step 2: Install M LED light sources on the roof for optical communication, and lay photoelectric sensors and wireless nodes for data acquisition and transmission indoors.

[0025] Step 3: Continuously measure the optical signal power emitted from M LEDs at each reference point. The vector composed of the M optical signal power values ​​can be represented as... in This represents the power of the optical signal emitted by the Mth LED at the i-th reference point.

[0026] Step 4: Measure V times as described in Step 3 and calculate the average value to represent the power of the light signal received by the photodetector from each LED. The average power P is then used to calculate the average power. i and the coordinates of the i-th reference point I i =(x i ,y i Store the fingerprint in the fingerprint database. This will create a fingerprint database P = [[P1, I1]]. T [P2,I2] T ,[P i ,I i ] T ,...,[P N ,I N ] T ]; where: P i I represents the average light intensity power at the i-th coordinate point. i This represents the position coordinates of the i-th point.

[0027] Step 5: After establishing the traditional fingerprint database, agglomerative hierarchical clustering is adopted, treating each fingerprint location in the fingerprint database as a single cluster, and single linkage is selected as the distance metric between two clusters.

[0028] Step 6: Perform several iterations, merging the two clusters with the smallest distance into one cluster, repeating this process until all data points are merged into one cluster.

[0029] Step 7: Select the desired number of clusters K, treating each cluster as a localization sub-region. Use the average of all fingerprints in the new cluster as the cluster center. Each cluster and its corresponding location fingerprint constitute a new sub-fingerprint database.

[0030] Step 8: The new clusters are further classified based on the light source power, dividing the sub-fingerprint database into high-power and low-power regions. The highest power received in the new sub-fingerprint database is P. max The minimum power is P min If the received optical power is less than The fingerprint is then included in the low-power region if the received optical power is greater than [a certain value]. The fingerprint is then included in the high-power region.

[0031] Step 9: In the localization phase, compare the optical power received by the mobile device with the Euclidean distance of the K new cluster centers, and assign the mobile device to the cluster with the closest distance.

[0032] Step 10: Then compare the mobile terminal optical power with... Compare and include them in the corresponding power range.

[0033] Step 11: In the corresponding power region, the KNN matching algorithm is used to traverse the fingerprint to find the target coordinate position.

Claims

1. An indoor visible light positioning method based on improved agglomerative hierarchical clustering, characterized by The method is used for measuring a certain positioning area, comprising the following steps: (1): evenly divide a certain positioning area into N grids, and select the center of the grid to form N reference points; (2): install M LED light sources used for optical communication on the roof of the positioning area, lay photoelectric sensors indoors, and arrange wireless nodes for data collection and transmission; (3): continuously measuring the optical signal power from M LED light sources in each reference point, forming a vector with M optical signal power values, and obtaining the power mean P i , expressed as wherein: represents the optical signal power from the Mth LED at the ith reference point; (4): Step (3) is repeated V times, and the average value is used to represent the optical detector received each LED light source emitted light signal power; the power average P i and the i-th reference point coordinates I i = (x i ,y i ) is stored in the fingerprint library; that is, a traditional fingerprint library P, P = [[P1,I1] T ,[P2,I2] T ,[P i ,I i ] T ,...,[P N ,I N ] T ] is established; wherein: P i represents the light intensity power average of the i-th coordinate point, I i represents the position coordinates of the i-th coordinate point; (5): after establishing a traditional fingerprint database, adopt condensed hierarchical clustering, take each position fingerprint in the fingerprint database as a single cluster, and select single linkage as the distance measurement standard between two clusters; (6): perform several iterations, merge two clusters with the smallest distance into one cluster, and repeat the process until all data points are merged into one cluster; (7): select the required cluster number K, regard each cluster as a positioning sub-area, and establish a new cluster; take the average value of all fingerprints in the new cluster as the cluster center of the new cluster; each new cluster and the corresponding position fingerprint constitute a new sub-fingerprint database; (8): According to the power of the optical signal, the new cluster is classified twice, and the sub-fingerprint database is divided into two areas of high power and low power; the highest power received in the new sub-fingerprint database is P max , and the lowest power is P min ; if the received optical signal power is less than , the fingerprint is included in the low power area, and if the received optical power is greater than , the fingerprint is included in the high power area; (9): in the positioning stage, compare the optical power received by the mobile terminal with the Euclidean distance of the K new cluster centers, and include the mobile terminal into the cluster closest to the mobile terminal; (10): the mobile terminal optical power is compared with the corresponding power region is included. (11): in the corresponding power area, adopt a KNN matching algorithm to traverse the fingerprints and find the target coordinate position.

Images