Progressive correction method and system for positioning deviation in indoor Wi-Fi fingerprint positioning technology

Abstract

The invention provides a progressive correction method and system for positioning deviation in an indoor Wi-Fi fingerprint positioning technology, and relates to the field of sensor signal processing, and the method comprises the steps: obtaining offline Wi-Fi signal collection data, extracting Wi-Fi fingerprint information from the collection data, and forming an offline information library; the method comprises the following steps: acquiring online Wi-Fi signal acquisition data, and positioning an online acquisition point according to an offline information base to obtain a preliminary positioning result; according to the preliminary positioning result, an acquisition point space relation graph is formed in combination with an offline information base; and according to the spatial relation graph, carrying out secondary positioning on the online acquisition points to obtain a secondary positioning result. According to the invention, through efficient and modularized positioning error correction, high-precision positioning based on indoor Wi-Fi fingerprint information is realized, so that the accuracy and reliability of a positioning result are improved.

Description

Technical Field

[0001] This invention relates to the field of sensor data processing technology, and in particular to a progressive correction method and system for positioning deviation in indoor Wi-Fi fingerprint positioning technology. Background Technology

[0002] In the field of indoor positioning, WiFi fingerprint positioning technology has become one of the most widely used positioning methods due to its advantages such as requiring no additional hardware deployment and low cost. However, the indoor environment is complex and dynamically changing, with many interference factors. Signals are susceptible to multipath effects during propagation, with signals from different paths superimposing each other, causing significant fluctuations in the signal strength at the receiving end; people walking and objects moving can also significantly alter the signal propagation path, further degrading signal stability.

[0003] When traditional machine learning / deep learning algorithms are applied to WiFi fingerprint positioning, they can uncover the relationship between signal features and location to some extent. However, because they do not fully consider the spatial relationship between collection points and the complex topology of the indoor environment, they are unable to effectively deal with these interferences, resulting in low positioning accuracy and poor stability, which cannot meet the growing demand for high-precision indoor positioning. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology, so as to improve positioning accuracy and enhance user experience.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: In a first aspect, a progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology, the method comprising: Acquire offline Wi-Fi signal data, extract Wi-Fi fingerprint information from the acquired data, and form an offline information database; Acquire online Wi-Fi signal data, locate the online collection points based on the offline information database, and obtain preliminary location results; Based on the preliminary location results and combined with the offline information database, a spatial relationship map of the collection points is generated; Based on the spatial relationship diagram, the online collection points are repositioned to obtain the repositioning results; Based on the secondary positioning results, the spatial relationship map of the collection points is updated, and a multi-source relationship map is formed by combining the signal characteristics of offline and online data. Based on the multi-source relationship diagram, the final positioning is performed, and the positioning deviation is progressively corrected. Furthermore, offline Wi-Fi signal collection data is acquired, and Wi-Fi fingerprint information is extracted from the collected data to form an offline information database, including: The Wi-Fi signal in the target area is collected from multiple points to obtain raw offline data.

[0006] Based on the raw data, obtain Wi-Fi signal source information and distinguish different signal sources according to BSSID; Obtain the signal power of each source, using decibels and milliwatts as the unit. Sources where no valid signal was detected are marked with special values. Obtain the location of each offline collection point and acquire the corresponding collection data; Based on the location of the collection points and the corresponding collection information, an offline database for Wi-Fi fingerprint positioning is established using structured storage.

[0007] Furthermore, online Wi-Fi signal data is acquired, and based on the offline database, the online acquisition points are located to obtain preliminary location results, including: Based on the characteristics of the acquired signals from the offline data, sample data is added to the localization model; Obtain online data collected by users of location services; Calculate the feature similarity between online collected data and offline data; Candidate offline collection points are selected based on similarity, and online collection locations are calculated.

[0008] Furthermore, based on the preliminary location results and combined with the offline information database, a spatial relationship map of the collection points is generated, including: Based on the preliminary positioning results of the online acquisition points, the characteristics of the online acquisition signals and the positioning results are combined with the offline data to form basic spatial relationship data; In spatial relationship data, the Euclidean distance from each collection point to other collection points is calculated as the spatial distance.

[0009] Based on spatial distance, candidate neighbor collection points are selected to form a local spatial relationship map for each collection point; Integrate all local spatial relationship diagrams to form an offline spatial relationship diagram containing only offline collection points and an overall spatial relationship diagram containing both online and offline collection points. The former can be completed in advance, while the latter needs to be generated based on user data during actual use.

[0010] Furthermore, based on the spatial relationship diagram, the online data collection points are repositioned to obtain secondary positioning results, including: The localization model is trained based on the offline spatial relationship graph; Based on the positioning results, the positioning results of the online collection points are extracted to form secondary positioning results.

[0011] Furthermore, based on the secondary positioning results, the spatial relationship map of the collection points is updated, and combined with the signal characteristics of offline and online data, a multi-source relationship map is formed, including: Based on the secondary positioning results, update the positions of the online collection points in the spatial relationship map and recalculate the local spatial relationships; Update the overall spatial relationship diagram based on the updated local spatial relationships; Based on the calculated feature similarity, candidate neighbor collection points are selected to form a local feature relationship graph for each collection point; By combining all local feature relationship graphs, an offline feature relationship graph containing only offline collection points and an overall feature relationship graph containing both online and offline collection points are formed. The former can be completed in advance, while the latter needs to be generated based on user data during actual use. By combining spatial relationship diagrams and feature relationship diagrams, a multi-source relationship diagram is formed, which includes offline relationship diagrams and overall relationship diagrams.

[0012] Furthermore, based on the multi-source relationship diagram, final positioning is performed to complete the progressive correction of positioning errors, including: The localization model is trained based on the offline multi-source relationship graph to obtain the trained model; Based on the trained model, input the overall multi-source relationship graph, and finally locate all the collection points in the graph to obtain the location results; Based on the positioning results, the positioning results of the online collection points are extracted to form the final positioning results.

[0013] Secondly, a progressive correction system for positioning deviation in indoor Wi-Fi fingerprint positioning technology includes: The acquisition module is used to acquire Wi-Fi signal data and extract the Wi-Fi source BSSID and signal power from the signal data to form Wi-Fi fingerprint data; based on the Wi-Fi fingerprint data and the location information of offline acquisition points, an offline fingerprint database is formed; based on the offline fingerprint database and the online fingerprint data, an acquisition point feature relationship diagram is formed, including an offline feature relationship diagram and an overall feature relationship diagram; based on the initial and secondary positioning results, an acquisition point spatial relationship diagram is formed, including an offline spatial relationship diagram and an overall spatial relationship diagram.

[0014] Processing module: Based on the offline information database, it locates the online collection points and obtains preliminary location results; based on the spatial relationship diagram, it performs secondary location of the online collection points to obtain secondary location results; based on the multi-source relationship diagram, it performs final location and completes the progressive correction of location deviation. Thirdly, a computing device includes: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0015] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0016] The above-described solution of the present invention has at least the following beneficial effects: By acquiring offline Wi-Fi signal data and extracting fingerprint information to form an offline database, a reliable data foundation is provided for subsequent positioning, ensuring the stability and integrity of the data source. After acquiring online Wi-Fi signal data, preliminary positioning results are obtained based on the offline database, enabling rapid initial positioning. A spatial relationship map of the acquisition points is formed based on the preliminary positioning results and the offline database, providing a spatial structure basis for secondary positioning and improving its rationality. Secondary positioning is performed based on the spatial relationship map, further optimizing the positioning results. The spatial relationship map of the acquisition points is updated based on the secondary positioning results, and a multi-source relationship map is formed by combining offline and online data signal characteristics, fully utilizing multi-source data information and enhancing the richness and relevance of the data. Finally, final positioning is performed based on the multi-source relationship map, completing a progressive correction of positioning deviations, significantly improving positioning accuracy, and providing users with more accurate and reliable indoor positioning services. Attached Figure Description

[0017] Figure 1 This is an example of the dataset UJIIndoorLoc used to illustrate the present invention, where blue dots represent the collection points for training data and red dots represent the collection points for test data.

[0018] Figure 2 The satellite image corresponding to the data collection location for this dataset shows it is located inside the University of Jaime I in Spain. The dataset contains 19,937 training data points and 1,111 test data points; it comprises three buildings, the tallest of which has five floors. Figure 1 The system displays 3D views of all data collection points, with the horizontal coordinates based on the WGS84 datum and the UTM projection plane coordinate system, divided into north and east coordinates, in meters, and the height coordinates representing the floor numbers.

[0019] Figure 3 This is a second illustration of the UTSIndoorLoc dataset used to illustrate the present invention, where blue dots represent the collection points for training data and red dots represent the collection points for test data.

[0020] Figure 4The satellite image corresponding to the data collection location for this dataset shows that the building belongs to the University of Technology Sydney, Australia. The dataset contains 9108 training data points and 388 test data points, covering 16 floors. Due to the large number of floors, it was difficult to display a 3D view.

[0021] Figure 5 This is a flowchart illustrating a progressive correction method for positioning deviation in an indoor Wi-Fi fingerprint positioning technology provided by an embodiment of the present invention.

[0022] Figure 6 This is a schematic diagram of a progressive correction system for positioning deviation in an indoor Wi-Fi fingerprint positioning technology provided by an embodiment of the present invention. Detailed Implementation

[0023] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0024] like Figure 5 As shown, an embodiment of the present invention proposes a progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology, the method comprising the following steps: Step 1: Acquire offline Wi-Fi signal data, extract Wi-Fi fingerprint information from the acquired data, and form an offline information database; Step 2: Obtain online Wi-Fi signal data, locate the online collection points based on the offline information database, and obtain preliminary location results; Step 3: Based on the preliminary positioning results and combined with the offline information database, a spatial relationship map of the collection points is generated; Step 4: Based on the spatial relationship diagram, perform secondary positioning of the online collection points to obtain the secondary positioning results; Step 5: Based on the secondary positioning results, update the spatial relationship map of the collection points, and combine the signal characteristics of offline and online data to form a multi-source relationship map; Step 6: Based on the multi-source relationship diagram, perform the final positioning and complete the progressive correction of the positioning deviation.

[0025] In this embodiment of the invention, offline Wi-Fi signal acquisition data in an indoor environment is first obtained. This offline acquisition data is typically collected at known locations and includes signal strength information from multiple access points. By preprocessing the offline Wi-Fi signal acquisition data, stable and representative Wi-Fi fingerprint information is extracted and organized and stored according to spatial location to construct an offline information database, providing basic reference data for subsequent positioning calculations. Next, online Wi-Fi signal acquisition data is obtained, i.e., real-time Wi-Fi signal information collected at unknown locations during actual user use. The system matches the online acquisition data with the Wi-Fi fingerprints in the offline information database. Based on fingerprint similarity or distance measurement methods, the online acquisition points are initially located, thus obtaining preliminary positioning results. This preliminary positioning result is used to roughly reflect the spatial location of the online acquisition points.

[0026] After obtaining the initial positioning result, the system constructs a spatial relationship map of the collection points based on this positioning result and the spatial distribution relationship of each fingerprint point in the offline information database. This spatial relationship map describes the spatial adjacency, relative position, and signal association relationships between the online collection point and surrounding offline fingerprint points, providing structured spatial constraint information for further optimization of the positioning result. Subsequently, the system performs secondary positioning of the online collection point according to the constructed spatial relationship map to obtain the secondary positioning result. By introducing spatial relationship constraints, the initial positioning result is corrected, ensuring that the positioning calculation not only relies on a single fingerprint matching result but also comprehensively considers the relative position of the collection point in the overall spatial structure, thereby effectively reducing positioning deviations caused by signal fluctuations or environmental interference and improving positioning accuracy. After completing the secondary positioning, the system dynamically updates the spatial relationship map of the collection points based on the updated positioning result and further combines Wi-Fi signal feature information from offline and online data to construct a multi-source relationship map. This multi-source relationship map comprehensively integrates spatial relationship information with signal feature information from multiple times and sources, used to more comprehensively characterize the association characteristics between collection points. Finally, the system performs the final positioning calculation of the online collection point based on the multi-source relationship map.

[0027] By combining multi-source information constraints and a progressive correction mechanism, the positioning results are further optimized, and the positioning deviation is gradually corrected to obtain a final positioning result with higher accuracy and stronger stability, thereby completing the progressive correction of positioning deviation in the indoor Wi-Fi fingerprint positioning process.

[0028] In a preferred embodiment of the present invention, step 1, acquiring offline Wi-Fi signal collection data, extracting Wi-Fi fingerprint information from the collection data, and forming an offline information database, includes: Step 11: Collect Wi-Fi signals from multiple points in the target area to obtain raw offline data. This includes: first, defining the physical boundaries of the target indoor area (e.g., a floor in an office building, a specific area in a shopping mall); determining the grid spacing (e.g., 1m×1m or 2m×2m) based on positioning accuracy requirements; marking all preset collection points on an electronic map and assigning a unique collection number to each point; preparing several mobile terminals (smartphones, professional collection equipment) with Wi-Fi signal collection capabilities; uniformly calibrating the Wi-Fi signal receiving modules of the terminals to ensure consistent collection parameters (sampling rate, receiving sensitivity) across different terminals; and establishing collection specifications, specifying the collection duration (e.g., continuous collection for 30 seconds), collection frequency (e.g., once per second), and terminal placement (e.g., screen horizontal, 1.2m above the ground) for each collection point to avoid operational discrepancies. This leads to data deviation. The data acquisition personnel, carrying calibrated terminals, arrive at the first preset acquisition point. After confirming that the terminal location completely overlaps with the map marker, the signal acquisition program is started. During the acquisition process, the terminal remains stationary and the antenna is unobstructed. The program records all raw Wi-Fi signal data (including original signal identifier, instantaneous strength value, acquisition timestamp, etc.) in real time. After completing the acquisition at that point, the raw data is named and saved with the acquisition number and timestamp. The process is then repeated at the next acquisition point until all preset points are acquired. The raw data files from all acquisition points are exported to the server. The correspondence between the acquisition numbers and preset points is verified, and invalid data caused by terminal malfunctions or operational errors (such as files with insufficient acquisition time, empty data, or all abnormal signal values) are removed. Finally, a complete raw offline Wi-Fi signal dataset for the target area is formed.

[0029] Step 12: Based on the raw data, obtain Wi-Fi source information and distinguish different sources according to BSSID. Specifically, this includes: based on the raw offline dataset obtained in Step 11, performing structured parsing on the raw data file of each collection point, filtering out the core fields related to Wi-Fi sources, focusing on extracting BSSID (Basic Service Set Identifier), which is the unique physical address of the Wi-Fi access point, and removing redundant information such as terminal system logs and invalid characters; performing BSSID deduplication on the parsing results of a single collection point to obtain a list of BSSIDs of all unique Wi-Fi sources that can be detected at that point; and summarizing the BSSID lists of all collection points. A comprehensive list of Wi-Fi sources in the target area is generated, and a unique source number (such as S001, S002) is assigned to each unique BSSID. A one-to-one correspondence between source number and BSSID is established to clearly distinguish different Wi-Fi sources. The format of the BSSIDs in the comprehensive list is checked to ensure that they conform to the MAC address standard (12 hexadecimal characters, divided into 6 segments), and invalid BSSIDs with incorrect formats are removed. Combined with the Wi-Fi facility ledger of the target area, external sources that are not in the target area (such as router signals penetrating from adjacent floors or outdoors) are marked to facilitate subsequent data filtering. Finally, a correspondence table of collection point - source number - BSSID is generated.

[0030] Step 13: Obtain the signal power of each source, using decibels and milliwatts (dBm) as the unit. Sources without detected valid signals are marked with special values. Specifically, this includes: based on the acquisition point-source number-BSSID correspondence table obtained in Step 12, extracting all raw signal strength values ​​corresponding to each BSSID at each acquisition point and converting them to decibels and milliwatts (dBm); calculating the average of all valid strength values ​​for each BSSID at that acquisition point as the final signal power value for that source at that acquisition point, reducing the impact of signal fluctuations in a single acquisition; traversing the total list of Wi-Fi sources in the target area, if... If the BSSID of a certain signal source does not appear in the parsing results of a certain collection point (i.e., the signal source signal is not detected), then a preset special value is marked for that signal source under that collection point (such as -100dBm, which is lower than the lower limit of the normal Wi-Fi signal strength, easy to distinguish and ensures data dimension consistency); check whether the signal power values ​​of all collection points are within a reasonable range (-100~0dBm), and remove outliers that are outside the range; if the proportion of outliers of a certain signal source at a certain collection point exceeds 50%, then directly mark it with the above special value, and finally form a complete data table of collection point-source number-BSSID-signal power.

[0031] Step 14: Obtain the location of each offline acquisition point and the corresponding acquisition data. Specifically, this includes: extracting the physical location information (such as two-dimensional coordinates with the southwest corner of the target area as the origin, floor number, area identifier, etc.) corresponding to each acquisition number from the pre-acquisition planning file, and organizing it into a structured location data table containing acquisition number-x coordinate-y coordinate-floor-area; matching the acquisition point-source number-BSSID-signal power data table generated in Step 13 with the location data table, using the acquisition number as the core association key; for a single acquisition number, supplementing its corresponding location information into the power data records of all sources under that number, forming a complete data record of acquisition number-location information-source number-BSSID-signal power; randomly selecting 10%-20% of the acquisition point data and manually verifying the consistency of the association between the location information and the power data (such as confirming whether the coordinates corresponding to the acquisition number match the on-site acquisition point); if problems such as misaligned acquisition numbers or incorrect location information are found, tracing back the original acquisition records and correcting them, finally forming an integrated dataset of location and acquisition data for all acquisition points in the target area.

[0032] Step 15: Based on the location of the collection points and the corresponding collection information, establish a Wi-Fi fingerprint positioning offline database using structured storage. Specifically, this includes: designing the core table structure of the offline database based on the integrated location-collection data dataset obtained in Step 14. The core table is named the Wi-Fi fingerprint offline table, and its fields include: collection point ID (primary key), collection number, x-coordinate, y-coordinate, floor, region, source number, BSSID, and signal power. At the same time, auxiliary tables are designed, such as the source information table (storing the source installation location and its operator) and the collection log table (storing the collection personnel, collection time, and terminal number), to facilitate data management and traceability.

[0033] MySQL, PostgreSQL, or other structured database systems were selected and deployed on the server. Data tables were created according to the designed table structure. A data import program was written to batch import the integrated location-collection data dataset into the core table, and auxiliary information into the corresponding auxiliary tables. During the import process, the field format was validated (e.g., coordinates are numeric, BSSID is character). After import, the amount of data imported into the database was checked against the original dataset to ensure no data loss or duplicate import. A regular database backup strategy was configured (daily full backup and hourly incremental backup) to prevent data corruption or loss. Data update rules were established so that when the Wi-Fi facilities in the target area changed (e.g., adding / removing routers), the corresponding area data was re-collected and the database was updated. Hierarchical access permissions were set (administrators can read and write, location service personnel can only read), ultimately forming a stable offline Wi-Fi fingerprint positioning database that can provide fingerprint matching services.

[0034] In this embodiment of the invention, an offline Wi-Fi fingerprint positioning database is constructed based on sensor-collected data. First, multi-point Wi-Fi signal collection is performed within the target area according to preset collection rules, acquiring raw offline data containing multiple collection points. Based on this, the raw offline data is parsed and processed to extract Wi-Fi source-related information, and different Wi-Fi sources are distinguished and marked using the unique identifier (BSSID) of the access point. Simultaneously, the signal power value corresponding to each Wi-Fi source is obtained, with decibels and milliwatts (dW) as the unit of measurement. For valid signals not detected at a certain collection point, a preset special value is used for labeling to ensure data consistency. Furthermore, the system acquires the spatial location information of each offline collection point and associates this location information with the corresponding Wi-Fi signal collection data. Finally, based on the collection point location and its corresponding Wi-Fi source name and signal power information, a structured data storage method is used to establish the Wi-Fi fingerprint positioning offline database, facilitating efficient execution of subsequent fingerprint matching and positioning calculations.

[0035] In a preferred embodiment of the present invention, step 2, acquiring online Wi-Fi signal acquisition data, locating the online acquisition point based on the offline information database, and obtaining preliminary location results, includes: Step 21: Add sample data to the positioning model based on the characteristics of the acquired signals from the offline data. Step 22: Obtain online data collected by the location service user; Step 23: Calculate the feature distance between the online collected data and the offline data using the following formula: ; in Represents the feature vector of online acquisition points With the feature vector of a certain offline collection point The Euclidean distance between them; d To represent a scalar value. Feature vectors of online collection points, It is a feature vector of a certain collection point in the offline database. The online data collection point is at the [number]th [location]. Signal strength values ​​from individual Wi-Fi signal sources; The offline collection point is at the 1st The signal strength value of a Wi-Fi signal source is an index. It represents the total number of signal sources; Step 24: Based on the feature distance, select candidate offline acquisition points and calculate the online acquisition location using the following formula: ; in Three-dimensional spatial coordinates; k This is the number of nearest-neighbor offline data collection points selected; i It is the summation index, indicating the index of the summation index. i The nearest neighbor; x i It is the first i The three-dimensional coordinates of the nearest neighbor offline acquisition point; y i It is the first i The three-dimensional coordinates (latitude and right) of the nearest neighbor offline acquisition point; z i It is the first i The three-dimensional coordinates of the nearest neighbor offline acquisition point. i The three-dimensional coordinates and height of the nearest neighbor offline acquisition point; d i It is an online collection point and the first i The feature distance between the nearest neighbor offline collection points.

[0036] In this embodiment of the invention, the positioning model is fitted based on the constructed Wi-Fi fingerprint positioning offline database, and the positioning model is used to perform initial positioning on the online collected data. First, according to the Wi-Fi signal feature information contained in the offline collected data, the fingerprint data of each offline collection point is added as a sample to the positioning model to describe the mapping relationship between different spatial locations and their corresponding signal features, thereby completing the training and construction of the positioning model.

[0037] Subsequently, online Wi-Fi signal data collected by the user during actual use of the location service is acquired. This online data includes signal power information from multiple Wi-Fi sources, and its data structure is consistent with the fingerprint data in the offline database to ensure comparability in subsequent matching calculations. During the positioning process, the system uses the K-nearest neighbor algorithm to calculate the feature similarity between the online data and fingerprint samples in the offline database. By measuring the similarity between the online fingerprint and each offline fingerprint, several candidate offline collection points with high similarity are selected. Further, based on the spatial location information of the selected candidate collection points, the location of the online collection points is calculated to obtain the preliminary positioning result corresponding to the online data.

[0038] In a preferred embodiment of the present invention, step 3, based on the preliminary positioning results and combined with the offline information database, forms a spatial relationship map of the collection points, including: Step 31: Based on the preliminary positioning results of the online acquisition points, combine the online acquisition signal characteristics and positioning results with the offline data to form basic spatial relationship data. Specifically, this includes: batch extracting the full data of all offline acquisition points from the Wi-Fi fingerprint positioning offline database, and filtering out the core fields: unique acquisition point number, x / y plane coordinates (with the southwest corner of the first floor of the target area as the origin), floor number, signal power value corresponding to each BSSID, and acquisition time; cleaning the extracted offline data, removing invalid offline acquisition points with empty coordinates or all signal power values ​​with special annotations (such as -100dBm), and retaining valid data; assigning a unique offline node identifier to each valid offline acquisition point (naming rule: LF-floor number-acquisition point sequence number, such as LF-01-0089), unifying the coordinate accuracy to the centimeter level and the signal power unit to dBm, and finally forming a structured and standardized offline node dataset. The dataset fields include: offline node identifier, floor number, x coordinate (cm), y coordinate (cm), BSSID list, and corresponding signal power list. The generated standardized offline node dataset is split according to floor number to obtain independent offline node subsets for each floor (e.g., floor 1 subset, floor 2 subset), ensuring that the data of each floor is processed independently and avoiding cross-floor coordinate confusion. For the offline node subset of a single floor, the x / y coordinates of all nodes are filtered out. First, the reference point is determined: traverse the y coordinates of all nodes on the floor and select the node with the smallest y coordinate (if there are multiple nodes with the same y coordinate, select the one with the smallest x coordinate) as the starting reference point for convex hull scanning. With the reference point as the center, calculate the azimuth angle of all other nodes on the floor relative to the reference point, and sort all nodes clockwise according to the azimuth angle from smallest to largest to form the sequence of nodes to be scanned.

[0039] Initialize a vertex stack, and push the pivot point and the first two sorted nodes onto the stack in sequence; Traverse each remaining node in the sequence of nodes to be scanned, and sequentially remove the top two nodes of the stack (denoted as node A and node B) and the currently traversed node (denoted as node C). Determine the turning direction from line segment AB to line segment BC: if it is a counterclockwise turn, pop node B from the stack and continue to determine the turning direction of the new top two nodes of the stack and node C; if it is a clockwise turn or collinear, push node C onto the stack. After the traversal is completed, the remaining nodes in the stack are the convex hull vertices of this floor. Record the x / y coordinates and offline node identifiers of these vertices. Connect the convex hull vertices in the stack in clockwise order to form the smallest convex polygon that encloses all offline collection points of the floor. This polygon is the effective spatial convex hull boundary of the floor. Confirm that all offline nodes of the floor are inside or on the boundary of the convex hull polygon. If a small number of nodes are outside the convex hull (≤1% of the total number of nodes), they are judged as abnormal offline nodes and removed. If the proportion of outliers exceeds 1%, the convex hull scanning process is re-executed to check for coordinate data errors. Repeat the above steps to complete the construction of the convex hull boundary of all floors in the target area and generate a correspondence table for each floor: floor number - convex hull vertex coordinate list - convex hull boundary range description.

[0040] Extract online data points from the current location request, including preliminary location results (3D location: floor number, x / y coordinates, height) and Wi-Fi signal characteristics (BSSID list and corresponding signal power). Based on the preliminary location floor number of the online data point, retrieve the convex hull boundary correspondence table for that floor and obtain the list of convex hull vertex coordinates. Determine whether the preliminary location x / y coordinates of the online data point are inside the convex hull polygon of that floor: Draw a ray from the online point in any direction (e.g., the positive x-axis direction), count the number of intersections between the ray and the boundary of the convex hull polygon, and if the number of intersections is odd, determine that it is inside the convex hull; if the number is even, determine that it is inside the convex hull. In addition, online acquisition points outside the convex hull are filtered out (these points have abnormal preliminary positioning results and no effective correction value), and only online acquisition points inside the convex hull are retained; a unique online node identifier is assigned to the retained online acquisition points (naming rule: ON-timestamp-random serial number, such as ON-202601181025-001), and the coordinate precision and signal power unit are unified to be consistent with the offline node dataset. The preliminary positioning coordinates after verification are supplemented to form a standardized online node dataset. The fields include: online node identifier, floor number, x coordinate (cm), y coordinate (cm), BSSID list, and corresponding signal power list.

[0041] The standardized offline node dataset and the standardized online node dataset were merged, and a node type field was added (offline nodes are labeled LF, online nodes are labeled ON). The field order and encoding format of the two datasets were unified to ensure that the BSSID list was arranged in the same order (arranged in ascending order of BSSID characters). Finally, a complete spatial relationship basic dataset was generated, with fields including: node identifier, node type, floor number, x-coordinate (cm), y-coordinate (cm), BSSID list, and signal power list. This dataset only contains valid offline / online acquisition points within the convex hull boundary of each floor, which limits the reasonable physical range for subsequent spatial distance calculations.

[0042] Step 32: Based on the spatial relationship data, calculate the Euclidean distance from each collection point to other collection points as the spatial distance. Specifically, this includes: based on the spatial relationship basic dataset generated in Step 31, standardizing the dimensionality of the location information of all collection points, converting floor information into corresponding height values ​​(e.g., 0m for floor 1, 3m for floor 2), so that the location of each collection point is represented by three-dimensional coordinates (x, y, height), ensuring the dimensional consistency of the spatial distance calculation; traversing each collection point in the dataset (including offline and online nodes), using that point as a reference point, calculating its Euclidean spatial distance to all other collection points in the dataset; during the calculation process, only collection points on the same floor (same height) are used for distance calculation, avoiding invalid distance calculations across floors, ensuring that the distance value can truly reflect the spatial proximity of collection points within the same physical plane; establishing a correspondence table of reference point identifier - target point identifier - spatial distance value for each reference point, associating all distance calculation results with the spatial relationship basic data, and supplementing the attribute fields of each node to form an extended dataset containing spatial distance information, which serves as the core basis for subsequent selection of neighboring collection points.

[0043] Step 33: Based on spatial distance, select candidate neighbor collection points to form a local spatial relationship map for each collection point. Specifically, this includes: based on the distance calculation results from Step 32, setting filtering rules for neighbor collection points, which can employ a distance threshold + K-nearest neighbor dual filtering: first, set a reasonable spatial distance threshold (e.g., 5m) to filter all collection points whose distance to the reference point is within the threshold; if the number of collection points within the threshold is less than the preset K value (e.g., K=10), then supplement with several nearest collection points up to K; if the number of collection points within the threshold exceeds the K value, then only the K nearest points are retained; for each collection point (reference point), treat itself as a local... The core node of the spatial relationship graph is selected as the candidate neighbor collection point and then as the adjacent node. An edge connection is established between the core node and each adjacent node, and the spatial distance value of the two is labeled with the attribute of the edge. At the same time, the basic attributes such as the position and signal characteristics of each node are retained to form a local spatial relationship subgraph centered on a single collection point. The generated local spatial relationship graph is validated. If the number of neighbor collection points of a certain reference point is 0 (such as an isolated online collection point), the distance threshold or K value is increased to ensure that each local graph contains at least a core node and more than 3 adjacent nodes, so as to avoid the failure of subsequent graph neural network training due to insufficient number of neighbors.

[0044] Step 34: Integrate all local spatial relationship graphs to form an offline spatial relationship graph containing only offline collection points and an overall spatial relationship graph containing both online and offline collection points. The former can be completed in advance, while the latter needs to be generated based on user data during actual use. Specifically, this includes: extracting the local spatial relationship subgraphs corresponding to all offline collection points in Step 33; integrating these subgraphs; merging all offline nodes into a global node set; merging all edge connections between offline nodes into a global edge set; and removing duplicate nodes and edges; standardizing the integrated graph structure, unifying node identifiers and edge attribute formats to form an offline spatial relationship graph containing only offline collection points. This graph can be pre-built and stored during the system deployment phase without repeated calculations; and combining the local spatial relationship subgraphs of all online collection points in Step 33 with the pre-generated... The offline spatial relationship graph is integrated, and online nodes are added to the global node set. Edges between online nodes and their adjacent nodes (offline / online) are connected to the global edge set. The type label (offline / online) of all nodes and the distance attribute of the edges are retained. During the integration process, the uniqueness of nodes and edges is verified in real time to avoid duplicate additions. Finally, an overall spatial relationship graph containing offline and online collection points is formed. This graph is dynamically generated according to the user's online positioning request. The graph structure is updated according to the new online collection points for each positioning request. A persistent storage index is established for the offline spatial relationship graph for quick retrieval. A temporary storage index is established for the overall spatial relationship graph. The temporary data can be released after positioning is completed, while the version information of the graph structure is retained to facilitate the tracking of the spatial relationship calculation process of different positioning requests and to provide a complete graph structure foundation for subsequent secondary positioning.

[0045] In this embodiment of the invention, a spatial relationship graph containing both online and offline collection points is constructed based on the preliminary location results of the online collection points and an offline database. First, the Wi-Fi signal characteristics and preliminary location results of the online collection points are fused with the collected data in the offline database to form a basic dataset for constructing spatial relationships. This basic data includes both known location information of the offline collection points and predicted location information of the online collection points, providing a basis for subsequent spatial relationship modeling.

[0046] Based on this, the system calculates the spatial distance between each collection point using the aforementioned spatial relationship data. This spatial distance is measured using Euclidean distance to quantify the relative proximity of different collection points in space. According to the calculated spatial distance results, several candidate neighbor collection points are selected for each collection point, thereby constructing a local spatial relationship graph corresponding to that collection point to describe its spatial association with surrounding collection points.

[0047] Furthermore, the system integrates the local spatial relationship maps of each collection point to form a complete spatial relationship map structure. The offline spatial relationship map, consisting only of offline collection points, can be pre-constructed during the system deployment phase; while the overall spatial relationship map, which includes both offline and online collection points, is dynamically generated based on user online data during the actual positioning process. The constructed spatial relationship map provides the necessary graph structure foundation for subsequent positioning correction based on graph neural networks.

[0048] In a preferred embodiment of the present invention, the online acquisition points are repositioned based on a spatial relationship diagram to obtain a secondary positioning result, including: Based on the offline spatial relationship graph, the localization model is trained. The core feature transformation formula for training the model is: ; in This is a neighbor feature aggregation algorithm for the localization model. This represents the layer index of the graph neural network; It represents a node variable; It is a target node; It is the node feature vector; It is the first The trainable weight matrix of the layer; It is a node The set of neighboring nodes; Based on the trained model, input the overall spatial relationship map and perform secondary localization on all the collection points in the map; Based on the positioning results, the positioning results of the online collection points are extracted to form secondary positioning results.

[0049] In this embodiment of the invention, the localization model is trained based on the constructed spatial relationship graph, and the trained model is used to perform secondary localization on online collected data. First, the system uses the offline spatial relationship graph as training data to learn the localization model. The localization model is based on the GraphSAGE structure and adopts an inductive modeling approach. By sampling and training the subgraph, the model can effectively learn the correlation patterns between spatial relationship features and signal features between collection points, thereby improving the model's generalization ability to unknown nodes.

[0050] After model training is complete, the overall spatial relationship graph containing both offline and online collection points is used as input. The trained localization model is then used to perform secondary localization calculations on all collection points in the graph. By aggregating and propagating information within the graph structure, the model can comprehensively consider the local spatial relationships of collection points and the features of adjacent nodes, further refining the initial localization results.

[0051] Finally, based on the results of the secondary positioning calculation, the positioning information of the corresponding online acquisition points is extracted from the overall spatial relationship map, serving as the secondary positioning result of the online acquisition data. Compared with the initial positioning result, the secondary positioning result can effectively reduce the positioning error, providing a reliable foundation for further improvement of subsequent positioning accuracy.

[0052] In a preferred embodiment of the present invention, based on the secondary positioning results, the spatial relationship map of the collection points is updated, and a multi-source relationship map is formed by combining the signal characteristics of offline and online data, including: Based on the secondary positioning results, update the positions of the online collection points in the spatial relationship map and recalculate the local spatial relationships; Update the overall spatial relationship diagram based on the updated local spatial relationships; Based on the calculated feature similarity, candidate neighbor collection points are selected to form a local feature relationship graph for each collection point; By combining all local feature relationship diagrams, an offline feature relationship diagram containing only offline acquisition points and an overall feature relationship diagram containing both online and offline acquisition points are formed. The former can be completed in advance, while the latter needs to be generated based on user data during actual use. Since the feature relationship in this invention specifically refers to the signal feature relationship, it can also be called a signal relationship diagram.

[0053] By combining spatial relationship diagrams and feature relationship diagrams, a multi-source relationship diagram is formed, which includes offline relationship diagrams and overall relationship diagrams.

[0054] In this embodiment of the invention, the spatial relationship map of the online acquisition points is updated based on the secondary positioning results of the online acquisition points, and further combined with the signal feature information of offline and online data to construct a multi-source relationship map for subsequent processing. First, the system corrects and updates the position information of the online acquisition points in the spatial relationship map based on the secondary positioning results, and recalculates the local spatial relationship between the online acquisition points and their neighboring acquisition points, thereby reflecting a more accurate spatial distribution.

[0055] Subsequently, based on the updated local spatial relationships, the overall spatial relationship graph is updated synchronously to dynamically reflect the latest spatial association structure between offline and online acquisition points. Building upon this, the system utilizes the feature similarity results calculated during the initial localization phase to select candidate neighbor acquisition points with similar signal features for each acquisition point, constructing corresponding local feature relationship graphs to describe the association between acquisition points in the signal feature dimension.

[0056] Furthermore, the local feature relationship graphs of each collection point are integrated to form an offline feature relationship graph containing only offline collection points, and an overall feature relationship graph containing both offline and online collection points. The offline feature relationship graph can be pre-built during system deployment, while the overall feature relationship graph is dynamically generated based on user online data during actual positioning. Finally, the updated spatial relationship graph and feature relationship graph are fused to construct a multi-source relationship graph, including an offline multi-source relationship graph and an overall multi-source relationship graph, providing the necessary graph structure foundation for subsequent positioning correction based on heterogeneous graph neural networks.

[0057] In a preferred embodiment of the present invention, final positioning is performed based on the multi-source relationship diagram to complete the progressive correction of positioning deviation, including: Based on the offline multi-source relationship graph, the localization model is trained to obtain a trained model; the core transformation formula of the localization model is: ; ; ; in For the shared input layer of the model, It is a signal relationship diagram; These are node features in the signal relationship diagram; and It is the set of edges between nodes in the spatial relationship graph and the signal relationship graph; It is a spatial graph passing through a shared layer The processed intermediate node feature matrix, The signal diagram passes through the shared layer. The processed intermediate node feature matrix; It is the final node feature matrix output by the two parallel branches. These are the final node feature matrices output from the two parallel branches, respectively. Based on the trained model, input the overall multi-source relationship graph, and finally locate all the collection points in the graph to obtain the location results; Based on the positioning results, the positioning results of the online collection points are extracted to form the final positioning results.

[0058] In this embodiment of the invention, the localization model is trained based on the constructed multi-source relationship graph, and the trained model is used to complete the final localization of online acquisition points. First, the system uses the offline multi-source relationship graph as training data to learn the localization model, enabling the model to fully capture the joint patterns of spatial relationships and signal feature relationships between acquisition points, thereby improving the generalization ability and localization accuracy for unknown acquisition points.

[0059] After model training is complete, a multi-source relationship graph containing both offline and online acquisition points is used as input. The trained localization model is then used to perform final localization calculations on all acquisition points in the graph. Through joint modeling of multi-source information, the model can comprehensively utilize spatial relationships and signal feature information to accurately infer the location of online acquisition points, thereby further reducing localization errors.

[0060] Finally, based on the final positioning calculation results, the positioning information of the online collection points is extracted from the overall multi-source relationship graph to form the final positioning result. The model design integrates multi-source information and performs feature sharing and parallel processing within the model, enabling the positioning calculation to more comprehensively utilize multi-dimensional data features, improving the stability and accuracy of positioning, while maintaining the model's scalability and adaptability.

[0061] For example, in practical applications, a method for progressively correcting positioning errors based on indoor Wi-Fi fingerprinting involves: acquiring offline Wi-Fi fingerprint information to form an offline fingerprint database; acquiring online Wi-Fi fingerprint information for preliminary positioning; combining the preliminary positioning with the offline fingerprint database to form a spatial relationship map of the collection points; performing secondary positioning on the online collection points based on the spatial relationship map to obtain secondary positioning results; updating the spatial relationship map of the collection points based on the secondary positioning results to form a multi-source relationship map; and performing final positioning based on the multi-source relationship map to complete the progressive correction of positioning errors. This improves the accuracy of indoor Wi-Fi fingerprint positioning. The specific implementation steps are as follows: Offline Wi-Fi fingerprint data is acquired, which mainly includes the Wi-Fi source SSID, Wi-Fi source BSSID, RSSI, collection point location, collection device, and collection time. Considering the repeatability of Wi-Fi names, the BSSID is used as the unique identifier of the source to form an offline database for Wi-Fi fingerprint positioning.

[0062] Acquire online Wi-Fi fingerprint data, which mainly includes the Wi-Fi source SSID, Wi-Fi source BSSID, RSSI, acquisition device, and acquisition time, but does not include the acquisition location. For a single online acquisition result, calculate and match the corresponding offline acquisition point in the offline database based on the BSSID and the corresponding signal power RSSI. Based on the location of the matched offline acquisition point, calculate the preliminary location result of the current online acquisition point.

[0063] In this process, the main calculation and matching algorithm used is the K-nearest neighbor algorithm, which ensures that the offline acquisition points matched by each online acquisition point maintain a close proximity relationship in terms of signal and space; the main online acquisition point localization algorithm used is the weighted centroid algorithm, which ensures that the preliminary localization results remain stable and interpretable.

[0064] The spatial relationship graph of the data collection points needs to reflect the relationships between them in real space. Spatial labels include latitude and longitude, floor level, and building information. The construction method for the spatial relationship graph is as follows: First, the spatial labels of the data collection points are used as node features input to the graph structure. These labels include latitude and longitude, floor level, and building information. Latitude and longitude are used to calculate the two-dimensional planar distance between nodes, and floor level labels are converted to three-dimensional spatial coordinates by assigning a fixed height to assist in distance calculation. Building labels are not directly involved in the graph construction process. Next, pairwise distance calculations are performed on all data collection points, including the actual location of offline data collection points and the preliminary predicted location of online data collection points, to quantify the spatial proximity between nodes. Based on the calculated distances, the K-nearest neighbor method is used to select several neighboring nodes for each node, establishing edge connections between nodes to form a local spatial adjacency structure. Finally, by integrating the adjacency relationships of all nodes, a complete spatial relationship graph is constructed. This graph can simultaneously reflect the relative positions of offline and online data collection points in real space, providing a basic graph structure for subsequent localization correction based on graph neural networks.

[0065] In this process, the metric used to calculate spatial distance is Euclidean distance, which reflects the shortest distance between two points in Euclidean space, i.e., the distance between the collection points in a real-world scenario. The constructed spatial relationship graph of the collection points contains nodes and edges, where each node represents an offline or online collection point, and edges represent the connection relationships between nodes.

[0066] For spatial relationship graphs, an inductive graph neural network is used for feature processing, and a multilayer perceptron is used to map the features to spatial labels. The data processing mode of the graph neural network model involves using the initial features of each collection point as node input and performing inductive feature aggregation using the GraphSAGE model. Specifically, for each node, the system collects feature information from its neighboring nodes and fuses the neighboring features with its own features through mean aggregation, thereby generating an updated feature representation for the node. This process can be iteratively performed in the multilayer graph neural network to fully capture the spatial relationships and feature information of the collection points both locally and over a wider range. The aggregated node features are then input into the multilayer perceptron, which maps the high-dimensional features to the target spatial label, i.e., three-dimensional coordinates or location representation, thereby enabling the prediction and correction of the spatial location of online collection points.

[0067] After secondary localization based on the spatial relationship map, the result can be considered a correction to the initial localization. The localization result is the updated spatial label of the online acquisition points, which can be used to update the constructed spatial relationship map. However, updating and tertiary localization based solely on the spatial relationship map cannot provide more accurate localization results. Therefore, a signal relationship map is added at this step to construct a multi-source relationship map that includes both offline and online acquisition points.

[0068] Multi-source relationship graphs, by introducing signal feature dimension information on top of the original spatial relationship graph, achieve a multi-faceted characterization of the relationships between acquisition points. On the one hand, the spatial relationship graph reflects the physical proximity and structural constraints of acquisition points; on the other hand, the signal relationship graph depicts the similarity and correlation of Wi-Fi signal features among different acquisition points. The multi-source relationship graph formed by fusing the two not only comprehensively utilizes the complementary information of spatial labels and signal features but also effectively alleviates the information deficiency problem caused by single relationship modeling.

[0069] Based on multi-source relational graphs, the model can simultaneously acquire information from both spatial and signal similarity neighborhoods during feature propagation and aggregation. This allows the localization process to move beyond relying solely on geometric distance and instead incorporate environmental perception features for joint inference. This multi-source modeling approach enhances the model's robustness to signal fluctuations, occlusion, and structural irregularities in complex indoor environments. Consequently, it provides a more comprehensive and reliable structured information foundation for subsequent localization calculations based on heterogeneous graph neural networks, further improving localization accuracy and stability.

[0070] For multi-source relation graphs, an inductive heterogeneous graph neural network is used for feature processing, and a multilayer perceptron is used to map the features to spatial labels. The data processing mode of the heterogeneous graph neural network model is as follows: First, the features of the collected points in the multi-source relation graph are input into a shared layer for unified processing. This shared layer, as the model's input layer, performs preliminary encoding of node features in both the spatial and signal relation graphs based on the GraphSAGE structure. This allows node features under different relation types to be mapped to the same feature space, thus providing a consistent feature representation foundation for subsequent multi-relation modeling.

[0071] After the shared layer processing is complete, the model sets up two parallel graph neural network branches, corresponding to the spatial relationship graph and the signal relationship graph, respectively. Each parallel layer is based on the GraphSAGE structure, independently processing and propagating the node features in its respective relationship graph to capture the feature representation of the acquisition points under spatial neighborhood relationships and signal similarity relationships. Through this parallel modeling approach, the model can learn structural information and feature associations from different relationship perspectives.

[0072] Subsequently, the node features output from the two parallel layers are concatenated and fused to form a feature representation that comprehensively reflects multi-source relationship information. Finally, the fused features are input into a multilayer perceptron, which performs nonlinear mapping on the node features and outputs corresponding spatial labels, achieving the final localization of the online acquisition points. Through a heterogeneous modeling approach combining shared and parallel layers, the model can fully utilize multi-source relationship information, improving the accuracy and stability of the localization results.

[0073] In this process, a basic positioning algorithm is first used to provide preliminary online point location results, which allows the construction of a spatial relationship map for the first error correction. Subsequently, a multi-source relationship map formed by the updated spatial relationship map and the signal relationship map is used to complete the second error correction. The preliminary positioning and the two error corrections form a complete progressive relationship, ultimately yielding a more accurate indoor Wi-Fi fingerprint positioning result.

[0074] The invention will be further explained using real-world datasets. Table 1 below provides examples of the datasets used.

[0075] The datasets used are the UJIIndoorLoc and UTSIndoorLoc public datasets, both of which are datasets formed by actual measurements inside large and complex buildings. Figure 1 and Figure 3 This displays all the data points collected in the dataset, with blue dots representing offline training data and red dots representing online test data.

[0076] For each online data collection point, offline data was used for data matching and positioning to obtain the preliminary positioning results of the online data collection point, as shown in Table 1.

[0077] Based on the preliminary location results and combined with offline data, a spatial relationship map of the collection points was constructed.

[0078] Based on the spatial relationship diagram, a deep learning model was used to locate the online collection points and obtain the secondary localization results, as shown in Table 1.

[0079] Based on the secondary positioning results and combined with the signal characteristic information of offline and online data, a multi-source relationship diagram is constructed, including a spatial relationship diagram and a signal relationship diagram.

[0080] Based on the multi-source relationship graph, a heterogeneous deep learning model was used to locate the online collection points, and the final location results were obtained, as shown in Table 1.

[0081] After the above steps, the positioning results in Table 1 were finally obtained. The positioning error shows that the proposed progressive error correction method can effectively correct the results of the positioning algorithm based on the traditional positioning method, thus improving the positioning accuracy.

[0082] Table 1 Positioning Error Correction Process

[0083] like Figure 5 As shown, embodiments of the present invention also provide a progressive correction system 20 for positioning deviation in indoor Wi-Fi fingerprint positioning technology, comprising: The acquisition module 21 is used to acquire Wi-Fi signal data and extract the Wi-Fi source BSSID and signal power from the signal data to form Wi-Fi fingerprint data; based on the Wi-Fi fingerprint data and the location information of offline acquisition points, an offline fingerprint database is formed; based on the offline fingerprint database and the online fingerprint data, an acquisition point feature relationship diagram is formed, including an offline feature relationship diagram and an overall feature relationship diagram; based on the initial and secondary positioning results, an acquisition point spatial relationship diagram is formed, including an offline spatial relationship diagram and an overall spatial relationship diagram.

[0084] Processing module 22 locates the online collection points based on the offline information database to obtain preliminary location results; it performs secondary location of the online collection points based on the spatial relationship diagram to obtain secondary location results; and it performs final location based on the multi-source relationship diagram to complete the progressive correction of the location deviation.

[0085] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0086] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0087] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0088] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0089] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0090] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0091] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0092] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0093] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0094] Furthermore, it should be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent solutions of the present invention. Moreover, the steps performing the above-described series of processes can naturally be executed in the order described, but are not necessarily required to be executed in chronological order; some steps can be executed in parallel or independently of each other. Those skilled in the art will understand that all or any step or component of the method and apparatus of the present invention can be implemented in any computing device or network of computing devices, in hardware, firmware, software, or a combination thereof. This can be achieved by those skilled in the art using basic programming skills after reading the description of the present invention.

[0095] Therefore, the object of the present invention can also be achieved by running a program or a set of programs on any computing device. The computing device can be a known general-purpose device. Therefore, the object of the present invention can also be achieved simply by providing a program product containing program code implementing the method or apparatus. That is, such a program product also constitutes the present invention, and the storage medium storing such a program product also constitutes the present invention. Obviously, the storage medium can be any known storage medium or any storage medium developed in the future. It should also be noted that in the apparatus and method of the present invention, it is obvious that the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered equivalent to the present invention. Furthermore, the steps performing the above series of processes can naturally be performed in the order described, but are not necessarily required to be performed in chronological order. Some steps can be performed in parallel or independently of each other.

[0096] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology, characterized in that, The method includes: Acquire offline Wi-Fi signal data, extract Wi-Fi fingerprint information from the acquired data, and form an offline information database; Acquire online Wi-Fi signal data, locate the online collection points based on the offline information database, and obtain preliminary location results; Based on the preliminary location results and combined with the offline information database, a spatial relationship map of the collection points is generated; Based on the spatial relationship diagram, the online collection points are repositioned to obtain the repositioning results; Based on the secondary positioning results, the spatial relationship map of the collection points is updated, and a multi-source relationship map is formed by combining the signal characteristics of offline and online data. Based on the multi-source relationship diagram, the final positioning is performed, and the positioning deviation is progressively corrected.

2. The progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology according to claim 1, characterized in that, Acquire offline Wi-Fi signal data, extract Wi-Fi fingerprint information from the acquired data, and form an offline information database, including: The Wi-Fi signal in the target area is collected from multiple points to obtain raw offline data; Based on the raw data, obtain Wi-Fi signal source information and distinguish different signal sources according to BSSID; Obtain the signal power of each source, using decibels and milliwatts as the unit. Sources where no valid signal was detected are marked with special values. Obtain the location of each offline collection point and acquire the corresponding collection data; Based on the location of the collection points and the corresponding collection information, an offline database for Wi-Fi fingerprint positioning is established using structured storage.

3. The progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology according to claim 2, characterized in that, Acquire online Wi-Fi signal data, locate the online collection points based on the offline information database, and obtain preliminary location results, including: Based on the characteristics of the acquired signals from the offline data, sample data is added to the localization model; Obtain online data collected by users of location services; Calculate the feature distance between the online collected data and the offline data; Candidate offline collection points are selected based on feature distance, and online collection locations are calculated.

4. The progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology according to claim 3, characterized in that, Based on the preliminary location results and combined with the offline information database, a spatial relationship map of the collection points was generated, including: Based on the preliminary positioning results of the online acquisition points, the characteristics of the online acquisition signals and the positioning results are combined with the offline data to form basic spatial relationship data; In spatial relationship data, the Euclidean distance from each collection point to other collection points is calculated as the spatial distance; Based on spatial distance, candidate neighbor collection points are selected to form a local spatial relationship map for each collection point; Integrate all local spatial relationship diagrams to form an offline spatial relationship diagram containing only offline collection points and an overall spatial relationship diagram containing both online and offline collection points.

5. The progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology according to claim 4, characterized in that, Based on the spatial relationship diagram, the online data collection points are repositioned to obtain the repositioning results, including: The localization model is trained based on the offline spatial relationship graph; Based on the trained model, input the overall spatial relationship map and perform secondary localization on all the collection points in the map; Based on the positioning results, the positioning results of the online collection points are extracted to form secondary positioning results.

6. The progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology according to claim 5, characterized in that, Based on the secondary positioning results, the spatial relationship map of the collection points is updated, and a multi-source relationship map is formed by combining the signal characteristics of offline and online data, including: Based on the secondary positioning results, update the positions of the online collection points in the spatial relationship map and recalculate the local spatial relationships; Update the overall spatial relationship diagram based on the updated local spatial relationships; Based on the calculated feature similarity, candidate neighbor collection points are selected to form a local feature relationship graph for each collection point; By combining all local feature relationship graphs, an offline feature relationship graph containing only offline collection points and an overall feature relationship graph containing both online and offline collection points are formed. By combining spatial relationship diagrams and feature relationship diagrams, a multi-source relationship diagram is formed, which includes offline relationship diagrams and overall relationship diagrams.

7. The progressive correction method for positioning deviation in indoor Wi-Fi fingerprint positioning technology according to claim 6, characterized in that, Based on the multi-source relationship diagram, final positioning is performed, and progressive correction of positioning errors is completed, including: The localization model is trained based on the offline multi-source relationship graph to obtain the trained model; Based on the trained model, input the overall multi-source relationship graph, and finally locate all the collection points in the graph to obtain the location results; Based on the positioning results, the positioning results of the online collection points are extracted to form the final positioning results.

8. A progressive correction system for positioning deviation in indoor Wi-Fi fingerprint positioning technology, the system being used to perform the method as described in any one of claims 1 to 7, characterized in that, include: The acquisition module is used to acquire Wi-Fi signal data and extract the Wi-Fi source BSSID and signal power from the signal data to form Wi-Fi fingerprint data; Based on Wi-Fi fingerprint data and the location information of offline collection points, an offline fingerprint database is formed; based on the offline fingerprint database and online fingerprint data, a collection point feature relationship diagram is formed, including an offline feature relationship diagram and an overall feature relationship diagram; based on the initial and secondary positioning results, a collection point spatial relationship diagram is formed, including an offline spatial relationship diagram and an overall spatial relationship diagram. The processing module is used to locate the online collection points based on the offline information database and obtain preliminary location results; to perform secondary location of the online collection points based on the spatial relationship diagram to obtain secondary location results; and to perform final location based on the multi-source relationship diagram to complete the progressive correction of location deviation.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to perform the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images