A method for indoor positioning based on Bluetooth beacons

Abstract

The present invention belongs to the field of positioning technology. Specifically, based on artificial intelligence technology, it adopts Bluetooth protocols and equipment to realize indoor positioning and recognition methods of different directions and gradient distances. Based on Bluetooth beacons, the present invention realizes a high-precision and universal indoor positioning model through machine learning technology, solving the problems of low indoor positioning accuracy and serious position deviation. The present invention adopts multiple classification and regression algorithms and realizes the identification of entry direction and prediction of gradient distance based on the target position. Through the self-development of indoor positioning technology, the dependence on other third-party ecological positioning technologies is reduced, and the implementation cost of the landing project is effectively reduced.

Description

Technical Field

[0001] The present invention belongs to the field of positioning technology. Specifically, based on artificial intelligence technology, the Bluetooth protocol and equipment are used to realize an indoor positioning recognition method with different orientations and gradient distances. Background Art

[0002] With the development of science and technology and the improvement of people's quality of life, the application of indoor positioning technology is increasing, and its technology is becoming more and more mature, bringing people more accurate, more matching, and more personalized services and experiences. It can provide indoor positioning capabilities for multiple industry scenarios such as smart hospitals, smart transportation, smart museums, warehousing and logistics, smart elderly care, smart parking, smart supermarkets, smart campuses, smart factories, etc., and the application prospects of the scenarios are very broad. Although the indoor positioning technology based on Bluetooth beacons has become increasingly mature, it is obviously unable to meet the needs of various complex application scenarios, such as the direction of people entering, different gradient distances, etc. At the same time, facing the complex indoor environment, problems such as high implementation cost and low positioning accuracy arise, which become problems that need to be urgently solved in indoor positioning technology. The present invention will reduce the overall indoor positioning cost by optimizing the Bluetooth beacon deployment strategy, and ensure accurate positioning in complex environments while realizing the identification of people entering the direction and gradient distance. Summary of the Invention

[0003] To overcome the above difficulties, the present invention provides a gradient positioning analysis method based on the Bluetooth protocol, which identifies the gradient distance and orientation information between random positioning points and fixed target points, and realizes multi-dimensional precise analysis such as proximity notification, route guidance, and passenger flow monitoring in indoor places such as museums, cultural centers, underground parking lots, smart supermarkets, and mines. It effectively solves the high cost problem of other positioning technologies and promotes the formation of a large-scale application ecosystem for indoor positioning.

[0004] A method for indoor positioning based on Bluetooth beacons comprises the following steps:

[0005] S1: Bluetooth Beacon Deployment

[0006] S11: Bluetooth beacon initialization

[0007] Equipment assembly and equipment parameter initialization,

[0008] S12: Bluetooth Beacon Deployment

[0009] After the Bluetooth beacon is initialized, it enters the indoor environment for field deployment. Two Bluetooth beacons are used to form a beacon group. This beacon group serves as the source of the RSSI value of the current target point. Similarly, N beacon groups are eventually deployed indoors.

[0010] S2: Model training

[0011] S21: Model training data collection

[0012] Collect model training data. The collection method is as follows:

[0013] 1) Clearly identify the target point's front, back, left, right, upper left, lower left, upper right, and lower right, a total of 8 directions;

[0014] 2) Clearly determine the distance information of the positioning point with the target point as the center;

[0015] 3) Determine the intersection of bearing and distance information;

[0016] 4) Mark the orientation and distance information of the intersection training data;

[0017] The format of collected data is as follows:

[0018] When the collection tool obtains RSSI, every 10 accumulated data are used as the value of a beacon group. Similarly, 8 groups of RSSI samples are constructed, and each sample represents a set of RSSI.

[0019] S22: Data Processing

[0020] Identify and preprocess abnormal data on the collected data, and store the processed data in a CSV file;

[0021] Step 1: Data preprocessing. The data processing steps are as follows:

[0022] 1) Missing value processing

[0023] The missing data mainly includes the missing Bluetooth beacon RSSI value, direction mark value, and gradient distance mark value. For the processing of missing values, the missing values are filled or deleted;

[0024] 2) Outlier processing

[0025] Outliers are eliminated by specifying the value range of specific attributes;

[0026] 3) Filtering

[0027] Filtering and noise reduction can be processed using the following formula. a(xi) represents the RSSI value of the i-th Bluetooth beacon. Assuming that the number of processed RSSI values is n, the number of RSSI values after filtering is n-1. Finally, the average of n-1 data is calculated, and the average is used as the RSSI value collected by the current Bluetooth beacon.

[0028]

[0029] Step 2: Data storage

[0030] The data processed by the above steps needs to be stored in the 1.csv file;

[0031] Step 3: Data Division

[0032] Randomly extract 70% of the data from the 1.csv file in the second step and store it in the 2.csv file. The remaining 30% of the data is stored in the 3.csv file. The 2.csv file is used as the model training data, and the 3.csv file is used as the model evaluation data.

[0033] S23: Model training

[0034] Model training mainly consists of the following steps: initializing the model, training and evaluating the model, and storing the model:

[0035] 1. Initialize the model:

[0036] Alink, a machine learning platform, is used as the foundation for model training and data initialization.

[0037] 2. Model training and evaluation:

[0038] 1) Orientation determination model

[0039] The orientation determination model adopts the Naive Bayes algorithm;

[0040] 2) Gradient distance model

[0041] The gradient distance model uses a decision tree regression algorithm;

[0042] 3. Model storage

[0043] After training is completed, the model is stored in a localized manner and saved to the file system.

[0044] S3: Result prediction

[0045] The RSSI value of the current UUID point is collected as the positioning point, and the algorithm is matched with the completed training model. After testing and configuring the push response time in advance, the user's detection result data set within a unit time is obtained. The point with the most matching times and no duplication with the last push can be identified as the push target.

[0046] The present invention has the following technical innovations based on the Bluetooth positioning capability:

[0047] 1. High precision: High-precision positioning can be achieved through real-time measurement of signal strength in multiple Bluetooth beacon environments and algorithm optimization.

[0048] 2. Low power consumption: Based on the adoption of Bluetooth low energy technology, the system also has the characteristics of long battery life and low energy consumption.

[0049] 3. Data security: When using cloud platforms for data storage and processing, multiple encryption technologies are used to ensure data security and prevent data leakage or theft.

[0050] 4. Real-time navigation: Through the application on the customer's mobile phone, the real-time navigation function is realized to provide customers with accurate route guidance and enhance the customer's shopping experience.

[0051] 6. Personalized recommendations: By analyzing customers' shopping data and behavior trajectories, personalized exhibit recommendations are provided to visitors, facilitating visitor reception and precise operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A collection distribution map of the Bluetooth beacon deployment of the present invention;

[0053] Figure 2 A diagram showing the deployment of a beacon group. Specific implementation plan

[0054] A method for indoor positioning based on Bluetooth beacons comprises the following steps:

[0055] S1: Bluetooth Beacon Deployment

[0056] S11: Bluetooth beacon initialization

[0057] Equipment assembly and equipment parameter initialization,

[0058] S12: Bluetooth Beacon Deployment

[0059] After the Bluetooth beacon is initialized, it enters the indoor environment for field deployment. Two Bluetooth beacons are used to form a beacon group. This beacon group serves as the source of the RSSI value of the current target point. Similarly, N beacon groups are eventually deployed indoors.

[0060] S2: Model training

[0061] S21: Model training data collection

[0062] Collect model training data. The collection method is as follows:

[0063] 1) Clearly identify the target point's front, back, left, right, upper left, lower left, upper right, and lower right, a total of 8 directions;

[0064] 2) Clearly determine the distance information of the positioning point with the target point as the center;

[0065] 3) Determine the intersection of bearing and distance information;

[0066] 4) Mark the orientation and distance information of the intersection training data;

[0067] The format of collected data is as follows:

[0068] When the collection tool obtains RSSI, every 10 accumulated data are used as the value of a beacon group. Similarly, 8 groups of RSSI samples are constructed, and each sample represents a set of RSSI.

[0069] S22: Data Processing

[0070] Identify and preprocess abnormal data on the collected data, and store the processed data in a CSV file;

[0071] Step 1: Data preprocessing. The data processing steps are as follows:

[0072] 1) Missing value processing

[0073] The missing data mainly includes the missing Bluetooth beacon RSSI value, direction mark value, and gradient distance mark value. For the processing of missing values, the missing values are filled or deleted;

[0074] 2) Outlier processing

[0075] Outliers are eliminated by specifying the value range of specific attributes;

[0076] 3) Filtering

[0077] Filtering and noise reduction can be processed using the following formula. a(xi) represents the RSSI value of the i-th Bluetooth beacon. Assuming that the number of processed RSSI values is n, the number of RSSI values after filtering is n-1. Finally, the average of n-1 data is calculated, and the average is used as the RSSI value collected by the current Bluetooth beacon.

[0078]

[0079] Step 2: Data storage

[0080] The data processed by the above steps needs to be stored in the 1.csv file;

[0081] Step 3: Data Partitioning

[0082] Randomly extract 70% of the data from the 1.csv file in the second step and store it in the 2.csv file. The remaining 30% of the data is stored in the 3.csv file. The 2.csv file is used as the model training data, and the 3.csv file is used as the model evaluation data.

[0083] S23 model training

[0084] Model training mainly consists of the following steps: initializing the model, training and evaluating the model, and storing the model:

[0085] 1. Initialize the model:

[0086] Alink, a machine learning platform, is used as the foundation for model training and data initialization.

[0087] 2. Model training and evaluation:

[0088] 1) Orientation determination model

[0089] The orientation determination model adopts the Naive Bayes algorithm;

[0090] 2) Gradient distance model

[0091] The gradient distance model uses a decision tree regression algorithm;

[0092] 3. Model storage

[0093] After training is completed, the model is stored in a localized manner and saved to the file system.

[0094] S3: Result prediction

[0095] The RSSI value of the current UUID point is collected as the positioning point, and the algorithm is matched with the completed training model. After testing and configuring the push response time in advance, the user's detection result data set within a unit time is obtained. The point with the most matching times and no duplication with the last push can be identified as the push target.

[0096] Specific use case description:

[0097] In terms of practicality, the Bluetooth positioning technology has been put into trial operation in three projects: China Telecom Museum, Beijing Shijingshan Intangible Cultural Heritage Center, and Gansu Provincial Museum, with a total of more than 100 implementation points. It mainly provides tourists with indoor real-time positioning and voice self-service push functions, with a positioning accuracy of up to 1.5 meters. Specific steps:

[0098] S1: Bluetooth Beacon Deployment

[0099] S11: Bluetooth beacon initialization

[0100] Complete the assembly of the equipment and the initialization of the equipment parameters. The specific initialization rules are as follows:

[0101]

[0102] S12: Bluetooth Beacon Deployment

[0103] After the Bluetooth beacon is initialized, it is deployed indoors. To prevent some Bluetooth beacons from malfunctioning or losing power during use, which would cause the RSSI signal value of the point to exceed the predetermined threshold value in subsequent data collection, two Bluetooth beacons are generally used to form a beacon group. This beacon group serves as the RSSI value source for the current target point. Similarly, N beacon groups are eventually deployed indoors. The descriptions of the target point, positioning point, direction, and distance are as follows: Figure 1 shown. Figure 1 In the figure, a is the center of the circle, and b, c, and d are auxiliary circles. The auxiliary circles are used to confirm the direction and distance of the positioning point.

[0104] The present invention finally predicts the target point through different orientations and gradient distances. Due to the different layouts and structures of indoor environments, in order to improve the accuracy of positioning, the following two deployment strategies are adopted: Figure 2 As shown:

[0105] Figure 2 Left: An indoor area is arranged in a triangle: beacon groups are deployed in a triangle pattern;

[0106] Figure 2 The diagram on the right shows that a certain area indoors has a right-angle layout: beacon groups are deployed at the corners.

[0107] S2: Model training

[0108] S21: Model training data collection

[0109] Use the Bluetooth data collection tool provided by the Bluetooth beacon manufacturer to collect model training data.

[0110] The collection method is as follows:

[0111] 1) Clearly define the target point's front, back, left, right, upper left, lower left, upper right, and lower right, a total of 8 directions.

[0112] 2) Clearly specify the distance information of 5m, 4m, 3m, 2m, 1.5m, 1m, and 0.5m with the target point as the center.

[0113] 3) Determine the intersection of the bearing and distance information.

[0114] 4) Use the Bluetooth data collection tool provided by the Bluetooth beacon manufacturer to collect training data at each intersection and mark the direction and distance information of the training data at that intersection.

[0115] The format of collected data is as follows:

[0116]

[0117] Assume there are 16 Bluetooth beacons, forming 8 beacon groups. When the acquisition tool obtains RSSI, every 10 accumulated data are used as the value of a beacon group. Similarly, 8 groups of RSSI samples are constructed, and each sample represents a set of RSSI.

[0118] S22: Data Processing

[0119] Abnormal data identification and preprocessing are performed on the collected data, and the processed data is stored in a CSV file.

[0120] Step 1: Data Preprocessing

[0121] The data processing steps are as follows:

[0122] 1) Missing value processing

[0123] The missing data mainly includes the missing Bluetooth beacon RSSI value, azimuth mark value, and gradient distance mark value. There may be cases where data is temporarily unavailable or some data is omitted. If left unattended, missing value data will cause chaos in the later model building process and lead to unreliable output results. For the processing of missing values, some methods are to fill 0, some are to fill with default values, or to directly delete them. For example, through analysis, it was found that the missing rate of certain key fields is less than 8%. Compared with the huge RSSI, the method of directly deleting this RSSI value is obviously the most effective. The specific processing measures are shown in the following table:

[0124]

[0125] 2) Outlier processing

[0126] Due to the complexity of the actual environment, it is inevitable that some of the characteristic parameters used for positioning will have abnormal values. Therefore, it is necessary to compare the positioning characteristic parameter value with the corresponding threshold value and eliminate bad values with large relative deviations to ensure the validity of the data as much as possible and reduce the impact on the system. The treatment of abnormal values is similar to that of missing values, except that the value range of the specific attribute needs to be defined in advance to determine whether the sampled value is abnormal. The treatment method for the abnormal value is to fill it with zero, fill it with the default value, or discard the fingerprint. The specific treatment measures are shown in the following table:

[0127]

[0128] 3) Filtering

[0129] Filtering and noise reduction can be processed using formula 1. a(xi) represents the RSSI value of the i-th Bluetooth beacon. Assuming that the number of processed RSSI values is n, the number of RSSI values after filtering is n-1. Finally, the average of n-1 data is calculated, and the average is used as the RSSI value collected by the current Bluetooth beacon.

[0130]

[0131] Formula 1

[0132] Step 2: Data storage

[0133] The data processed by the above steps needs to be stored in the 1.csv file. The stored data sample is as follows:

[0134]

[0135] Step 3: Data Partitioning

[0136] Randomly extract 70% of the data from the 1.csv file in the second step and store it in the 2.csv file, and store the remaining 30% of the data in the 3.csv file. The 2.csv file is used as the model training data, and the 3.csv file is used as the model evaluation data.

[0137] S23 model training

[0138] Model training mainly consists of the following steps: initializing the model, training and evaluating the model, and storing the model:

[0139] 1. Initialize the model:

[0140] The training model is an important tool for overall prediction. To better assist in building the training model, a series of parameter data must be set at the initial stage of model creation. The machine learning platform Alink is used as the foundation for model training. The specific initialization parameters of the model to be trained are shown in the following table:

[0141]

[0142] 2. Model training and evaluation:

[0143] The overall model training is based on the machine learning capability platform to complete the training of various models. The present invention focuses on solving the problem of predicting the direction of personnel entry and different gradient distances based on reference points. Therefore, the trained models are divided into direction determination models and gradient distance models.

[0144] 1) Orientation determination model

[0145] A variety of multi-classification algorithms were used to train the orientation determination model. The Naive Bayes algorithm was ultimately chosen. This algorithm has stable classification efficiency, can handle multi-classification tasks, is suitable for incremental training, is not very sensitive to missing data, and the evaluation indicators of other multi-classification algorithms are not as good as this algorithm. The indicators are shown in the following table:

[0146]

[0147] As can be seen from the above table, the naive Bayes method has the highest indicators and the shortest time. The other classification algorithms are far behind it. Therefore, using naive Bayes classification can achieve better classification results.

[0148] 2) Gradient distance model

[0149] We used multiple regression algorithms to train the gradient distance model, and ultimately adopted the decision tree regression algorithm. This algorithm is fast, highly accurate, can handle continuous and categorical fields, and is suitable for high-dimensional data. The evaluation indicators of other regression algorithms are not as good as this algorithm. The indicators are shown in the following table:

[0150]

[0151] As can be seen from the table above, decision tree regression significantly outperforms other regression algorithms. While other regression algorithms improve various metrics, the improvements are small and do not yield better results. Therefore, in practical applications, we need to consider both regression performance and computational cost when selecting an appropriate regression algorithm.

[0152] 3. Model storage

[0153] After training is completed, the model is stored in a localized manner and saved to the file system.

[0154] S3: Result prediction

[0155] Based on the orientation determination and gradient distance training models, the RSSI value and UUID of the current location are collected as the positioning point and algorithmically matched with the completed training model. After pre-testing and configuring the push response time, the user's detection result data set within a unit time (usually 5 seconds) is collected. The point with the most matching times and no duplication with the previous push is identified as the push target.

[0156] 1. Based on Bluetooth beacons, the present invention realizes a high-precision and universal indoor positioning model through machine learning technology, solving the problems of low indoor positioning accuracy and serious position deviation.

[0157] 2. The present invention adopts multiple classification and regression algorithms and realizes the recognition of entry direction and prediction of gradient distance based on target position.

[0158] 3. Through self-developed indoor positioning technology, we reduce our reliance on other third-party ecological positioning technologies and effectively reduce the implementation costs of projects.

Claims

1. A method for indoor positioning based on Bluetooth beacons, characterized in that The process includes the following steps: S1: Bluetooth Beacon Deployment S11: Bluetooth beacon initialization Equipment assembly and equipment parameter data initialization; S12: Bluetooth Beacon Deployment After the Bluetooth beacon is initialized, it enters the indoor environment for field deployment. Two Bluetooth beacons are used to form a beacon group. This beacon group serves as the source of the RSSI value of the current target point. Similarly, N beacon groups are eventually deployed indoors. S2: Model training S21: Model training data collection Collect model training data. The collection method is as follows: Clearly define the target point's front, back, left, right, upper left, lower left, upper right, and lower right, a total of 8 position information; Clearly determine the distance information of the positioning point with the target point as the center; Determine the intersection of bearing and distance information; Mark the orientation and distance information of the intersection training data; The format of collected data is as follows: When the collection tool obtains RSSI, every 10 accumulated data are used as the value of a beacon group. Similarly, 8 groups of RSSI samples are constructed, and each sample represents a set of RSSI. S22: Data Processing Identify and preprocess abnormal data on the collected data, and store the processed data in a CSV file; Step 1: Data preprocessing. The data processing steps are as follows: 1) Missing value processing Missing data includes missing Bluetooth beacon RSSI values, direction mark values, and gradient distance mark values. For the processing of missing values, missing values are filled or deleted; 2) Outlier processing Outliers are eliminated by specifying the value range of specific attributes; 3) Filtering Filtering and noise reduction can be processed using the following formula. a(xi) represents the RSSI value of the i-th Bluetooth beacon. Assuming that the number of processed RSSI values is n, the number of RSSI values after filtering is n-1. Finally, the average of n-1 data is calculated, and the average is used as the RSSI value collected by the current Bluetooth beacon. Step 2: Data storage The data processed through the above steps needs to be stored in the 1.csv file.

2. The method for indoor positioning based on Bluetooth beacons according to claim 1, characterized in that Also includes the following steps: S22 data processing also includes the third step: data division Randomly extract 70% of the data from the 1.csv file in the second step and store it in the 2.csv file. The remaining 30% of the data is stored in the 3.csv file. The 2.csv file is used as the model training data, and the 3.csv file is used as the model evaluation data. S23: Model training Model training mainly consists of the following steps: initializing the model, training and evaluating the model, and storing the model:

1. Initialize the model: Alink, a machine learning platform, is used as the foundation for model training and data initialization.

2. Model training and evaluation: 1) Orientation determination model The orientation determination model adopts the Naive Bayes algorithm; 2) Gradient distance model The gradient distance model uses a decision tree regression algorithm; 3. Model storage: After training is completed, the model is stored in a localized manner and saved to the file system; S3: Result prediction The RSSI value of the current UUID point is collected as the positioning point, and the algorithm is matched with the completed training model. After testing and configuring the push response time in advance, the user's detection result data set within a unit time is obtained. The point with the most matching times and no duplication with the last push can be identified as the push target.

Images