Multi-target positioning method based on Wi-Fi, electronic equipment and storage medium

Abstract

The invention relates to the technical field of wireless network positioning, in particular to a multi-target positioning method based on Wi-Fi, electronic equipment and a storage medium. According to the method, single-target fingerprint data are acquired, multiple (two or more) single-target fingerprint data are fused to obtain fused fingerprint data capable of simulating multi-target fingerprint data, and a fused fingerprint generation model and a multi-label positioning classification model are jointly trained according to the fused fingerprint data. And processing the multi-target fingerprint data through the trained multi-label positioning classification model so as to realize multi-target positioning.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless network positioning, and in particular to a Wi-Fi-based multi-target positioning method, electronic equipment and storage medium. Background Art

[0002] Indoor location-based services (LBS), including applications such as security monitoring, smart home automation, and asset tracking, require accurate and reliable positioning systems to track multiple target groups in complex indoor environments. Traditional device-based positioning methods rely on users to carry dedicated transceiver devices or tags, which face practicality and user cooperation issues, thus driving people to turn to device-free passive positioning (DFPL) methods. DFPL uses wireless signals in the environment, which are subtly modulated by the presence and movement of target groups, to achieve target positioning without carrying any hardware. However, reliable multi-target positioning remains a major challenge due to factors such as nonlinear signal interactions, multipath signal superposition, and occlusion caused by walls, furniture, and other obstacles.

[0003] Wi-Fi-based DFPL approaches take advantage of the ubiquitous Wi-Fi infrastructure, providing a more accessible and economically viable solution. Techniques leveraging channel state information (CSI) have explored compressed sensing, channel diversity, and sparsity-driven model selection (e.g., SA-M-SBL, SBL) to improve positioning accuracy and robustness under incomplete prior knowledge. Deep learning solutions combined with Wi-Fi CSI enable simultaneous crowd counting and positioning. These methods typically require large multi-target fingerprint databases and computing resources, which complicates their deployment in real-time or resource-constrained scenarios. Methods based on time of flight (ToF) and round-trip time (RTT) can achieve high positioning accuracy but require specialized hardware.

[0004] The recently emerged MTFLoc method treats multi-target localization as a multi-label classification problem and combines it with high-resolution fingerprint recognition technology; this method may be more efficient, but lacks advanced domain adaptation and synthetic sample generation capabilities, thus limiting its adaptability in dynamic environments. In contrast, the MLDA-MultiLoc method uses unsupervised domain adaptation and fusion generation techniques to handle complex indoor scenes; however, its computational intensity and complex training and tuning process may hinder its deployment in practical scenarios. Summary of the invention

[0005] Based on this, the present invention provides a Wi-Fi-based multi-target positioning method, an electronic device, and a storage medium, which at least solve one problem in the prior art.

[0006] In a first aspect, the present invention provides a Wi-Fi-based multi-target positioning method, which comprises the following steps: Acquire single target fingerprint data of training points based on Wi-Fi signals; Construct a fusion fingerprint generation model and a multi-label positioning classification model, and use single-target fingerprint data to collaboratively train the fusion fingerprint generation model and the multi-label positioning classification model; wherein, the single-target fingerprint data is fused to obtain fusion fingerprint data; the fusion fingerprint data is classified using the training points in the single-target fingerprint data as labels; the mutual information loss of the fusion fingerprint generation model is calculated based on the Hilbert-Schmidt independence criterion, the cross entropy loss of the multi-label positioning classification model is calculated, and the parameters of the fusion fingerprint generation model and the multi-label positioning classification model are updated according to the mutual information loss and the cross entropy loss; Obtain multi-target fingerprint data and use the trained multi-label localization classification model to predict the probability of the target location.

[0007] It should be noted that this method obtains single-target fingerprint data, fuses multiple (two or more) single-target fingerprint data, and obtains fused fingerprint data that can simulate multi-target fingerprint data. Based on this, the fused fingerprint generation model and the multi-label positioning classification model are jointly trained, and the multi-target fingerprint data is processed by the trained multi-label positioning classification model to achieve multi-target positioning. Among them, when training the fused fingerprint generation model, the Hilbert-Schmidt independence criterion (HSIC) is used to measure the dependency between the single-target fingerprint data and the fused fingerprint data to determine the mutual information loss (MIL).

[0008] In some optional embodiments, acquiring single-target fingerprint data of a training point based on a Wi-Fi signal includes: Arrange a pair of Wi-Fi signal transceiver devices in the positioning area; Divide the positioning area into multiple training points; For each training point, wireless signal data is collected through a Wi-Fi signal transceiver, and the wireless signal data is preprocessed to obtain the single target fingerprint data of the training point.

[0009] It should be noted that the positioning area refers to an area where multi-target positioning is required, such as indoors. The Wi-Fi signal transceiver device can be a wireless router and / or a computer with a wireless network card.

[0010] In some optional embodiments, preprocessing the wireless signal data includes: denoising and phase calibrating the channel state information data in the wireless signal data, then splicing the obtained amplitude data and phase data into feature fingerprint data, and obtaining single target fingerprint data after normalization.

[0011] In some optional embodiments, the denoising is wavelet domain denoising (WDD).

[0012] In some optional embodiments, phase calibration includes a linear transformation as shown in equation (1): in, and denote the slope and variance of the linear correction, respectively; Indicates i The actual index value of the subcarrier; Indicates i The calibrated phase data of each subcarrier; Indicates i The original phase data obtained by measuring the subcarriers; and Respectively represent the original phase data measured for the 30th and 1st subcarriers; and Represent the actual index values ​​of the 30th and 1st subcarriers respectively; i and j They represent the subcarrier subscripts respectively.

[0013] In some optional embodiments, calculating the mutual information loss of the fusion fingerprint generation model based on the Hilbert-Schmidt independence criterion includes: The Hilbert-Schmidt independence criterion is used to measure the dependency between single target fingerprint data and fused fingerprint data, as shown in formula (2); Calculate the mutual information loss as shown in formula (4); in, represents mutual information loss; X Represents the input single target fingerprint data; Z Represents the output fusion fingerprint data; represents the batch size; tr(·) represents the sum of the main diagonal elements of the matrix; H represents a positive definite matrix; , They represent the kernel matrix calculated using the radial basis function kernel (RBF), Elements and The calculation method of the radial basis function kernel of the elements is shown in formula (3); Among them, exp(·) represents the power of the natural constant e, Represents the width parameter of the kernel function.

[0014] In some optional embodiments, calculating the cross entropy loss of the multi-label localization classification model includes: Calculate the cross entropy loss of the multi-label localization classification model according to formula (6); in, Indicates the number of samples; Indicates the number of categories; represents the probability that sample i is predicted to be label n, as shown in formula (5); represents an indicator variable; in, Represents the functionalization of the layer before the output layer.

[0015] In some optional embodiments, using a trained multi-label positioning classification model to predict the probability of a target location includes: inputting multi-target fingerprint data into a trained multi-label positioning classification model, and having the trained multi-label positioning classification model output the probability of the target location; wherein, if the probability of the target location is greater than or equal to 0.5, it indicates that the target location is occupied by the target.

[0016] In a second aspect, the present invention provides an electronic device, comprising: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores instructions, which, when executed by at least one processor, implement the Wi-Fi-based multi-target positioning method as described above.

[0017] In a third aspect, the present invention provides a computer-readable storage medium storing instructions, which, when executed by a processor, implement the Wi-Fi-based multi-target positioning method as described above.

[0018] Due to the adoption of the above technical solution, the embodiments of the present invention have at least the following beneficial effects: (1) Only a pair of Wi-Fi signal transceiver devices are needed, and the hardware requirements are extremely low; (2) Create virtual samples for multi-target scenarios by synthesizing and fusing fingerprint data from single-target fingerprint data. This allows multi-target localization using only single-target fingerprint data, eliminating the need for a large multi-target fingerprint database. (3) The fingerprint generation model and the multi-label localization classification model (localization model) are jointly trained to enhance the robustness to nonlinear interactions, signal superposition, and occlusion effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1Schematic diagram of the architecture of the fusion fingerprint generation model and the multi-label positioning classification model in an embodiment of the present invention.

[0020] Figure 2 It is a schematic diagram of the layout plan of the positioning scene in an embodiment of the present invention.

[0021] Figure 3 4 is a diagram comparing system performance under different target numbers in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following is a clear and complete description of the concept of the present invention and the technical effects produced, so as to fully explain the purpose, scheme and effect of the present invention.

[0023] According to an embodiment of the present invention, a multi-target positioning method based on Wi-Fi is provided. In this method, single-target fingerprint data of known training points are first obtained, each training point corresponds to a specific indoor location, and a complete single-target fingerprint database is established by controlling conditions and recording target signals one by one. A pair of Wi-Fi signal transceivers are used to regularly collect wireless signal data in the environment, focusing on channel state information (CSI). Unlike systems that rely on dedicated hardware (such as UWB or millimeter wave arrays), this configuration is more cost-effective and practical.

[0024] The preprocessing process of wireless signal data includes noise filtering, phase calibration, feature normalization and dimensionality reduction to generate stable high-dimensional fingerprint data representing spatial patterns. Noise filtering uses wavelet domain denoising (WDD) to remove noise by thresholding the coefficients, which can more effectively retain the core features of the signal than traditional low-pass filters. After denoising, phase calibration is performed to correct the original phase data that is usually affected by time synchronization errors and environmental noise, in which a linear transformation as shown in equation (1) is performed, which produces stable and usable phase information.

[0025] in, and denote the slope and variance of the linear correction, respectively; Indicates i The actual index value of the subcarrier; Indicates i The calibrated phase data of each subcarrier; Indicates i The original phase data obtained by measuring the subcarriers; and Respectively represent the original phase data measured for the 30th and 1st subcarriers; and Represent the actual index values ​​of the 30th and 1st subcarriers respectively; i and j They represent the subcarrier subscripts, and their values ​​can be 1 to 30.

[0026] After that, the amplitude data and the calibrated phase data are spliced ​​into 180-dimensional feature fingerprint data, and the features are normalized to make them consistent within the numerical range. These processed and calibrated features are filled into a single-target fingerprint database, and the single-target fingerprint database is divided into multiple training points. Each training point contains single-target fingerprint data, which provides a solid foundation for multi-target fingerprint data generation, classifier training, and improved location-based services.

[0027] In order to simulate multi-target scenarios, a fusion fingerprint generation model is constructed to fuse single-target fingerprint data into multi-target fingerprint data. By maximizing mutual information, the fusion fingerprint generation model can retain basic spatial details even in the case of nonlinear interactions and occlusions. In this way, the method can flexibly adapt to different numbers of targets without requiring a large number of multi-target samples.

[0028] At the same time, a multi-label positioning and classification model is constructed to discretize the indoor area into multiple training points, each of which serves as a "label". The multi-label positioning and classification model allows the simultaneous prediction of the existence of multiple targets. By jointly training the fusion fingerprint generation model and the multi-label positioning and classification model, complex fusion features are mapped to multiple target locations in real time.

[0029] The architecture of the fusion fingerprint generation model and the multi-label positioning classification model is as follows Figure 1 As shown in the figure. In the fusion fingerprint generation model, three 180-dimensional single-target fingerprint features are encoded into 256-dimensional representations by a separate encoder (a fully connected layer followed by a ReLU activation function). The three encoded features are concatenated into a 768-dimensional vector and then compressed into a 180-dimensional fusion fingerprint. In order to minimize the mutual information loss, the Hilbert-Schmidt independence criterion (HSIC) is used to measure the input and The dependency relationship between them is shown in formula (2); in, represents the batch size; tr(·) represents the sum of the main diagonal elements of the matrix; H represents a positive definite matrix; , They represent the kernel matrix calculated using the radial basis function kernel (RBF), Elements and The calculation method of the radial basis function kernel of the elements is shown in formula (3); Among them, exp(·) represents the power of the natural constant e, Represents the width parameter of the kernel function.

[0030] Calculate the mutual information loss as shown in formula (4); in, represents the mutual information loss; X represents the input single target fingerprint data; Z represents the output fusion fingerprint data. By minimizing ,The model retains the basic spatial features of the fused fingerprint.

[0031] In the multi-label localization classification model, the fully connected neural network (followed by the sigmoid activation function) predicts independent binary probabilities for each label independently. Unlike softmax, the sigmoid activation function ensures independent binary probabilities, as shown in Equation (5); in, Represents the functionalization of the layer before the output layer.

[0032] A threshold (0.5 in this example) is applied to determine which positions are occupied by the target. The training target uses the cross entropy loss, as shown in equation (6); in, Indicates the number of samples; Indicates the number of categories; represents the probability that sample i is predicted to be label n, as shown in formula (5); Represents an indicator variable, which is 1 if the category is the same as the true label of sample i, otherwise it is 0.

[0033] The total training loss of the fusion fingerprint generation model and the multi-label positioning classification model is: Here, λ represents the equilibrium parameter.

[0034] The joint training process of the fusion fingerprint generation model and the multi-label positioning classification model is as follows: Input training set, the training set is a single target fingerprint database ; Set the learning rate ; Multi-label localization and classification model initialized by pre-training of single-target fingerprint database ; Initialize the fusion fingerprint generation model ; Repeat the following steps until the training is complete: Three batches are randomly selected , each batch size is , and input it into the model; Get fusion tags ; Calculate the loss according to formula (4): ; Update using backpropagation ; According to formula (6) and formula (7), the loss is calculated and ; Update using backpropagation and .

[0035] During the training process, key optimization details include standard parameter initialization (He initialization in this example) to ensure stability, and pre-training the multi-label localization classification model on single target fingerprints before joint training. The Adam optimizer is used with an initial learning rate of 0.001. , , to accelerate convergence and maintain stability. Balance parameters By performing a grid search on the validation set, we select ), to achieve a proper balance between classification accuracy and information retention. An appropriate batch size (128 in this embodiment) ensures the stability and representativeness of HSIC estimation. This joint training framework enables the method to efficiently transform single-target fingerprint data into a reliable basis for multi-target positioning, achieving accurate, robust and efficient indoor positioning under a variety of complex conditions.

[0036] In order to evaluate the Wi-Fi based multi-target localization method (the method of the embodiment of the present invention), a prototype was implemented using off-the-shelf Wi-Fi devices, and extensive experiments were conducted in a representative indoor environment.

[0037] The experiments were conducted in a standard laboratory, such as Figure 2 As shown in the figure, a TP-LINK WR841N router was used as a transmitter and a laptop (running Ubuntu 14.04) equipped with an Intel 5300 network card was used as a receiver to record the raw CSI measurement data. 28 training points (RPs) were defined and distributed in an open space.

[0038] To collect CSI data, CSI Tools was used and maintained at a transmission rate of 100 packets per second. For the single-target scenario, 5 training points were selected and approximately 900 CSI measurements were recorded for each point. For the multi-target scenario, 5 sets of CSI data were collected for 2, 3, and 4 targets located at different training points, each containing 900 measurements. Participants were allowed to move naturally (e.g., turn around, stretch) at designated training or test locations. To evaluate stability, a five-day experiment was conducted, with training and test data collected on the first day and only test data collected on the next few days, ultimately forming a combined test dataset.

[0039] In order to evaluate the accuracy of the method of the embodiment of the present invention under different numbers of targets, it is compared with the most advanced multi-target localization methods (including ComDec, SA-M-SBL and MTFLoc). All methods rely on the single-link CSI data collected in the above manner, and each method uses its recommended parameters.

[0040] Table 1 shows the accuracy of each method in estimating the number of targets. Although the accuracy naturally decreases as the number of targets increases, the method of the embodiment of the present invention outperforms the other methods, reaching 0.91 for a single target and 0.75 for four targets. In contrast, MTFLoc, SA-M-SBL, and ComDec show lower accuracy, highlighting the advantages of the method of the embodiment of the present invention, which is not limited by the sparsity assumption.

[0041] Table 1 Accuracy of each method in estimating the number of targets like Figure 3 As shown, as the number of targets increases, the average positioning error gradually increases (only samples with an accurate estimate of the number of targets will perform positioning error calculations). Despite this, the method of the embodiment of the present invention still has higher accuracy than the most advanced multi-target positioning method. In the case of four targets, the average error of the method of the embodiment of the present invention is 0.39 meters, while MTFLoc is 0.42 meters, SA-M-SBL is 0.64 meters, and ComDec is 0.61 meters. Thanks to the fusion fingerprint generation model, the method of the embodiment of the present invention remains stable when the number of targets increases. In addition, the method of the embodiment of the present invention can maintain high accuracy even in the case of a single target, achieving an average positioning error as low as 0.13 meters. This demonstrates the ability of the method of the embodiment of the present invention to consistently provide accurate positioning results under different conditions, thanks to the pre-training initialization of the multi-label positioning classification model.

[0042] Experimental results show that the method of the embodiments of the present invention consistently outperforms the state-of-the-art methods in terms of accuracy and scalability, and maintains reliable performance even when the number of targets increases.

[0043] The above is only a preferred embodiment of the present invention. The present invention is not limited to the above implementation. As long as the technical effect of the present invention is achieved by the same or equivalent means, it should belong to the protection scope of the present invention. Within the protection scope of the present invention, its technical scheme and / or implementation method can have various modifications and changes.

Claims

1. A multi-target positioning method based on Wi-Fi, characterized in that: The following steps are involved: Acquire single target fingerprint data of training points based on Wi-Fi signals; Constructing a fusion fingerprint generation model and a multi-label positioning classification model, and using single-target fingerprint data to collaboratively train the fusion fingerprint generation model and the multi-label positioning classification model; wherein the single-target fingerprint data is fused to obtain fusion fingerprint data; The training points in the single-target fingerprint data are used as labels to classify the fused fingerprint data; the mutual information loss of the fused fingerprint generation model is calculated based on the Hilbert-Schmidt independence criterion, and the cross entropy loss of the multi-label positioning classification model is calculated; the parameters of the fused fingerprint generation model and the multi-label positioning classification model are updated according to the mutual information loss and the cross entropy loss; Obtain multi-target fingerprint data and use the trained multi-label localization classification model to predict the probability of the target location.

2. The method according to claim 1, characterized in that: The single target fingerprint data of the training point obtained based on Wi-Fi signals includes: Arrange a pair of Wi-Fi signal transceiver devices in the positioning area; Divide the positioning area into multiple training points; For each training point, wireless signal data is collected through a Wi-Fi signal transceiver, and the wireless signal data is preprocessed to obtain the single target fingerprint data of the training point.

3. The method according to claim 2, characterized in that The preprocessing of the wireless signal data includes: denoising and phase calibration of the channel state information data in the wireless signal data, then splicing the obtained amplitude data and phase data into feature fingerprint data, and obtaining single target fingerprint data after normalization.

4. The method according to claim 3, characterized in that Denoising is done in wavelet domain.

5. The method according to claim 3, characterized in that: Phase calibration involves a linear transformation as shown in equation (1): in, and denote the slope and variance of the linear correction, respectively; Indicates i The actual index value of the subcarrier; Indicates i The calibrated phase data of each subcarrier; represents the original phase data measured for the i-th subcarrier; and Respectively represent the original phase data measured for the 30th and 1st subcarriers; and Represent the actual index values ​​of the 30th and 1st subcarriers respectively; i and j They represent the subcarrier subscripts respectively.

6. The method according to claim 1, characterized in that The mutual information loss of the fusion fingerprint generation model is calculated based on the Hilbert-Schmidt independence criterion, including: The Hilbert-Schmidt independence criterion is used to measure the dependency between single target fingerprint data and fused fingerprint data, as shown in formula (2); Calculate the mutual information loss as shown in formula (4); in, represents mutual information loss; X represents the input single target fingerprint data; Z represents the output fusion fingerprint data; represents the batch size; tr(·) represents the sum of the main diagonal elements of the matrix; H represents a positive definite matrix; , They represent the kernel matrices calculated using the radial basis function kernel, Elements and The calculation method of the radial basis function kernel of the elements is shown in formula (3); Among them, exp(·) represents the power of the natural constant e, Represents the width parameter of the kernel function.

7. The method according to claim 6, characterized in that Calculating the cross entropy loss for a multi-label localization classification model involves: Calculate the cross entropy loss of the multi-label localization classification model according to formula (6); in, Indicates the number of samples; Indicates the number of categories; represents the probability that sample i is predicted to be label n, as shown in formula (5); represents an indicator variable; in, Represents the functionalization of the layer before the output layer.

8. The method according to claim 7, characterized in that Predicting the probability of the target location using the trained multi-label positioning classification model includes: inputting the multi-target fingerprint data into the trained multi-label positioning classification model, and the trained multi-label positioning classification model outputs the probability of the target location; wherein, if the probability of the target location is greater than or equal to 0.5, it indicates that the target location is occupied by the target.

9. An electronic device, characterized in that: include: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores instructions, and when the instructions are executed by at least one processor, the Wi-Fi-based multi-target positioning method as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, characterized in that: Instructions are stored, and when the instructions are executed by the processor, the Wi-Fi-based multi-target positioning method as described in any one of claims 1 to 8 is implemented.

Images