A method and apparatus for fingerprint augmentation for positioning

Abstract

The embodiment of the present application provides a fingerprint expansion method and device for positioning, and relates to the field of indoor fingerprint positioning. The method comprises the following steps: collecting CSI fingerprint signals of each reference point in the environment under initial space conditions, then collecting a small amount of CSI signals after space conditions are changed, converting the CSI signals containing multiple links into a fingerprint feature map, and establishing a fingerprint database; encoding the CSI feature map of the initial space conditions through a position feature encoder to obtain a position feature code; sampling the Gaussian distribution multiple times to obtain multiple space feature codes after the space conditions are changed; and combining the position feature code and different space feature codes to send into a generator to obtain the fingerprint feature map after the space conditions are changed. The scheme provided by the embodiment of the present application can separate and recombine the position feature and the space feature of the fingerprint data, realize the expansion of the fingerprint data after the change of the environmental space conditions, and reduce the data collection cost.

Description

Technical Field

[0001] This invention relates to the field of indoor fingerprint positioning, and in particular to a fingerprint augmentation method and apparatus for positioning. Background Technology

[0002] With the advent of the information and intelligent era, accurate indoor location information has become a necessary requirement for mobile intelligent sensing and IoT device applications. However, most indoor environments are complex and highly time-varying, posing a significant challenge to high-precision indoor positioning. Positioning methods such as TOA and AOA are susceptible to multipath and non-line-of-sight effects in complex environments, while fingerprint positioning, with its advantages of high accuracy, strong multipath resistance, and ease of deployment, has been widely used in indoor positioning. In fingerprint positioning systems, Received Signal Strength (RSS) or Channel State Information (CSI) is typically used as the fingerprint, which is then further processed to estimate the location of the target. Compared to the coarse-grained RSS information at the physical layer, CSI can represent the subcarrier channel state of the transmitter antenna pair, such as signal scattering, multipath fading, and shadow fading, providing higher sensing capabilities.

[0003] Data-driven fingerprint localization requires the pre-establishment of a fingerprint database, with samples corresponding only to the indoor environment at the sampling time. Typically, an initial fingerprint database is collected under only one spatial condition within the environment. However, indoor environments are complex and variable; the wireless signal characteristics of fingerprint data change with movement of people and the influence of obstacles such as walls and furniture. This leads to significant distributional differences between offline and online fingerprint databases in practical applications, posing a serious challenge to fingerprint localization. Re-collecting fingerprint databases in the changed environment is both labor-intensive and time-consuming, and fails to make efficient use of the initial database. Furthermore, due to time and spatial limitations, the amount of data collected at different fingerprint points and in different environments may be uneven, degrading the overall quality of the fingerprint database. With the rise of Generative Adversarial Networks (GANs) in deep learning and their excellent data generation capabilities, some studies have applied them to fingerprint data generation with some success. However, current GAN-based fingerprint data augmentation methods are all focused on augmenting fingerprint samples in static environments, neglecting data augmentation after environmental changes. These methods primarily predict or generate fingerprints of reference points under the same fingerprint data distribution and require a large number of reference point samples in the environment to be augmented for network training. However, environmental changes can be considered as altering the spatial information contained in fingerprint data. To expand fingerprint data across data distributions after environmental changes, the spatial and location information in the fingerprint data should be separated. The initial fingerprint database should then be used to accommodate the data distribution after environmental changes while preserving the location information. Therefore, this invention expands fingerprint data after environmental changes by separating and recombining spatial and location information, while ensuring the location information remains unchanged.

[0004] The information disclosed in this background section is intended only to enhance the understanding of the general background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a fingerprint augmentation method and apparatus for positioning, thereby solving the technical problems existing in the prior art. To achieve the above objective, this invention adopts the following technical solution:

[0006] In a first aspect, the present invention provides a fingerprint augmentation method for positioning, comprising the following steps:

[0007] S1: Collect CSI fingerprint signals of each reference point in the environment under the initial spatial conditions, and then collect a small number of CSI signals after spatial condition transformation. Convert the CSI signals containing multiple links into fingerprint feature maps.

[0008] S2: The CSI feature map of the initial spatial conditions is encoded by the position feature encoder to obtain the position feature code.

[0009] S3: Multiple samplings of the Gaussian distribution yield multiple spatial feature codes after spatial condition changes.

[0010] S4: The location feature code and different spatial feature codes are combined and fed into the generator to obtain the fingerprint feature map after the spatial conditions change.

[0011] In one embodiment of the present invention, step S1 includes steps S11-S12:

[0012] S11: Continuously acquire M data packets of CSI amplitude values, with each antenna link containing N subcarriers.

[0013] csi m,n =|csi m,n |(1≤m≤M,1≤n≤N) (1)

[0014] Among them, |csi m,n | represents the amplitude value corresponding to the nth subcarrier in the mth data packet.

[0015] S12: Arrange all amplitude values ​​of the t-th antenna link into an M×N amplitude matrix to construct the CSI amplitude matrix.

[0016]

[0017] Among them, AMP t Let be the CSI magnitude matrix of the t-th link.

[0018] S13: Construct a fingerprint feature map from the amplitude matrices of the T antenna links.

[0019] MAP = [AMP1,AMP2,…,AMP] T (3)

[0020] MAP is the fingerprint feature map, which is T×M×N dimensional.

[0021] In one embodiment of the present invention, step S2 includes:

[0022] A location feature encoder network is constructed to separate spatial information from fingerprint feature maps and extract location information. The input fingerprint feature map is downsampled and feature extracted through stride convolutional layers with layer normalization, and then the location feature code is obtained through a residual module.

[0023] In one embodiment of the present invention, step S3 includes:

[0024] By sampling the Gaussian distribution function multiple times, multiple spatial feature codes corresponding to changes in spatial conditions are output, providing more inputs for generating diverse fingerprint samples.

[0025] In one embodiment of the present invention, step S4 includes:

[0026] A generator network is constructed to combine the obtained location feature code and spatial feature code. The spatial feature code is passed through a fully connected layer to calculate the adaptive normalization layer parameters of the residual layer. Then, the location feature code is passed through the residual layer, upsampled, and the generated fingerprint feature map is output.

[0027] In a second aspect, the present invention provides a fingerprint expansion device for positioning, the device comprising:

[0028] Database building module: Collects CSI fingerprint signals of each reference point in the environment under initial spatial conditions, then collects a small number of CSI signals after spatial condition transformation, converts CSI signals containing multiple links into fingerprint feature maps, and establishes a fingerprint database.

[0029] Location feature encoding module: The location feature encoder encodes the CSI feature map of the initial spatial conditions to obtain the location feature code.

[0030] Environmental feature sampling module: Multiple samplings of the Gaussian distribution are performed to obtain multiple spatial feature codes after changes in spatial conditions.

[0031] Fingerprint data generation module: It feeds the location feature code and different spatial feature codes into the generator to obtain the fingerprint feature map after the spatial conditions change.

[0032] In one embodiment of the present invention, the database creation module is specifically used for:

[0033] Data packets containing M CSI amplitude values ​​are continuously collected, where each antenna link contains N subcarriers; then, the amplitude values ​​of the t-th antenna link are arranged into an M×N amplitude matrix to construct the CSI amplitude matrix; finally, the amplitude matrices of the T antenna links are used to construct a fingerprint feature map.

[0034] In one embodiment of the present invention, the location feature encoding module is specifically used for:

[0035] A location feature encoder network is constructed to separate spatial information and extract location information from the fingerprint feature map. The input fingerprint feature map is downsampled through a stride convolutional layer with modified linear unit activation function and layer normalization to achieve data dimensionality reduction and feature extraction. Then, the location feature code is obtained by passing it through a convolutional residual module with modified linear unit activation function and layer normalization.

[0036] In one embodiment of the present invention, the environmental feature sampling module is specifically used for:

[0037] A Gaussian error sampling module was built to sample the Gaussian distribution function multiple times, obtaining multiple spatial feature codes after the corresponding spatial conditions change, thus simulating some random errors in fingerprint data.

[0038] In one embodiment of the present invention, the fingerprint data generation module is specifically used for:

[0039] A fingerprint data generator network is constructed. This network feeds the given location feature code into the residual module, and the environmental feature code is passed through a fully connected layer to obtain adaptive normalization layer parameters which are then embedded into the residual module. Finally, the generated fingerprint data is obtained through an upsampling module.

[0040] By adopting the above technical solution, the present invention has the following beneficial effects:

[0041] This invention constructs fingerprint feature maps from the CSI amplitude matrix according to different communication links. It separates the location feature code and environment feature code of the fingerprint data within the initial environment through location feature encoding and environmental feature separation. Then, it performs multiple samplings of the Gaussian distribution to obtain different environment feature codes after environmental changes. Finally, it combines the location features of the initial environment with the different environment feature codes after environmental changes and inputs them into a generator to obtain the fingerprint samples corresponding to the environmental changes. This expands the quantity and richness of fingerprint data even when only a small amount of data exists after environmental changes. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings.

[0043] Figure 1 This is a flowchart illustrating a fingerprint augmentation method for positioning provided in an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram of the structure of the CSI amplitude fingerprint feature map provided in an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the structure of a location feature encoding network provided in an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram of the generator network structure provided in an embodiment of the present invention;

[0047] Figure 5 This is a schematic diagram of a fingerprint expansion device for positioning provided in an embodiment of the present invention. Detailed Implementation

[0048] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Indoor environments are complex and variable. The wireless signal characteristics of fingerprint data change with the movement of people and the influence of obstacles such as walls and furniture. This leads to significant distribution differences between offline fingerprint databases and online data in practical applications, posing a serious challenge to fingerprint localization. The only solution is to re-collect fingerprint databases in the changed environment, which is both labor-intensive and time-consuming, and fails to make efficient use of the initial database.

[0050] To address the aforementioned problems, embodiments of the present invention provide a fingerprint augmentation method and apparatus for positioning, which will be described in detail below.

[0051] First, an indoor fingerprint expansion method provided by an embodiment of the present invention will be described.

[0052] See Figure 1The above is a flowchart illustrating an indoor fingerprint positioning method provided by an embodiment of the present invention. The method is applied to an electronic device with computing capabilities. For example, the method is applied to a computer. The method includes the following steps S1 to S4.

[0053] S1: Collect CSI fingerprint signals of each reference point in the environment under the initial spatial conditions, then collect a small number of CSI signals after spatial condition transformation, convert the CSI signals containing multiple links into fingerprint feature maps, and establish a fingerprint database.

[0054] In fingerprint localization systems, RSS or Channel State Information (CSI) is typically used as a fingerprint, which is then further processed to estimate the location of the target. Compared to the coarse-grained RSS information at the physical layer, CSI can represent the subcarrier channel state of the transmitter antenna pair, such as signal scattering, multipath fading, and shadowing fading, providing higher sensing capabilities. Figure 2 As shown, the processing procedure for CSI data in this invention is as follows:

[0055] Step 1: Continuously acquire M data packets representing CSI amplitude values. Each antenna link contains N subcarriers.

[0056] csi m,n =|csi m,n |(1≤m≤M,1≤n≤N) (1)

[0057] Among them, |csi m,n | represents the amplitude value corresponding to the nth subcarrier in the mth data packet.

[0058] Step 2: Arrange all amplitude values ​​of the t-th antenna link into an M×N amplitude matrix to construct the CSI amplitude matrix.

[0059]

[0060] Among them, AMP t Let be the CSI magnitude matrix of the t-th link.

[0061] Step: Construct a fingerprint feature map from the amplitude matrices of the T antenna links.

[0062] MAP = [MAP1,AMP2,…,AMP] T (3)

[0063] MAP is the fingerprint feature map, which is T×M×N dimensional.

[0064] S2: The CSI feature map is encoded by a position feature encoder to obtain the position feature code.

[0065] Specifically, a location feature encoder network is constructed to separate spatial information from the fingerprint feature map and extract location information. The input fingerprint feature map is downsampled and feature extracted, and then normalized layer by layer. Finally, the location feature code is obtained.

[0066] For example, such as Figure 3 As shown, the original fingerprint features are passed through three stride convolutional layers to downsample the input, and then further processed by four residual blocks to obtain the location feature code. All convolutional layers are followed by instance normalization layers.

[0067] S3: Multiple samplings of the Gaussian distribution yield multiple spatial feature codes after spatial condition changes.

[0068] Specifically, a feature extraction network is built to sample the Gaussian distribution function multiple times and output multiple spatial feature codes corresponding to changes in spatial conditions, providing more inputs for generating diverse fingerprint samples.

[0069] For example, q(s) B ) is the prior distribution Different spatial feature codes from q(s) B (Sampling)

[0070] S4: The location feature code and different spatial feature codes are combined and fed into the generator to obtain the fingerprint feature map after the spatial conditions change.

[0071] Specifically, a generator network is built to combine the obtained location feature code and spatial feature code. The spatial feature code is passed through a fully connected layer to calculate the adaptive normalization layer parameters of the residual layer. Then, the location feature code is passed through the residual layer, upsampled, and the generated fingerprint feature map is output.

[0072] For example, such as Figure 4 As shown, the environmental feature code is fed into the fully connected layer to calculate the adaptive normalization parameters of the four residual layers. Then, the location feature code is passed through the four residual layers, and finally, it is upsampled through three two-dimensional convolutional layers that use modified linear unit activation functions and instance normalization to obtain the generated fingerprint data after environmental changes.

[0073] To train the location feature encoding module effectively in the offline phase, this invention designs the following loss function:

[0074]

[0075] Among them, the position feature encoder E L It was trained independently before the rest of the network. This indicates that data is extracted from the training fingerprint database before and after the environmental changes. Let y represent the total number of samples, and y represent the one-hot location label corresponding to the real sample x. This represents the probability distribution of the predicted labels. The following loss function is used to jointly train other network modules:

[0076]

[0077] in G is a spatial feature encoder for the environment before and after changes. A G B For the generator before and after environmental changes, D A D B It serves as a discriminator for distinguishing between before and after environmental changes. D A D B It is an auxiliary module for training the generator. To combat the losses, For fingerprint feature map reconstruction loss, For location feature code reconstruction loss, For environmental feature code reconstruction loss, λ x , λ c , λ s These are the control weights for the corresponding loss terms.

[0078] Corresponding to the aforementioned indoor fingerprint expansion method, this embodiment of the invention also provides an indoor fingerprint expansion device, such as... Figure 5 As shown.

[0079] Database Building Module B1: Collects CSI fingerprint signals of each reference point in the environment under initial spatial conditions, then collects a small number of CSI signals after spatial condition transformation, converts CSI signals containing multiple links into fingerprint feature maps, and establishes a fingerprint database.

[0080] Location feature encoding module B2: Encodes the CSI feature map of the initial spatial conditions through a location feature encoder to obtain the location feature code.

[0081] Environmental feature sampling module B3: Performs multiple samplings of the Gaussian distribution to obtain multiple spatial feature codes after spatial condition changes.

[0082] Fingerprint data generation module B4: It feeds the location feature code and different spatial feature codes into the generator to obtain the fingerprint feature map after the spatial conditions change.

[0083] In one embodiment of the present invention, the database creation module is specifically used for:

[0084] Data packets containing M CSI amplitude values ​​are continuously collected, where each antenna link contains N subcarriers; then, the amplitude values ​​of the t-th antenna link are arranged into an M×N amplitude matrix to construct the CSI amplitude matrix; finally, the amplitude matrices of the T antenna links are layered to construct a fingerprint feature map.

[0085] In one embodiment of the present invention, the location feature encoding module is specifically used for:

[0086] A location feature encoder network is constructed to separate spatial information and extract location information from the fingerprint feature map. The input fingerprint feature map is downsampled through a stride convolutional layer with modified linear unit activation function and layer normalization to achieve data dimensionality reduction and feature extraction. Then, the location feature code is obtained by passing it through a convolutional residual module with modified linear unit activation function and layer normalization.

[0087] In one embodiment of the present invention, the environmental feature sampling module is specifically used for:

[0088] A Gaussian error sampling module was built to sample the Gaussian distribution function multiple times, obtaining multiple spatial feature codes after the corresponding spatial conditions change, thus simulating some random errors in fingerprint data.

[0089] In one embodiment of the present invention, the fingerprint data generation module is specifically used for:

[0090] A fingerprint data generator network is constructed. The environmental feature code is passed through a fully connected layer to obtain adaptive normalization layer parameters which are then embedded into the residual module. The given location feature code is then fed into the residual module with a modified linear unit activation function. Finally, the generated fingerprint data is obtained through an upsampling module.

Claims

1. A fingerprint augmentation method for positioning, characterized in that, The steps include the following: S1: Collect CSI fingerprint signals of each reference point in the environment under the initial spatial conditions, and then collect a small number of CSI signals after spatial condition transformation; convert the CSI signals containing multiple links into CSI fingerprint feature maps and establish a fingerprint database; S2: Construct a location feature encoder network, and downsample and extract features from the CSI fingerprint feature map through a convolutional layer with layer normalization to separate the spatial information in the CSI fingerprint feature map and obtain the location feature code. S3: Build a Gaussian error sampling module to sample the Gaussian distribution function multiple times and obtain multiple spatial feature codes after spatial condition changes; S4: Build a generator network, combine the obtained location feature code and spatial feature code, calculate the adaptive normalization layer parameters of the residual layer after passing the spatial feature code through a fully connected layer, and then pass the location feature code through the residual layer, upsample it and output the generated CSI fingerprint feature map.

2. The fingerprint augmentation method for positioning according to claim 1, characterized in that, Step S1 includes: S11: Continuously acquire M data packets representing CSI amplitude values, with each antenna link containing N subcarriers. （1） in, This represents the amplitude value corresponding to the nth subcarrier in the mth data packet; S12: Arrange all amplitude values ​​of the t-th antenna link into an M×N amplitude matrix to construct the CSI amplitude matrix. （2） S13: Construct a CSI fingerprint feature map from the amplitude matrices of the T antenna links. MAP=[AMP1,AMP2,…,AMP T ] (3) The CSI fingerprint feature map is T×M×N dimensional.

3. The fingerprint augmentation method for positioning according to claim 1, characterized in that, In step 2, the CSI fingerprint feature map is encoded using a location feature encoder to obtain a location feature code, including: A location feature encoder network is constructed, and the spatial information in the CSI fingerprint feature map is separated and the location information is extracted using the location feature encoder network. The input CSI fingerprint feature map is downsampled and feature extracted through a stride convolutional layer with layer normalization. The extracted features are processed by residual mapping through a residual module to obtain the location feature code representing the location feature.

4. The fingerprint augmentation method for positioning according to claim 1, characterized in that, In step S3, multiple samplings of the Gaussian distribution are performed to obtain multiple spatial feature codes after spatial condition changes, including: A Gaussian error sampling module is constructed to perform multiple random samplings from a standard Gaussian distribution to obtain a latent space vector that follows a Gaussian distribution. The latent space vector is output as the spatial feature code to the generator network to characterize the environmental interference features under different spatial conditions, and is also used as the input to the generator network to generate a CSI fingerprint feature map with environmental diversity.

5. The fingerprint augmentation method for positioning according to claim 1, characterized in that, In step S4, the location feature code and different spatial feature codes are combined and fed into the generator to obtain the CSI fingerprint feature map after spatial condition changes, including: A generator network is constructed to combine the obtained location feature code and spatial feature code. The spatial feature code is passed through a fully connected layer to calculate the adaptive normalization layer parameters of the residual layer. Then, the location feature code is passed through the residual layer, upsampled, and the generated CSI fingerprint feature map is output.

6. A fingerprint augmentation device for positioning, the device comprising: Fingerprint database construction module: used to collect CSI fingerprint signals of each reference point in the environment under initial spatial conditions, and then collect a small number of CSI signals after spatial condition transformation; convert CSI signals containing multiple links into CSI fingerprint feature maps and establish a fingerprint database; Location feature encoding module: used to build a location feature encoder network, downsample and extract features from the CSI fingerprint feature map through a convolutional layer with layer normalization, separate the spatial information in the CSI fingerprint feature map, and obtain the location feature code; Environmental feature sampling module: used to sample the Gaussian distribution multiple times to obtain multiple spatial feature codes after changes in spatial conditions, so as to characterize the spatial interference features under different environments; Fingerprint data generation module: Build a generator network, combine the obtained location feature code and spatial feature code, calculate the adaptive normalization layer parameters of the residual layer after passing the spatial feature code through a fully connected layer, and then pass the location feature code through the residual layer, upsample it and output the generated CSI fingerprint feature map.

7. The fingerprint expansion device for positioning according to claim 6, characterized in that, The fingerprint database construction module is specifically used for: Data packets containing M CSI amplitude values ​​are continuously collected, where each antenna link contains N subcarriers; then, the amplitude values ​​of the t-th antenna link are arranged into an M×N amplitude matrix to construct the CSI amplitude matrix; finally, the amplitude matrices of the T antenna links are used to construct a CSI fingerprint feature map.

8. The fingerprint expansion device for positioning according to claim 6, characterized in that, The location feature encoding module is specifically used for: A location feature encoder network is constructed to separate spatial information and extract location information from the CSI fingerprint feature map. The input CSI fingerprint feature map is downsampled through a stride convolutional layer with modified linear unit activation function and layer normalization to achieve data dimensionality reduction and feature extraction. Then, the location feature code is obtained by passing it through a convolutional residual module with modified linear unit activation function and layer normalization.

9. The fingerprint expansion device for positioning according to claim 6, characterized in that, The environmental feature sampling module is specifically used for: A Gaussian error sampling module was built to sample the Gaussian distribution function multiple times, obtaining multiple spatial feature codes after the corresponding spatial conditions change, thus simulating some random errors in fingerprint data.

10. The fingerprint expansion device for positioning according to claim 6, characterized in that, The fingerprint data generation module is specifically used for: A fingerprint data generator network is constructed. Spatial feature codes are passed through a fully connected layer to obtain adaptive normalization layer parameters, which are then embedded into a residual module. The given location feature codes are then fed into the residual module with a modified linear unit activation function. Finally, the generated fingerprint data is obtained through an upsampling module.

Images