A robot indoor positioning method based on WiFi fingerprint

Abstract

The present invention relates to the field of robot technology, and in particular to a robot indoor positioning method based on WiFi fingerprint, which divides the operation area into grids, establishes a WiFi fingerprint library through pre-measurement, randomly selects a small part of the grid, measures the WiFi signal strength of the point, and establishes a calibration data set based on the environment, performs preliminary outlier processing, and establishes a regression calibration model using a multi-layer DNN based on the established calibration data set, establishes multiple features based on each grid of the positioning data set, establishes a robot positioning model, and the robot receives the signal strength of WiFi transmission at each position, generates features, and finally puts the generated features into the robot positioning model to obtain positioning results. The method combines grid division with data denoising to reduce the influence of RSSI fluctuations on the model, thereby realizing the indoor positioning of the robot relying solely on the WiFi signal itself, effectively improving the positioning accuracy without excessive cost.

Description

Technical Field

[0001] The present invention relates to the technical field of robots, and in particular to a robot indoor positioning method based on WiFi fingerprint. Background Art

[0002] Autonomous indoor navigation is the core function of robots, and accurate positioning technology is the key. Due to the complexity of the indoor environment and related limitations, accurate positioning of the robot is an important prerequisite for robot application, so robot indoor positioning is one of the hot topics in robot research. Currently, there are many ways to locate robots indoors based on wireless signals, typical methods include TOA and TDOA. Among them, the method based on WiFi signal fingerprint is one of the more popular positioning methods because of its simplicity.

[0003] However, the current WiFi fingerprint positioning method has the problem of low positioning accuracy due to the fluctuation of RSSI. Other fusion positioning methods that combine ultrasonic and geomagnetic information require more sensors, which increases the cost of the robot. Summary of the invention

[0004] The purpose of the present invention is to provide a robot indoor positioning method based on WiFi fingerprint, which solves the problem that the current WiFi fingerprint positioning method has low positioning accuracy due to RSSI fluctuations. Other fusion positioning methods combining ultrasonic and geomagnetic information require more sensors, which increases the cost of the robot.

[0005] To achieve the above object, the present invention provides a robot indoor positioning method based on WiFi fingerprint, comprising the following steps:

[0006] Divide the operation area into M*N grids and establish a WiFi fingerprint library through pre-measurement;

[0007] Using professional channel measurement equipment, randomly select a small number of grids, measure the WiFi signal strength of the small number of grids, and establish an environment-based calibration data set;

[0008] The method of calculating the ε neighborhood is used to perform preliminary outlier processing;

[0009] Based on the established calibration data set, a regression calibration model is established using a multi-layer DNN;

[0010] Based on each grid of the positioning data set, multiple features are established;

[0011] Use the training feature dataset to build a robot positioning model;

[0012] The robot receives the signal strength transmitted by WiFi at each location, removes outliers, calibrates the data, and generates features. Finally, the generated features are put into the robot positioning model to obtain the grid number of the location, thereby obtaining the positioning result.

[0013] The operation area is divided into M*N grids, and a WiFi fingerprint library is established through pre-measurement. The steps also include:

[0014] Divide the plane of the operating area into M*N grids, and each grid has the same size;

[0015] Use a simple measuring device to measure the WiFi signal strength at each grid and record it as a WiFi fingerprint;

[0016] Perform multiple measurements in each grid, save the received measurement data, and build a WiFi fingerprint library.

[0017] Among them, using professional channel measurement equipment, randomly selecting a small part of the grid, measuring the WiFi signal strength of the point, and establishing an environment-based calibration data set, the steps also include:

[0018] The selected grid points are measured jointly using a channel meter and a simple measurement device to obtain two signal strengths at the same time, one is the RSSI collected by the simple measurement device r , one is the RSSI collected by the channel meter p , the RSSI r As a feature, RSSI p As labels, they are combined into a calibration dataset.

[0019] Wherein, the method of calculating the ε neighborhood is used to perform preliminary outlier processing, and the steps further include:

[0020] Randomly select a measurement point in the WiFi fingerprint library and calculate the number of neighboring points within the range of ε. If the number of its neighboring points is less than the threshold T, the point is marked as a noise point. Repeat this process until all noise points are found.

[0021] According to the established calibration data set, a regression calibration model is established using a multi-layer DNN, and the step further includes:

[0022] The WiFi fingerprint library data with outliers removed is put into the calibration model for calibration, and the calibrated WiFi fingerprint library is used as the positioning data set.

[0023] Wherein, based on each grid of the positioning data set, multiple features are established, and the steps further include:

[0024] The various features include distance features, digital features, and radio frequency image features. The features and grid ids are combined to form a training feature dataset.

[0025] The invention discloses a robot indoor positioning method based on WiFi fingerprint, step 1: divide the operation area into M*N grids, establish a WiFi fingerprint library by pre-measurement. Step 2: use professional channel measurement equipment to randomly select a small part of the grid, measure the WiFi signal strength of the point, and establish an environment-based calibration data set. Step 3: use the method of calculating the ε neighborhood to perform preliminary outlier processing. Select the data of a measurement point in the WiFi fingerprint library, calculate the number of neighboring points within the ε range, and if the number of its neighboring points is less than the threshold T, the point is marked as a noise point. Repeat this process until all noise points are found. ε and T are manually adjusted according to experimental results. Step 4: according to the calibration data set established in step 2, a regression calibration model is established using a multi-layer DNN. The WiFi fingerprint library data with outliers removed is placed in the calibration model for calibration processing. The calibrated WiFi fingerprint library is used as a positioning data set. Step 5: Based on each grid of the positioning data set, multiple features are established. These features include 1) distance features: the estimated distance from the grid to the transmitter. 2) digital features: the mean, maximum, minimum, median, and quartile of the RSSI corresponding to each SSID group. 3) RF image features. These features are combined with the grid id to form a training feature data set. Step 6: Use the training feature data set to build a robot positioning model. Step 7: When the robot receives the signal strength transmitted by WiFi at each location, it will perform outlier removal, data calibration, and feature generation according to steps 2, 3, 4, and 5 respectively. Finally, these features are put into the robot positioning model to obtain the grid number of the location, thereby obtaining the positioning result. The method of combining grid division with data denoising can reduce the impact of RSSI fluctuations on the model, thereby realizing the indoor positioning of the robot relying solely on the WiFi signal itself, effectively improving the positioning accuracy without excessive cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0027] Figure 1 It is a flowchart of the steps of the robot indoor positioning method based on WiFi fingerprint according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0028] Embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0029] The first embodiment of the present application is:

[0030] See also Figure 1 ,in, Figure 1 It is a flowchart of the steps of the robot indoor positioning method based on WiFi fingerprint according to the first embodiment of the present invention.

[0031] The present invention provides a robot indoor positioning method based on WiFi fingerprint, comprising the following steps:

[0032] S101: Divide the operation area into M*N grids, and establish a WiFi fingerprint library through pre-measurement;

[0033] Specifically, the plane of the operating area is divided into M*N grids, and each grid is the same size. The grid size of the edge of the area may be smaller due to the occlusion of special-shaped objects. M and N can be adjusted according to the size of the area and the positioning accuracy. If higher accuracy is required, M and N can be set larger, that is, more grids can be set. Use a mobile phone or other simple measurement equipment, handheld or portable, to measure at the grid and record the strength (RSSI) of the received WiFi signal. Measure the RSSI signal strength emitted by the WiFi router in each grid and record it as a four-tuple {Time Stamp, GID, SSID, RSSI}, which is called WiFi fingerprint. Since there are multiple WiFi routers in the environment, multiple RSSI fingerprints may be received in one measurement. Perform K measurements in each grid, save the received measurement data, and establish a WiFi fingerprint library. The more measurements are made, the more grids are covered, and the higher the positioning accuracy will be.

[0034] S102: Use professional channel measurement equipment to randomly select a small part of the grid, measure the WiFi signal strength of the point, and establish an environment-based calibration data set;

[0035] Specifically, a professional channel meter and the simple measuring device used in step S101 are fixed on a live broadcast bracket, and the distance between the two is no more than 10 cm. The bracket is carried by humans to perform measurements at some grid points. The selection of measurement grid points must cover the main regional characteristics. With the operating area as the center, the operating area is divided into 4 quadrants, and each quadrant is further divided into 4 small areas, resulting in a total of 16 small areas. Then, the grid points in each small area are sampled, and 20% of the grid points are extracted as measurement grid points. If the operating area is large, each of the 16 small areas can be further divided into 4 smaller areas. This can be adjusted according to actual conditions. The extracted measurement grid points are jointly measured using a channel meter and a simple measuring device. Thereby, two signal strengths are obtained at the same time, one is the RSSI collected by the simple measuring device. r , one is the RSSI collected by the channel analyzer p . Set RSSI r As a feature, RSSI p As labels, they are combined into a calibration dataset.

[0036] S103: Perform preliminary outlier processing by using a method for calculating the ε neighborhood;

[0037] Specifically, define an ε field:

[0038]

[0039] p, q represent the RSSI values ​​of the two measurement points, and distance represents the distance function. The distance function distance is calculated using the following formula. ε is a manually set value that can be adjusted according to actual conditions, for example, ε=0.5 can be set.

[0040]

[0041] For the data of any measurement point in the WiFi fingerprint library, calculate the number of neighboring points within the ε range. If the number is less than the threshold T, the point is marked as an outlier and can be removed from the WiFi fingerprint library. The T value can be adjusted according to the specific situation, for example, T=2 can be set. Repeat the above operation to check the outliers of all points in the fingerprint library until all outliers are found.

[0042] S104: Establishing a regression calibration model using a multi-layer DNN according to the established calibration data set;

[0043] Specifically, according to the calibration data set established in step S102, the RSSI collected by the simple measurement device is used as a feature, and the RSSI collected by the channel analysis is used as a label to establish a supervised regression learning problem, and a multi-layer DNN is used to establish an RSSI calibration model. All data of the WiFi fingerprint library with outliers removed are placed in the RSSI calibration model for calibration. The calibrated WiFi fingerprint library is used as the WiFi positioning data set.

[0044] S105: establishing multiple features based on each grid of the positioning data set;

[0045] Specifically, the multiple features include distance features, digital features, and radio frequency image features. Distance features: Using the path loss model, the height of the WiFi transmitter, the height of the robot, and the transmission power of the WiFi, the Cartesian distance from each transmitter to the robot is calculated as the estimated distance from the grid to the transmitter. Digital features: All RSSI fingerprint data received from a single scan of each grid are grouped by SSID. The mean, maximum, minimum, median, and quartile of the RSSI corresponding to each SSID group are calculated, and the distance is estimated to establish the digital features of the location. Radio frequency image features: The RSSI data corresponding to the fingerprint matrix obtained by a single scan of each grid is converted into a W*W radio frequency image, where W is the size of the image, that is, the length and width. W can be set according to the data sampling situation, and a larger value can be set if more data is collected. For example, W=10 is set. Then, the CNN neural network is used to extract features from the image, and finally a K*1-dimensional radio frequency image feature is formed. This K can be set with reference to the value of W, for example, K=10 is set. A supervised learning data set is established. Set the number of each grid as the Label, and then combine the distance feature, digital feature and radio frequency image feature to establish a supervised training feature dataset, where grid is the grid and grid id is the grid ID.

[0046] S106: Using the training feature data set, establish a robot positioning model;

[0047] Specifically, the training feature data set established in step S105 is divided into 80% training set and 20% test set. XGBoost is used to establish a machine learning classification model, and training and verification are performed on the training feature data set to generate a positioning model.

[0048] S107: The robot receives the signal strength transmitted by the WiFi at each location, performs outlier removal, data calibration, and feature generation, and finally puts the generated features into the robot positioning model to obtain the grid number of the location, thereby obtaining the positioning result.

[0049] Specifically, during the operation of the robot, the RSSI signal emitted by the WiFi router is collected at the point to be positioned. Outliers are removed, data is calibrated, and features are generated according to steps S102, S103, S104, and S105. Finally, these features are put into the robot positioning model, and the model predicts the corresponding grid number to obtain the positioning result.

[0050] Combining grid division with data noise reduction methods, the impact of RSSI fluctuations on the model is reduced, so that the robot's indoor positioning can be achieved relying solely on the WiFi signal itself, effectively improving the positioning accuracy without excessive cost.

[0051] What is disclosed above is only one or more preferred embodiments of the present application, and cannot be used to limit the scope of rights of the present application. Ordinary technicians in this field can understand that all or part of the processes of implementing the above embodiments and equivalent changes made according to the claims of the present application are still within the scope covered by the present application.

Claims

1. A robot indoor positioning method based on WiFi fingerprint, characterized in that: The following steps are involved: Divide the operation area into M*N grids and establish a WiFi fingerprint library through pre-measurement; Use professional channel measurement equipment to randomly select a small number of grids, measure the WiFi signal strength of the small number of grids, and establish an environment-based calibration data set; use a channel meter and a simple measurement device to jointly measure the selected grid points, and obtain two signal strengths at the same time, one is the RSSI collected by the simple measurement device r , one is the RSSI collected by the channel meter p , the RSSI r As a feature, RSSI p As labels, combined into a calibration dataset; The method of calculating the ε neighborhood is used to perform preliminary outlier processing; the data of a measurement point is randomly selected in the WiFi fingerprint library, and the number of neighboring points within the ε range is calculated. If the number of its neighboring points is less than the threshold T, the point is marked as a noise point, and this process is repeated until all noise points are found; Based on the established calibration data set, a regression calibration model is established using a multi-layer DNN; Based on each grid of the positioning data set, multiple features are established; Use the training feature dataset to build a robot positioning model; The robot receives the signal strength transmitted by WiFi at each location, removes outliers, calibrates the data, and generates features. Finally, the generated features are put into the robot positioning model to obtain the grid number of the location, thereby obtaining the positioning result.

2. The robot indoor positioning method based on WiFi fingerprint as claimed in claim 1, characterized in that: The operation area is divided into M*N grids, and a WiFi fingerprint library is established through pre-measurement. The steps further include: Divide the plane of the operating area into M*N grids, and each grid has the same size; Use a simple measuring device to measure the WiFi signal strength at each grid and record it as a WiFi fingerprint; Perform multiple measurements in each grid, save the received measurement data, and build a WiFi fingerprint library.

3. The robot indoor positioning method based on WiFi fingerprint as claimed in claim 1, characterized in that: According to the established calibration data set, a regression calibration model is established using a multi-layer DNN, and the step further includes: The WiFi fingerprint library data with outliers removed is put into the calibration model for calibration, and the calibrated WiFi fingerprint library is used as the positioning data set.

4. The robot indoor positioning method based on WiFi fingerprint as claimed in claim 3, characterized in that: Based on each grid of the positioning data set, multiple features are established, and the steps further include: The multiple features include distance features, digital features, and radio frequency image features. The features and grid id are combined to form a training feature data set, where grid is a grid and grid id is a grid ID.

Images