An indoor fingerprint positioning method and device
By constructing a spatial and temporal feature learning network based on CSI data, fusing weighted features and predicting location probabilities, the problem of insufficient accuracy and stability of Wi-Fi fingerprint positioning methods in complex dynamic environments is solved, and high-precision indoor positioning is achieved.
Patent Information
- Application Number
- CN202410065525.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-17
- Publication Date
- 2026-01-09
- Estimated Expiration
- 2044-01-17
AI Technical Summary
Existing Wi-Fi fingerprint positioning methods suffer from insufficient positioning accuracy and stability in complex and dynamic indoor environments. This is mainly due to the time sensitivity and environmental dependence of wireless signals, which leads to large fluctuations in the collected signal characteristics and frequent mismatches.
By constructing fingerprint features that include CSI data amplitude and phase information, learning sub-networks by combining spatial and temporal features, calculating weighted representation coefficients and performing feature fusion, and using pre-trained models to predict the matching probability of location information, the robustness and discriminativeness of fingerprint features are improved.
It significantly improves the accuracy and stability of indoor positioning, enhances the feasibility of fingerprint positioning systems in complex and dynamic environments, and improves the performance of location services.
Smart Images
Figure CN120343700B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] Embodiments of the present application relate to the technical field of data processing, in particular to an indoor fingerprint positioning method and device. TECHNICAL BACKGROUND
[0002] With the rapid development of the Internet of Things (IOT) industry, Location Based Services (LBS) have shown great application demand and market potential in precision marketing, smart logistics, smart home and other fields. Indoor positioning technology, as the main technical foundation of LBS, has attracted widespread attention from the academic community. In order to obtain high-precision and high-stability indoor location services, various sensing technologies have been explored and applied to various indoor scenarios, such as Inertial Measurement Unit (IMU), geomagnetism, radio frequency identification (RFID), camera, Wireless fidelity (Wi-Fi), etc. Among them, Wi-Fi is favored by researchers due to its unique advantages of low device price and widespread deployment.
[0003] The method based on Wi-Fi fingerprint matching is divided into an offline phase and an online phase. In the offline phase, wireless signals are collected in the space to be positioned and features are extracted, and a mapping relationship is established with the label of the corresponding position. In the online phase, the wireless signal features of the target to be positioned are extracted and matched with the position label to estimate the target position. This kind of method usually extracts fingerprint features from the received signal strength indicator (RSSI) and channel state information (CSI) of wireless sensing signals. Compared with RSSI, CSI has fine-grained perception ability, and by describing the amplitude and phase characteristics, it provides more detailed multipath propagation information. By utilizing the position specificity of CSI, related technologies use methods of manually extracting features or extracting features through deep learning algorithms to construct fingerprints for specific positions in the pre-positioning space.
[0004] Although the extracted features can theoretically achieve high-precision matching results, their actual performance will be greatly limited in actual scenarios. The time sensitivity and environment dependence of wireless signals will cause the collected signal metrics to fluctuate due to environmental dynamic changes, such as changes in furniture layout, human movement, closing of doors and windows, etc. This will cause the collected wireless signals to contain abnormal noise, exacerbating the occurrence of fingerprint false matching phenomena, causing problems such as decreased positioning accuracy and poor stability, which seriously affect the performance of indoor location services. Therefore, related technologies face the following problems: the fingerprint features extracted from the data collected in the offline phase and the online phase are different, and thus have insufficient robustness and discriminability, making it difficult to obtain accurate matching results. SUMMARY
[0005] The embodiments of the present application aim to provide an indoor fingerprint positioning method and device to increase the accuracy and stability of an indoor wireless fingerprint positioning system. The specific technical solutions are as follows.
[0006] In a first aspect, the embodiments of the present application provide an indoor fingerprint positioning method, which comprises:
[0007] Obtaining CSI data of Wi-Fi at a reference point with known coordinates in an indoor space and at a to-be-positioned point, applying CSI data collected in a continuous time sequence, and constructing fingerprint features containing amplitude and phase information in the CSI data;
[0008] Calculating weighted representation coefficients of each feature point in each channel of the CSI fingerprint features, updating the fingerprint features based on the coefficients, and obtaining spatial features of the CSI fingerprint;
[0009] Calculating weighted representation coefficients of each channel of the CSI fingerprint features, updating the fingerprint features based on the coefficients, obtaining time features of the CSI fingerprint, and fusing the features with the spatial features of the CSI fingerprint to update the fingerprint features representing the position information of the reference point;
[0010] Using the known coordinate information of the reference point to train a model to predict the matching probability between the input features and the position information and determine the ability of the fingerprint features to belong to the position based on the matching probability.
[0011] In an embodiment of the present application, the obtaining of the CSI data of Wi-Fi at the reference point with known coordinates in the indoor space and at the to-be-positioned point, the application of the CSI data collected in the continuous time sequence, and the construction of the fingerprint features containing the amplitude and phase information in the CSI data comprise:
[0012] Continuously obtaining a plurality of CSI data packets through a Wi-Fi device, retaining amplitude and phase information from all subcarriers and multi-links in each CSI data packet as single-layer features, and splicing the CSI single-layer features in the continuous time sequence to construct the fingerprint features.
[0013] In an embodiment of the present application, the calculation of the weighted representation coefficients of each feature point in each channel of the CSI fingerprint features, the updating of the fingerprint features based on the coefficients, and the obtaining of the spatial features of the CSI fingerprint comprise:
[0014] Building a spatial feature learning subnetwork, the subnetwork first performs pooling representation on each channel of the CSI fingerprint features, and then calculates spatial weighting coefficients M s through a fully connected layer, the coefficients being used to weight each feature point in each channel of the CSI fingerprint features, thereby updating the CSI fingerprint features and obtaining the spatial features of the CSI fingerprint.
[0015] In one embodiment of the present application, the calculation of the weighted representation coefficient of each channel of the CSI fingerprint feature is based on the coefficient to update the fingerprint feature, obtain the time feature of the CSI fingerprint, and fuse the feature with the spatial feature of the CSI fingerprint, update the fingerprint feature representing the position information of the reference point, including:
[0016] A time feature learning subnetwork is built, the subnetwork performs average pooling on all feature points of each channel of the CSI fingerprint feature, and then calculates a time weighting coefficient M through a fully connected layer t The coefficient is used to weight each channel of the CSI fingerprint feature, thereby updating the CSI fingerprint feature and obtaining the time feature of the CSI fingerprint.
[0017] The obtained time feature is used to fuse with the spatial feature obtained in the above-mentioned, all information of the two features is retained, and the fusion is performed in a splicing manner, and the fingerprint feature obtained after the fusion replaces the original fingerprint feature to represent the position information of the reference point.
[0018] In one embodiment of the present application, the ability of the model trained by using the coordinate information of the reference point to predict the matching probability between the input feature and the position information and determine the position to which the fingerprint feature belongs includes:
[0019] The CSI data obtained by the reference point and the coordinate of the reference point are jointly input into an indoor fingerprint positioning model, and a pre-trained model updates and adjusts model parameters; after the above-mentioned pre-trained model is obtained, the CSI data obtained at an unknown coordinate in the same indoor scene is input, and the position information of the to-be-positioned fingerprint feature output by the fingerprint positioning model is obtained.
[0020] In a second aspect, an embodiment of the present application provides an indoor fingerprint positioning device, and the device includes:
[0021] The acquisition module is configured to acquire CSI data of Wi-Fi at a reference point with a known coordinate in an indoor environment and at a to-be-positioned position, and construct a fingerprint feature containing amplitude and phase information in the CSI data by using CSI data collected in a continuous time sequence.
[0022] The spatial feature calculation module is configured to calculate a weighted representation coefficient of each feature point in each channel of the CSI fingerprint feature, update the fingerprint feature based on the coefficient, and obtain a spatial feature of the CSI fingerprint.
[0023] The time feature calculation module is configured to calculate a weighted representation coefficient of each channel of the CSI fingerprint feature, update the fingerprint feature based on the coefficient, obtain a time feature of the CSI fingerprint, fuse the feature with the spatial feature of the CSI fingerprint, and update the fingerprint feature representing the position information of the reference point.
[0024] Position prediction module: the ability of training model to predict the matching probability between input features and position information by using the coordinate information of the known reference point and determine the position of the fingerprint feature based on the matching probability.
[0025] In one embodiment of the application, the acquisition module is specifically configured to:
[0026] A plurality of CSI data packets are continuously acquired through the Wi-Fi device, the amplitude and phase information from all subcarriers and multi-links in each CSI data packet are reserved as single-layer features, and the CSI single-layer features in the continuous time sequence are spliced to construct the fingerprint features.
[0027] In one embodiment of the application, the spatial feature calculation module is specifically configured to:
[0028] A spatial feature learning subnetwork is built, the subnetwork first performs pooling representation on each channel of the CSI fingerprint feature, and then calculates a spatial weighting coefficient M s through a fully connected layer, the coefficient is used for weighting each feature point in each channel of the CSI fingerprint feature, so as to update the CSI fingerprint feature and obtain the spatial feature of the CSI fingerprint.
[0029] In one embodiment of the application, the time feature calculation module is specifically configured to:
[0030] A time feature learning subnetwork is built, the subnetwork performs average pooling on all feature points of each channel of the CSI fingerprint feature, and then calculates a time weighting coefficient M t through a fully connected layer, the coefficient is used for weighting each channel of the CSI fingerprint feature, so as to update the CSI fingerprint feature and obtain the time feature of the CSI fingerprint.
[0031] The obtained time feature is used for fusion with the spatial feature obtained in the above-mentioned, all information of the above-mentioned two features is reserved, the fusion is performed in a splicing manner, and the fingerprint feature obtained after the fusion replaces the original fingerprint feature to represent the position information of the reference point.
[0032] In one embodiment of the application, the position prediction module is specifically configured to:
[0033] The CSI data obtained by the reference point and the reference point coordinates are jointly input into an indoor fingerprint positioning model, a pre-trained model updates and adjusts model parameters, after the above-mentioned pre-trained model is obtained, the CSI data obtained at an unknown coordinate in the same indoor scene is input, and the position information of the to-be-positioned fingerprint feature output by the fingerprint positioning model is obtained.
[0034] The embodiment of the application has the following beneficial effects:
[0035] The application provides an indoor fingerprint positioning method and device, simultaneously calculates time features and space features of CSI and linearly fuses the features, and uses the fused features as fingerprints for indoor position estimation, which contributes an important technical tool for indoor position service development. Compared with related technologies, the indoor fingerprint positioning method and device can significantly improve the accuracy and stability of fingerprint positioning in a complex dynamic indoor scene, and increase the feasibility of deploying a fingerprint positioning system in an indoor actual scene. BRIEF DESCRIPTION OF DRAWINGS
[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only some embodiments of the present application, and other embodiments can be obtained by those skilled in the art based on these drawings.
[0037] Figure 1 A flowchart of an indoor fingerprint positioning method provided by an embodiment of the present application is shown in the figure.
[0038] Figure 2 A structure diagram of an original fingerprint feature provided by an embodiment of the present application is shown in the figure.
[0039] Figure 3 A structure diagram of a space feature learning subnetwork provided by an embodiment of the present application is shown in the figure.
[0040] Figure 4 A structure diagram of a time feature learning subnetwork provided by an embodiment of the present application is shown in the figure.
[0041] Figure 5 A structure diagram of an indoor fingerprint positioning device provided by an embodiment of the present application is shown in the figure. Specific implementation method
[0042] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art based on the present application belong to the scope of protection of the present application.
[0043] The wireless signal acquired by the indoor fingerprint positioning system is affected by the dynamic change of the environment and fluctuates, resulting in abnormal noise in the data, aggravating the occurrence of fingerprint error matching phenomenon, causing the problems of positioning accuracy decline, poor stability and the like, and seriously affecting the performance of indoor location service. Therefore, the indoor positioning method based on Wi-Fi fingerprint matching faces the following problems: the fingerprint features extracted from the data collected in the offline stage and the online stage are different, and the robustness and discriminability are insufficient, so that accurate matching results are difficult to obtain.
[0044] To solve the above problems, the embodiment of the application provides an indoor fingerprint positioning method and device, which are described below.
[0045] Firstly, an indoor fingerprint positioning method provided by the embodiment of the application is described.
[0046] Referring to Figure 1 A flowchart of an indoor fingerprint positioning method provided by the embodiment of the application is shown, the method is applied to an electronic device with computing capability, for example, the method is applied to a computer, and the method includes the following steps S101 to S104.
[0047] Step S101: acquiring CSI data of Wi-Fi at a reference point with a known coordinate in the room and at a to-be-positioned point, applying the CSI data collected in a continuous time sequence to construct fingerprint features containing amplitude and phase information in the CSI data.
[0048] In the process of indoor fingerprint positioning, a plurality of CSI data packets are continuously acquired by a Wi-Fi device, amplitude and phase information from all subcarriers and multi-link in each CSI data packet are reserved as single-layer features, and the CSI single-layer features in a continuous time sequence are spliced to construct fingerprint features.
[0049] For example, each acquired CSI data packet contains fine-grained physical information, which records the amplitude and phase information of the wireless signal propagating between the transmitter and the receiver, and the number of antennas of the receiver and the transmitter determines the complex propagation link characteristics of multiple output and multiple reception. The CSI received from each propagation link contains multiple subcarrier information.
[0050] Therefore, the amplitude data of one received CSI can be represented as a matrix CSI A :
[0051]
[0052] Each row vector in the matrix represents the amplitude of the CSI corresponding to multiple propagation links in a data packet, where n is the number of propagation links; each column vector represents the amplitude of each subcarrier of a single link CSI, where m is the number of subcarriers.
[0053] Similarly, the phase matrix CSI can be obtained from the acquired CSI data. P Its structure is as follows:
[0054]
[0055] To preserve the multi-link and multi-carrier information of wireless signals propagating in space, such as Figure 2 As shown, in the embodiments of the present invention, the CSI amplitude matrix CSI A and phase matrix CSI P These are concatenated together to form one layer of the fingerprint feature. Furthermore, the CSI amplitude matrix (CSI) mentioned above is obtained from multiple continuously acquired data packets. A and phase matrix CSI P The single-layer features of T data packets are then concatenated together to obtain the original fingerprint feature CSI. in .
[0056] Step S102: Calculate the weighted representation coefficient of each feature point in each channel of the CSI fingerprint feature, update the fingerprint feature based on this coefficient, and obtain the spatial features of the CSI fingerprint.
[0057] Specifically, a spatial feature learning subnetwork is constructed. This subnetwork first performs pooling representation on each channel of the CSI fingerprint feature, and then calculates the spatial weighting coefficients M through a fully connected layer. s The coefficients are used to weight each feature point in each channel of the CSI fingerprint feature, thereby updating the CSI fingerprint feature and obtaining the spatial features of the CSI fingerprint.
[0058] For example, the original fingerprint features described above are passed through a basic convolutional layer to obtain the output feature CSI. basic Its size is C×H×W. For example... Figure 3 As shown, firstly, regarding CSI basic The channel dimension is subjected to average pooling and max pooling to obtain two feature matrices of size 1×H×W. These two feature matrices are then concatenated to obtain a 2×H×W intermediate layer feature matrix m. A convolutional layer is applied to the intermediate layer feature matrix m, followed by activation using the sigmoid function to obtain the spatial feature weight map M. s Its size is 1×H×W. Finally, the above spatial feature weight map is compared with the feature CSI. basic By multiplying layer by layer, the robust spatial features after spatial attention weighting are obtained. The calculation formula for the spatial attention module is as follows:
[0059]
[0060] Where σ is the sigmoid operation, f CNN It is the size of the convolution kernel.
[0061] Step S103: Calculate the weighted representation coefficients of each channel of the CSI fingerprint feature, update the fingerprint feature based on these coefficients, obtain the temporal feature of the CSI fingerprint, and fuse this feature with the spatial feature of the CSI fingerprint to update the fingerprint feature representing the reference point location information.
[0062] Specifically, a temporal feature learning sub-network is constructed. This sub-network performs average pooling on all feature points of each channel of the CSI fingerprint feature, and then calculates the temporal weighting coefficients M through a fully connected layer. t The coefficients are used to weight each channel in the CSI fingerprint feature, thereby updating the CSI fingerprint feature and obtaining the time feature of the CSI fingerprint.
[0063] For example, the original fingerprint features described above are also passed through a basic convolutional layer to obtain the output feature CSI. basic Its size is C×H×W. For example... Figure 4 As shown, firstly, the obtained CSI basic Perform global average pooling to obtain the feature vector CSI with a global receptive field. z Its size is 1×1×C. The calculation formula for the global average pooling is as follows:
[0064]
[0065] CSI basic-C It is a feature CSI basic Features of the c-th layer channel.
[0066] Then, a fully connected network is used to perform a nonlinear transformation on the result obtained after the global average pooling, resulting in the channel weight vector M. t Its size is also 1×1×C. Finally, the obtained channel weight vector M... t As weights, the input features of the corresponding channels are weighted to obtain stable temporal features. The calculation formula for the above temporal attention module is as follows:
[0067] M t (CSI basic )=σ(g(CSI z ,W))
[0068] =σ(W2δ(W1CSI) z ))
[0069] where σ is sigmoid operation, g is fully connected network, After that, the CSI spatial feature and time feature vectors obtained by the above are linearly spliced to obtain a CSI space-time feature with a size of 2CxHxW. The space-time feature is used to update the original fingerprint feature, which is used as the fingerprint feature representing the reference point position information.
[0070] Step S104: The model is trained to predict the matching probability between the input feature and the position information using the known coordinate information of the reference point, and to determine the position to which the fingerprint feature belongs based on the matching probability.
[0071] Specifically, the CSI data obtained by the reference point and the reference point coordinates are jointly input into the indoor fingerprint positioning model, and the pre-trained model is updated to adjust the model parameters. After obtaining the pre-trained model, the CSI data obtained at an unknown coordinate in the same indoor scene is input to obtain the position information of the to-be-positioned fingerprint feature output by the fingerprint positioning model.
[0072] For example, the updated fingerprint feature is subjected to a convolution operation to obtain a one-dimensional feature vector h with a dimension equal to the number of reference points with known positions, i.e., d h = num_RPs. The feature vector is input into a SoftMax classifier for position classification prediction. The probability of the positioning prediction result for the i-th position point is:
[0073]
[0074] where h i represents the i-th feature value in the feature vector h.
[0075] In order to train the model corresponding to the method in the offline stage, cross entropy is used as the overall loss function, which is represented as:
[0076]
[0077] where y i and are the true label and the prediction result of the to-be-positioned feature, respectively.
[0078] In the application stage, the collected CSI data is input into the above model, and the model outputs the matching probability value of the to-be-positioned fingerprint feature and each reference point. In order to further obtain the accurate position estimation of the fingerprint feature, the positions of the top three reference points in the prediction probability are selected, and the final position prediction is obtained by a weighted centroid method:
[0079]
[0080] where L is the final predicted coordinate, L i , Pi respectively correspond to the three most probable predicted reference point coordinates and the corresponding probability values.
[0081] Corresponding to the foregoing indoor fingerprint positioning method, the embodiment of the application further provides an indoor fingerprint positioning device.
[0082] Referring to Figure 5 A structural schematic diagram of an indoor fingerprint positioning device provided by the embodiment of the application is shown in the figure, and the device is applied to an electronic device with computing capability.
[0083] The acquisition module 501 is configured to acquire CSI data of Wi-Fi at a reference point with a known coordinate in an indoor space and at a to-be-positioned position, apply CSI data collected in a continuous time sequence, and construct fingerprint features containing amplitude and phase information in the CSI data.
[0084] The spatial feature calculation module 502 is configured to calculate weighted representation coefficients of each feature point in each channel of the CSI fingerprint features, update the fingerprint features based on the coefficients, and obtain spatial features of the CSI fingerprint.
[0085] The time feature calculation module 503 is configured to calculate weighted representation coefficients of each channel of the CSI fingerprint features, update the fingerprint features based on the coefficients, obtain time features of the CSI fingerprint, and fuse the features with the spatial features of the CSI fingerprint to update the fingerprint features representing position information of the reference point.
[0086] The position prediction module 504 is configured to train a model to predict matching probability between input features and position information by using known coordinate information of the reference point and determine the position to which the fingerprint features belong based on the matching probability.
[0087] In one embodiment of the application, the acquisition module 501 is specifically configured to:
[0088] A plurality of CSI data packets are continuously acquired by the Wi-Fi device, amplitude and phase information from all subcarriers and multiple links in each CSI data packet are reserved as single-layer features, and the CSI single-layer features in a continuous time sequence are spliced to construct the fingerprint features.
[0089] In one embodiment of the application, the spatial feature calculation module 502 is specifically configured to:
[0090] A spatial feature learning subnetwork is built, the subnetwork first performs pooling representation on each channel of the CSI fingerprint features, and then calculates a spatial weighting coefficient M s by using a full connection layer, the coefficient is used to weight each feature point in each channel of the CSI fingerprint features, thereby updating the CSI fingerprint features and obtaining spatial features of the CSI fingerprint.
[0091] In one embodiment of the present application, the time feature calculation module 503 is specifically used for:
[0092] The time feature learning subnetwork is built, the subnetwork performs average pooling on all feature points of each channel of the CSI fingerprint feature, and then calculates a time weighting coefficient M through a full connection layer t The coefficient is used for weighting each channel in the CSI fingerprint feature, thereby updating the CSI fingerprint feature to obtain the time feature of the CSI fingerprint. The obtained time feature is used for fusion with the spatial feature obtained in the above-mentioned manner, all information of the two features is retained, and the two features are fused in a splicing manner. The fingerprint feature obtained after the fusion replaces the original fingerprint feature to represent the reference point position information.
[0093] In one embodiment of the present application, the position prediction module 504 is specifically used for:
[0094] The CSI data obtained by the reference point and the reference point coordinates are jointly input into an indoor fingerprint positioning model, and a pre-trained model updates and adjusts model parameters. After obtaining the pre-trained model, the CSI data obtained at an unknown coordinate in the same indoor scene is input, and the position information of the to-be-positioned fingerprint feature output by the fingerprint positioning model is obtained.
Claims
1. A method of indoor fingerprinting, characterized in that, The method comprises: Applying the CSI data collected in continuous time sequence, constructing the fingerprint feature containing the amplitude and phase information in the CSI data, comprising: Through the Wi-Fi device, continuously acquiring multiple CSI data packets at the reference point with known coordinates in the indoor and at the to-be-positioned point, reserving the amplitude and phase information from all subcarriers and multi-links in each CSI data packet as a single-layer feature, splicing the CSI single-layer features in continuous time sequence to construct the fingerprint feature; Calculating the weighted representation coefficients of each feature point in each channel of the CSI fingerprint feature, updating the fingerprint feature based on the coefficients, obtaining the spatial feature of the CSI fingerprint, comprising: A spatial feature learning subnetwork is built, which firstly performs pooling on each channel of the CSI fingerprint feature, and then calculates a spatial weighting coefficient M through a fully connected layer, the coefficient being used to weight each feature point in each channel of the CSI fingerprint feature. s , the coefficient being used to weight each feature point in each channel of the CSI fingerprint feature. Calculating the weighted representation coefficients of each channel of the CSI fingerprint feature, updating the fingerprint feature based on the coefficients, obtaining the time feature of the CSI fingerprint, and fusing the feature with the spatial feature of the CSI fingerprint, updating the fingerprint feature representing the position information of the reference point, comprising: A time feature learning subnetwork is built, which performs average pooling on all feature points of each channel of the CSI fingerprint feature, and then calculates a time weighting coefficient M through a fully connected layer, which is used to weight each channel in the CSI fingerprint feature t . Obtaining the time feature and the spatial feature of the CSI fingerprint, reserving all the information, fusing in a splicing manner, and replacing the original fingerprint feature representing the position information of the reference point with the fused fingerprint feature; The ability of training the model using the known coordinate information of the reference point to predict the matching probability between the input feature and the position information and determine the position of the fingerprint feature based on the matching probability.
2. The method of claim 1, wherein, The ability of training the model using the known coordinate information of the reference point to predict the matching probability between the input feature and the position information and determine the position of the fingerprint feature based on the matching probability, comprising: Inputting the CSI data acquired by the reference point and the reference point coordinates into the indoor fingerprint positioning model together, updating and adjusting the model parameters of the pre-trained model; after obtaining the pre-trained model, inputting the CSI data acquired at the unknown coordinates in the same indoor scene, obtaining the position information of the to-be-positioned fingerprint feature output by the fingerprint positioning model.
3. An indoor fingerprinting apparatus, comprising: The device comprises: The acquisition module is specifically configured to: Continuously acquire multiple CSI data packets at the reference point with known coordinates in the indoor and at the to-be-positioned point, splice the amplitude and phase information in each CSI data packet based on the CSI data collected in continuous time sequence, and construct the fingerprint feature; The spatial feature calculation module is specifically configured to: The space feature learning subnetwork is built, first, the channels of the CSI fingerprint features are pooled, and then the weighted representation coefficients, i.e., the space weighted coefficients M, of the feature points in each channel are calculated through a fully connected layer s , the coefficients are used to weight the feature points in each channel of the CSI fingerprint features, so as to update the CSI fingerprint features and obtain the space features of the CSI fingerprint; The time feature calculation module is specifically configured to: The time feature learning subnetwork is built, first, all feature points of each channel of the CSI fingerprint feature are average-pooled, and then weighted representation coefficients, i.e., time weighting coefficients M of each channel are calculated through a fully connected layer t The coefficients are used for weighting each channel in the CSI fingerprint feature, thereby updating the CSI fingerprint feature, obtaining the time feature of the CSI fingerprint, and fusing the feature and the spatial feature of the CSI fingerprint in a splicing manner to update the fingerprint feature representing the reference point position information. The position prediction module has the ability of training the model using the known coordinate information of the reference point to predict the matching probability between the input feature and the position information and determine the position of the fingerprint feature based on the matching probability.
4. The apparatus of claim 3, wherein, The acquisition module, the spatial feature calculation module, the time feature calculation module, and the position prediction module are specifically configured to: Input the CSI data acquired by the reference point and the reference point coordinates into the indoor fingerprint positioning model together, update and adjust the model parameters of the pre-trained model; after obtaining the pre-trained model, input the CSI data acquired at the unknown coordinates in the same indoor scene, and obtain the position information of the to-be-positioned fingerprint feature output by the fingerprint positioning model.