A target positioning method and device combining CSI coarse positioning and Y-type CCD fine correction

Abstract

The application belongs to the technical field of positioning and measurement, and particularly relates to a target positioning method and device combining CSI coarse positioning and Y-type CCD fine correction, which comprises the following steps: S1, CSI acquisition equipment is arranged in a target area, and CSI data between a sending end and a receiving end is acquired; S2, a Y-type CCD sensor is geometrically calibrated and a coordinate system is corrected, and coordinate reconstruction is completed; S3, the CSI data is preprocessed, noise is removed, and input is normalized; S4, time domain features of the CSI are extracted by using a time domain branch; S5, frequency domain features of the CSI are extracted by using a frequency domain branch; S6, a gating fusion module is used to fuse the time domain and frequency domain features, and the approximate position coordinates of a target are output; S7, the Y-type CCD sensor is controlled to optically acquire a target area, and point light source coordinate information is acquired; and S8, error correction is performed on the data acquired by the Y-type CCD, and the accurate position coordinates of the target are output.

Description

Technical Field

[0001] This invention relates to the field of positioning and measurement technology, specifically to a target positioning method and apparatus that combines CSI coarse positioning with Y-type CCD fine correction. Background Technology

[0002] With the development of wireless communication, intelligent manufacturing, and intelligent robotics, target localization technology is playing an increasingly important role in applications such as indoor navigation, logistics handling, precision manufacturing, and human-computer interaction. Efficient and accurate localization methods are one of the key technologies for achieving intelligent operations.

[0003] Among existing wireless positioning technologies, channel state information (CSI)-based positioning methods have received widespread attention. CSI reflects the wireless channel characteristics between a target and a base station or access point. Using CSI data, the approximate location of a target can be predicted through time of arrival (TOA), angle of arrival (AGA), or machine learning algorithms. CSI positioning offers advantages such as wide coverage, flexible deployment, and low cost, making it suitable for rapid target detection over large areas. However, due to factors such as multipath propagation, non-line-of-sight effects, and dynamic interference in the wireless environment, CSI positioning accuracy typically only reaches the decimeter or even meter level, making it difficult to meet the needs of applications requiring high-precision positioning.

[0004] In the field of optical positioning, CCD-based visual measurement methods can achieve high positioning accuracy. In particular, Y-type CCD positioning sensors, composed of linear CCD arrays and cylindrical lenses, work by using linear CCD arrays in different directions to acquire the target's optical signals. After signal processing, the precise position of the target on a two-dimensional coordinate plane can be obtained. Y-type CCD sensors have the advantages of high resolution, fast response speed, and high positioning accuracy, achieving millimeter-level positioning accuracy in the detection of static or low-speed moving targets. However, the measurement range of Y-type CCDs is limited by their field of view, making it difficult to quickly locate targets in large environments, thus limiting their efficiency in practical applications.

[0005] In summary, while a single CSI positioning method can achieve wide coverage, its accuracy is insufficient; a single Y-type CCD positioning method can achieve high accuracy, but it struggles to quickly search for targets over large areas. Therefore, existing technologies cannot simultaneously meet the dual requirements of "wide coverage" and "high accuracy." To address this issue, there is an urgent need to propose a positioning method and device that combines coarse CSI positioning with fine Y-type CCD correction, balancing speed and high accuracy to meet the target positioning needs in complex environments. Summary of the Invention

[0006] The purpose of this invention is to address the problems of insufficient CSI positioning accuracy and limited CCD positioning range in the prior art by proposing a target positioning method and device that combines CSI coarse positioning with Y-type CCD fine correction, so as to take into account both wide coverage and high-precision positioning requirements.

[0007] To achieve the above objectives, this invention provides a target localization method using CSI coarse localization and Y-type CCD fine correction, the method comprising the following steps:

[0008] Step S1: Deploy CSI acquisition equipment in the target area to ensure signal coverage and acquire CSI data between the transmitting and receiving ends;

[0009] Step S2: Perform geometry and coordinate system correction for the Y-type CCD sensor to complete coordinate reconstruction;

[0010] Step S3: Preprocess the collected CSI data to remove noise and normalize the input;

[0011] Step S4: Extract the temporal features of CSI using multi-scale temporal convolutional blocks stacked with temporal branches;

[0012] Step S5: Extract the frequency domain features of CSI through the spectral selection attention module of the frequency domain branch and the multi-scale convolutional block;

[0013] Step S6: The time domain and frequency domain features are fused using a gated collaborative fusion module, and the approximate position coordinates of the target are output.

[0014] Step S7: Control the Y-type CCD sensor to capture the coordinate information of the point light source in the target area;

[0015] Step S8: Correct the error of the data collected by the Y-type CCD sensor to obtain the precise position coordinates of the target.

[0016] Preferably, in step S2, coordinate reconstruction includes the following sub-steps:

[0017] Step S2.1: The geometric model is derived by using the imaging relationship between the three linear CCDs and the cylindrical lens, and the least squares method is used to solve the model coefficients to reduce calibration error.

[0018] Step S2.2: Establish the transformation relationship between the body coordinate system, the cylindrical mirror coordinate system and the geodetic coordinate system to achieve coordinate alignment and unification.

[0019] Preferably, in step S3, the preprocessing includes the following sub-steps:

[0020] Step S3.1: Calculate the amplitude based on the real and imaginary parts of the subcarrier;

[0021] Step S3.2 involves smoothing the CSI amplitude sequence using Gaussian filtering to improve the signal-to-noise ratio. Specifically, ,in, It is the output after Gaussian filtering. It is the original input signal. It is the offset within the current window. The weights of the Gaussian filter are defined as follows: ,in, It is the standard deviation of the Gaussian filter;

[0022] Step S3.3: Perform Min-Max normalization on the denoised data. This enhances the comparability of features.

[0023] Preferably, in step 4, the multi-scale time-aware module of the time-domain branch includes the following sub-steps:

[0024] Step S4.1: Linearly project the preprocessed CSI data onto a unified channel, using the following formula: .in, Indicates batch size. , , These are the number of antennas, the number of subcarriers, and the number of time frames, respectively.

[0025] Step S4.2 involves extracting hierarchical temporal features through multi-scale temporal convolutional blocks and enhancing feature propagation through residual connections. Each multi-scale temporal convolutional block contains multiple parallel branches, each branch containing two identical convolutional layers, with different branches using different convolutional kernels. The first convolutional block Each branch can be expressed as: The multi-scale features of the convolutional block are obtained by averaging and fusing the outputs of multiple branches. The final output of each multi-scale temporal convolutional block is: ;

[0026] Step S4.3 involves stacking multiple convolutional blocks layer by layer to expand the receptive field, and finally extracting temporal features through global average pooling. The dilation rate of each convolutional block is... Increasing layer by layer, This is the stacking layer index. The final time characteristics of the time-domain branch output are: .

[0027] Preferably, in step S5, the spectrum selection attention module of the frequency domain branch includes the following sub-steps:

[0028] Step S5.1: Perform a real-number fast Fourier transform on the time-domain signal to obtain the amplitude spectrum, as shown in the formula: 0. Wherein, The length of the frequency domain;

[0029] Step S5.2: Obtain the initial frequency domain features through convolution mapping, using the following formula: ;

[0030] Step S5.3 introduces a spectral attention mechanism to enhance key frequency information. First, global average pooling is performed along the frequency dimension. Channel weights are then generated through fully connected operations. ,in, , , The compression ratio is... This is the Sigmoid function. Finally, the channel weights are applied to the original features to obtain the channel-enhanced feature representation: .in, This represents the element-wise weighted sum of the channel dimensions.

[0031] Step S5.4 involves further extracting higher-order information from the frequency domain using a multi-scale convolutional block, which is then fused to obtain the final spectral representation. This convolutional block contains multiple parallel convolutional branches, each using convolutional kernels of different scales. Each branch also contains two layers of one-dimensional convolutional operations with ReLU activation and Dropout. Features from each branch are fused using an element-wise averaging method: The frequency domain characteristics of the final output from the frequency domain branch are as follows: .

[0032] Preferably, step S6 includes the following sub-steps:

[0033] Step S6.1, splicing temporal features and frequency domain features Forming joint features ;

[0034] Step S6.2: Construct a gating network to generate dynamic fusion weights. ,in, , , Used to output dynamic gating coefficients Based on the gating weights, feature weighting and fusion are performed, resulting in the following fused features: ;

[0035] Step S6.3 introduces residual connections and normalization operations to improve the stability and generalization of the fused features. The final feature representation is as follows: ;

[0036] Step S6.4: Input the fusion result into the discrimination layer to obtain the approximate location coordinates of the target.

[0037] Preferably, in step S8, a filtering algorithm is used to suppress dynamic errors caused by target or sensor jitter, so as to smooth the jitter and improve coordinate stability, and obtain the corrected accurate target position coordinates.

[0038] The present invention also provides a target positioning device for CSI coarse positioning and Y-type CCD fine correction, the device comprising:

[0039] Module M1 is used to collect CSI data between the transmitter and receiver to reflect the impact of the target on the wireless signal propagation path;

[0040] Module M2 is used to process CSI data to obtain the approximate location coordinates of the target. The processing includes data preprocessing, time-domain feature extraction, frequency-domain feature extraction and feature fusion, and outputs the approximate location coordinates.

[0041] Module M3 is used to control the Y-type CCD sensor to acquire optical data of the target area;

[0042] Module M4 is used to perform dynamic error compensation on the data acquired by the CCD to suppress fluctuations caused by target motion or sensor jitter, and output stable and accurate position coordinates.

[0043] Preferably, the module M2 includes:

[0044] Submodule M2.1 is used to preprocess CSI data;

[0045] Submodule M2.2 is used to extract the time-domain features of CSI data;

[0046] Submodule M2.3 is used to extract the frequency domain features of CSI data;

[0047] Submodule M2.4 is used to fuse the time-domain features and frequency-domain features and output the approximate location coordinates of the target.

[0048] This invention provides a target positioning method and apparatus that combines CSI coarse positioning with Y-type CCD fine correction, which has the following beneficial effects:

[0049] 1. Achieving a balance between wide coverage and high precision: By constructing a two-level architecture of "CSI coarse positioning + Y-type CCD fine correction", the technical contradiction between insufficient accuracy of single wireless positioning (CSI) (usually at the decimeter level) and limited search range of single optical positioning (CCD) is effectively resolved, enabling the system to quickly lock onto targets in a wide range of scenarios and obtain high-precision positioning results at the millimeter level.

[0050] 2. Significantly improved positioning efficiency: The CSI coarse positioning results provide precise guidance for the Y-type CCD, eliminating the need for the optical sensor to perform ineffective scanning across the entire scene; it only needs to focus on a local area for measurement. This "on-demand acquisition" working mode greatly shortens the response time of high-precision positioning and reduces system energy consumption.

[0051] 3. Enhanced robustness in complex environments: In the CSI processing stage, a time-frequency dual-branch feature extraction and gating fusion mechanism is adopted, which can adaptively integrate information from different dimensions, effectively suppress the impact of wireless environmental noise such as multipath effect and non-line-of-sight interference on the positioning results, and improve the stability and reliability of the coarse positioning stage.

[0052] 4. Improved data processing effectiveness: Through multi-scale convolution and spectral attention mechanisms, key location-related features in CSI signals can be accurately captured, avoiding information redundancy or insufficient feature extraction in traditional methods, and providing a more accurate initial range for subsequent optical precision positioning.

[0053] 5. Ensures the stability of precise positioning results: The introduction of a dynamic error compensation mechanism (such as multiple filtering algorithms) in the optical measurement process can effectively suppress position fluctuations caused by the target's own movement or slight sensor jitter, ensuring the continuity and smoothness of the final output precise position coordinates.

[0054] 6. Excellent adaptability and scalability: The device adopts a modular design, and the CSI processing module and Y-type CCD module can be configured and upgraded according to actual needs. Meanwhile, the error compensation module supports switching between multiple filtering algorithms, enabling flexible adaptation to different application scenarios such as low-speed personnel movement and high-speed robot navigation. Attached Figure Description

[0055] 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 drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a schematic diagram of the method flow of the present invention.

[0057] Figure 2 This is a schematic diagram of the system framework of the present invention.

[0058] Figure 3 This is a schematic diagram of the CSI processing network structure in this invention.

[0059] Figure 4This is a schematic diagram of the multi-scale convolutional block structure in this invention.

[0060] Figure 5 This is a schematic diagram of the Y-type CCD sensor structure in this invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0062] Example 1: Standard Configuration Implementation

[0063] like Figure 1-5 As shown, this embodiment provides a target localization method based on CSI coarse localization and Y-type CCD fine correction, including the following steps:

[0064] Step S1: CSI Acquisition Equipment Deployment: Deploy at least one base station within the positioning area for signal interaction with the target and CSI data acquisition. This embodiment uses a general-purpose Wi-Fi communication module as the CSI acquisition equipment to ensure low system cost and ease of deployment.

[0065] Step S2: Y-type CCD Calibration and Coordinate Correction: During the system initialization phase, the Y-type CCD sensor is geometrically calibrated. An optical geometric model is established using the imaging relationship between the three linear CCDs and the cylindrical lens, and the least squares method is used for parameter fitting to reduce calibration errors. Furthermore, the transformation relationship between the CCD body coordinate system, the mirror coordinate system, and the geodetic coordinate system is established to complete coordinate alignment, providing a unified reference benchmark for subsequent fine measurements.

[0066] Step S3: CSI Data Preprocessing: The acquired CSI signals are preprocessed. First, the CSI amplitude sequence is calculated based on the real and imaginary parts of the subcarriers; second, a Gaussian filter is used to smooth the amplitude sequence to suppress instantaneous high-frequency noise interference; finally, the Min-Max normalization method is used to map the data to a uniform range to enhance the comparability between different samples.

[0067] Step S4: Temporal Feature Extraction: The preprocessed CSI signal is input into the temporal branch, and feature representations at different time scales are extracted using multi-scale temporal convolutional blocks. Through the multi-branch convolutional structure and residual connection mechanism, joint modeling of short-term fluctuations and long-term dependencies is achieved.

[0068] Step S5: Frequency Domain Feature Extraction: Perform a Fast Fourier Transform on the preprocessed CSI signal to obtain a spectral representation; then, use convolution operations to extract initial frequency domain features and introduce a spectral selection attention mechanism to enhance key frequency components. Finally, further extract higher-order spectral features through multi-scale convolutional blocks to obtain a frequency domain feature representation.

[0069] Step S6: Feature Fusion and Coarse Localization: The temporal and frequency domain features are concatenated and input into the gated fusion module to generate dynamic fusion weights, achieving adaptive weighted integration of features. The fused features are processed by the discriminant layer to output the approximate location coordinates of the target, with a localization accuracy of up to decimeter level.

[0070] Step S7: CCD Optical Acquisition: Guided by the CSI coarse positioning results, the Y-type CCD sensor is adjusted to the target area, and the two-dimensional coordinate data of the point light source is acquired. Due to the high resolution and fast response characteristics of the CCD, high-precision measurement can be achieved in a local area.

[0071] Step S8: Dynamic Error Compensation and Fine Positioning: To address position fluctuations caused by target movement or sensor jitter during CCD acquisition, a smoothing filtering algorithm is used to suppress dynamic errors, resulting in stable optical measurement results. Finally, the coarse positioning results are combined with the CCD fine-correction data to obtain the precise position coordinates of the target.

[0072] like Figure 2 As shown, the corresponding device, namely a target positioning device for CSI coarse positioning and Y-type CCD fine correction, includes:

[0073] Module M1: CSI data acquisition module;

[0074] Module M2: CSI data processing module (including preprocessing submodule, time-domain feature submodule, frequency-domain feature submodule, and feature fusion submodule);

[0075] Module M3: Y-type CCD optical acquisition module;

[0076] Module M4: Dynamic error compensation module.

[0077] Through the above steps and apparatus, this embodiment can achieve an organic combination of large-scale rapid coarse positioning and local high-precision fine correction while maintaining low system cost and easy deployment. The positioning accuracy is improved from the decimeter level of the traditional CSI method to the millimeter level, thereby verifying the basic effectiveness of the present invention.

[0078] This application proposes a target positioning method and device that combines coarse wireless signal positioning with fine optical measurement. Its core working principle can be summarized as a two-level progressive positioning strategy of "coarse positioning guidance and fine positioning correction".

[0079] Phase 1: Wide-area coarse localization based on CSI

[0080] The system first collects Channel State Information (CSI) data using devices such as Wi-Fi base stations deployed in the target area. Since the movement of the target along the wireless signal propagation path affects the amplitude and phase of the CSI data, the target's location information is implicitly contained within the CSI data. To accurately calculate the approximate location of the target from the complex wireless signal, this application designs a dedicated CSI data processing flow:

[0081] Data preprocessing: The raw CSI data is filtered, denoised, and normalized to suppress multipath interference and noise in the environment.

[0082] Dual-branch feature extraction:

[0083] Temporal branch: By stacking multi-scale temporal convolutional blocks, hierarchical features of CSI signals in the time dimension are extracted to capture the temporal changes of signals caused by target motion.

[0084] Frequency domain branch: The spectrum is obtained by performing a fast Fourier transform on the CSI signal, and a spectrum attention mechanism and multi-scale convolution are introduced to enhance the perception of key frequency components in order to distinguish the unique impact of different locations on the channel.

[0085] Gated fusion and output: A gated fusion module is used to dynamically calculate the fusion weights of time-domain and frequency-domain features based on data characteristics, so as to achieve adaptive information integration and finally output the approximate location coordinates of the target within the coverage area.

[0086] Phase Two: Local Fine-tuning Based on Y-type CCD

[0087] After obtaining the approximate location of the target, the system controls a Y-type CCD sensor to perform targeted optical acquisition of that local area. The Y-type CCD consists of three linear CCD arrays combined with cylindrical lenses, offering advantages such as high resolution and fast response speed. After acquiring the point light source coordinate information, the system combines the results of previous geometric calibration and coordinate system correction of the Y-type CCD to perform error compensation on the optical data (such as using Kalman filtering algorithms to suppress motion jitter), ultimately outputting the precise position coordinates of the target.

[0088] The entire workflow enables rapid searching of "areas" and precise measurement of "points," organically combining the wide-area coverage capability of wireless positioning with the high-precision characteristics of optical positioning.

[0089] Example 2: Indoor Personnel Positioning

[0090] In this embodiment, the method of the present invention is applied to a real-time positioning scenario of indoor personnel.

[0091] 1. System Deployment: In a 10 m × 10 m laboratory, a Wi-Fi base station and receiver were deployed as CSI data acquisition devices, and a Y-type CCD sensor was installed in the center of the ceiling. The experimental subjects were people moving freely indoors, whose bodies disturbed the wireless signal propagation path, thus affecting the CSI data collected by the receiver.

[0092] 2. CSI Coarse Positioning: While personnel move within the area, continuous wireless communication occurs between the Wi-Fi base station and the receiver. The receiver extracts CSI data when parsing the received signal frames. This data is first preprocessed (including subcarrier amplitude calculation, Gaussian filtering, and Min-Max normalization), then input into the time-domain and frequency-domain branches respectively to extract multi-scale features. Finally, the approximate location coordinates of the personnel are output through a gated fusion module. In one optional implementation, the multi-scale convolutional blocks in the time-domain branch employ a three-branch parallel structure, with kernel sizes of 3, 9, and 15 for each branch, to simultaneously capture short-term fluctuations and long-term dependency features. The frequency-domain branch can also adopt the same multi-scale design to enhance the modeling capability of key frequency components. Through the above design, this embodiment can achieve an average positioning accuracy of approximately 0.6 m, thereby completing fast and reliable coarse positioning over a large area and providing a reference for subsequent fine measurements.

[0093] 3. CCD Optical Acquisition: Guided by the coarse positioning results from CSI, a Y-type CCD sensor is controlled to perform optical imaging of the target area, extracting the two-dimensional coordinate information of the personnel reflection markers. Due to the high resolution and fast response characteristics of CCD, millimeter-level positioning capability can be achieved within a local area of ​​the target.

[0094] 4. Coordinate mapping and registration: The geometric calibration and coordinate system correction of the Y-type CCD have been completed during the system initialization phase. Therefore, after optical acquisition, the CCD measurement results are directly mapped to a unified coordinate system using the calibration model to achieve coarse and fine positioning.

[0095] 5. Dynamic error compensation: To address position fluctuations caused by slight personnel movement or sensor jitter, a smoothing filtering algorithm is used to dynamically suppress errors in CCD measurement results, thereby outputting continuous and stable target coordinates.

[0096] Experimental results: Comparative experiments show that when using only the CSI method, the average positioning error is about 0.6 m; after adopting the CSI coarse positioning and Y-type CCD fine correction method proposed in this invention, the average error is significantly reduced to the millimeter level, verifying the high precision, stability and real-time advantages of this invention in indoor personnel positioning tasks.

[0097] Example 3: Robot Navigation Application

[0098] In this embodiment, the method of the present invention is applied to the autonomous navigation task of a robot in a warehouse environment.

[0099] 1. System Deployment: In a 20 m × 15 m warehouse environment, Wi-Fi base stations and receivers were deployed as CSI data acquisition devices. Y-type CCD sensors could move around the perimeter of the ceiling. The experimental subject was an autonomous mobile robot with optical reflective markers mounted on its top to enhance the optical acquisition effect of the CCD.

[0100] 2. CSI Coarse Localization: During robot movement, the Wi-Fi base station and receiver continuously interact to acquire CSI signals in real time. The collected data undergoes preprocessing such as amplitude calculation, Gaussian filtering, and normalization before being input into the time-domain and frequency-domain branches for feature extraction. Adaptive integration is then achieved in the gated fusion module to output the robot's approximate position coordinates. Specifically, the time-domain branch stacks five multi-scale convolutional blocks, each employing a four-branch parallel structure with kernel sizes set to 3, 9, 17, and 33 respectively to cover feature information at different time scales. The frequency-domain branch uses the same design for its multi-scale convolutional blocks to fully exploit the hierarchical features of the spectral domain. Through this configuration, this stage enables rapid robot positioning in a wide-ranging environment, effectively controlling the positioning error within 0.8–1.2m, providing a reliable initial reference for subsequent fine CCD measurements.

[0101] 3. Navigation and CCD Triggering: After obtaining the approximate range of the robot's position through CSI coarse positioning, the control system guides the Y-type CCD sensor to collect optical reflection markers on the top of the robot, thereby achieving high-precision fine positioning in a local area.

[0102] 4. Coordinate Correction: During the system deployment phase, the geometric calibration and coordinate system transformation of the Y-type CCD have been completed. Therefore, after acquiring optical data, the existing model and transformation relationship can be directly used to map the point light source coordinates measured by the CCD to a unified coordinate system, thereby obtaining a more accurate two-dimensional position.

[0103] 5. Dynamic Compensation: Due to mechanical vibrations that easily occur during robot movement, CCD measurement results may fluctuate. This embodiment employs a Kalman filter method to dynamically compensate for optical acquisition results, smoothing out jitter and improving the stability of the output results.

[0104] Experimental Results: The experimental results show that this embodiment can achieve positioning accuracy from centimeters to millimeters in robot navigation scenarios, giving full play to the advantages of CSI in large-area coverage and the complementary characteristics of CCD in local high-precision measurement, thereby significantly improving the reliability and accuracy of robot autonomous navigation.

[0105] Example 4: Modular Device Structure

[0106] In this embodiment, the present invention optimizes the structure of the target positioning device and proposes a modular implementation scheme to improve the system's scalability, maintainability, and application value.

[0107] 1. Overall Structure of the Device

[0108] like Figure 2 As shown, this device consists of four main modules:

[0109] Module M1: CSI data acquisition module;

[0110] Module M2: CSI data processing module;

[0111] Module M3: Y-type CCD optical acquisition module;

[0112] Module M4: Dynamic error compensation module.

[0113] Each module is connected through a standardized interface and can be independently upgraded, expanded, or replaced to adapt to the performance requirements of different application scenarios.

[0114] 2. CSI Data Acquisition Module (M1)

[0115] Module M1 consists of a Wi-Fi base station and a receiver, used for real-time acquisition of CSI data. This module employs a hardware abstraction layer design, enabling different Wi-Fi chip models to extract CSI data through a unified driver, thereby enhancing system compatibility and portability.

[0116] 3. CSI Data Processing Module (M2)

[0117] Module M2 performs end-to-end processing of CSI data, and is further subdivided internally into:

[0118] M2.1 Preprocessing Submodule: Performs amplitude calculation, noise filtering, and normalization;

[0119] M2.2 Temporal Feature Extraction Submodule: Extracts hierarchical temporal features based on multi-scale temporal convolution;

[0120] M2.3 Frequency Domain Feature Extraction Submodule: Obtains frequency domain features through Fast Fourier Transform and Spectral Attention Mechanism;

[0121] The M2.4 feature fusion and coarse localization submodule utilizes a gated fusion mechanism to achieve adaptive feature integration and output approximate location coordinates.

[0122] The hierarchical submodule design facilitates algorithm optimization and upgrades, and can be tailored according to hardware resources, thus balancing performance and cost.

[0123] 4. Y-type CCD optical acquisition module (M3)

[0124] Module M3 is responsible for optical measurements of the target area. Its core structure consists of an optical lens, a three-line CCD array, and cylindrical lenses. Based on the coarse positioning results from CSI, the control unit can dynamically adjust the acquisition range of the CCD, allowing it to operate only within a local area, thereby reducing energy consumption and improving efficiency.

[0125] 5. Dynamic Error Compensation Module (M4)

[0126] Module M4 dynamically corrects CCD measurement results to suppress fluctuations caused by target motion, sensor jitter, or installation errors. Multiple methods can be used in different scenarios:

[0127] In low-speed scenarios: moving average filtering achieves fast smoothing;

[0128] High-speed scenarios: Dynamic compensation is performed using Kalman filtering or extended Kalman filtering;

[0129] Interference scenarios: Particle filtering enhances robustness.

[0130] Finally, M4 maps the corrected optical results to a unified coordinate system and outputs high-precision two-dimensional coordinates.

[0131] 6. System Advantages and Effects

[0132] The modular design in this embodiment achieves software and hardware decoupling and flexible expansion, and has the following advantages:

[0133] Maintainability: Upgrading the CSI or CCD module will not affect the overall system operation;

[0134] Scalability: CCD modules can be added as needed to expand the coverage area;

[0135] Applicability: It can be used for indoor personnel positioning, as well as robot navigation, industrial workpiece measurement and other scenarios.

[0136] Experimental results show that, using the modular device of this embodiment, the system can still achieve millimeter-level high-precision positioning while maintaining low cost and easy deployment.

[0137] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A target localization method combining CSI coarse localization and Y-type CCD fine correction, characterized in that, Includes the following steps: S1 deploys CSI acquisition equipment in the target area and acquires CSI data between the transmitting and receiving ends; S2 performs geometric calibration and coordinate system correction on the Y-type CCD sensor to complete coordinate reconstruction; S3 preprocesses the CSI data to remove noise and normalize the input; S4 uses time-domain branching to extract time-domain features of CSI; S5 uses frequency domain branching to extract the frequency domain features of CSI; S6 uses a gated fusion module to fuse time-domain and frequency-domain features and outputs the approximate location coordinates of the target. S7 controls the Y-type CCD sensor to perform optical acquisition of the target area and obtain the coordinate information of the point light source; S8 performs error correction on the data acquired by the Y-type CCD and outputs the precise position coordinates of the target.

2. The method according to claim 1, characterized in that, The preprocessing of S3 includes: S3.1 Calculate the amplitude of the CSI subcarrier; S3.2 A filtering method is used to smooth the CSI amplitude sequence in order to suppress noise and improve the signal-to-noise ratio; S3.3 Normalizes the denoised data to enhance feature comparability.

3. The method according to claim 1, characterized in that, The time-domain branch of S4 includes: S4.1 maps CSI data to a unified channel; S4.2 utilizes multi-scale temporal convolutional blocks to extract hierarchical temporal features and enhances feature propagation through residual connections; S4.3 stacks multiple convolutional blocks to expand the receptive field layer by layer, and finally obtains temporal features through pooling operations.

4. The method according to claim 1, characterized in that, The frequency domain branch of S5 includes: S5.1 performs a Fast Fourier Transform on the CSI signal to obtain its spectral representation; S5.2 obtains the initial frequency domain features through convolution mapping; S5.3 utilizes a spectrum attention mechanism to enhance key frequency information; S5.4 extracts high-order spectral features through multi-scale convolutional blocks to obtain the final frequency domain features.

5. The method according to claim 1, characterized in that, The gated fusion module in S6 includes: S6.1 The time-domain features and frequency-domain features are concatenated to form a joint feature; S6.2 Utilize a gating network to generate dynamic fusion weights, and perform weighted fusion on the joint features; S6.3 introduces residual connection and normalization operations to improve the stability and generalization of fused features; S6.4 inputs the fusion result into the discriminant layer and outputs the approximate location coordinates of the target.

6. The method according to claim 1, characterized in that, The error correction of S8 includes: using a filtering algorithm to compensate for dynamic errors caused by target motion or sensor jitter.

7. A target positioning device combining CSI coarse positioning and Y-type CCD fine correction, characterized in that, include: Module M1 is used to collect CSI data between the transmitter and receiver to reflect the impact of the target on the wireless signal propagation path; Module M2 is used to process the CSI data and output the approximate location coordinates of the target; Module M3 is used to control the Y-type CCD sensor to acquire optical data of the target area; Module M4 is used to perform error compensation and correction on the optical data and output the precise position coordinates of the target.

8. The apparatus according to claim 7, characterized in that, The module M2 includes: The M2.1 preprocessing submodule is used to perform CSI amplitude calculation, noise filtering, and normalization. The M2.2 temporal feature extraction submodule is used to extract temporal features based on multi-scale convolutional blocks; The M2.3 frequency domain feature extraction submodule is used to extract frequency domain features through spectral attention mechanism and multi-scale convolutional blocks; The M2.4 feature fusion submodule is used to integrate multimodal features using a gated fusion mechanism and output coarse localization results.

9. The apparatus according to claim 7, characterized in that, The module M3 includes a combination structure of three linear CCDs and cylindrical lenses. The module dynamically adjusts the acquisition area according to the CSI coarse positioning results to reduce invalid scans and improve processing efficiency.

10. The apparatus according to claim 7, characterized in that, The module M4 supports switching between multiple filtering algorithms, including moving average filtering, Kalman filtering, and particle filtering, to adapt to different target velocities and environmental interference conditions.

Images