On-device ai-based hybrid positioning method, apparatus, and server for supporting on-device hybrid positioning in user terminals

Abstract

One aspect of the present invention discloses an on-device composite positioning method using a plurality of wireless signals at a terminal. The method comprises the steps of: collecting a composite wireless signal including at least two of a cellular base station signal, a WiFi AP signal, and a BLE signal; extracting cellular cell identification information (cellular cell ID) from the collected composite wireless signal; transmitting the extracted cellular cell identification information to a server; receiving from the server a distilled on-device AI positioning model corresponding to a region related to the current location of the terminal among the AI ​​positioning models held by the server; and determining the current location of the terminal based on the collected composite wireless signal using the received on-device AI positioning model, wherein the distilled on-device AI positioning model is provided to the terminal by analyzing the cellular cell identification information at the server.

Description

Technology Field

[0001] The present invention relates to a composite positioning method, and more specifically, to a method for precise positioning using AI based on a plurality of wireless signals. Background Technology

[0002] In modern society, location-based services have become an essential part of daily life, and various positioning technologies have been developed alongside the proliferation of smartphones. Traditionally, the Global Positioning System (GPS) was the most widely used positioning technology in outdoor environments, but there were problems with significantly reduced accuracy in indoor environments or urban areas due to signal shielding and multipath phenomena.

[0003] To overcome the limitations of GPS, various alternative technologies have been developed, such as WiFi fingerprinting, cellular base station-based positioning, and positioning using BLE beacons. In particular, WiFi fingerprinting technology estimates location by database-ing signal strength patterns of WiFi APs in indoor environments, offering the advantage of enabling indoor positioning without the need for separate infrastructure. However, these existing technologies operate independently and rely on a single signal source, which makes them vulnerable to environmental changes and results in unstable positioning accuracy.

[0004] Recently, hybrid positioning technology, which improves positioning accuracy by fusing multiple signal sources, has been attracting attention. Hybrid positioning is a method that utilizes various signal sources simultaneously—such as cellular base stations, WiFi APs, and BLE beacons—to combine the strengths of each source and compensate for their weaknesses. Furthermore, with the advancement of artificial intelligence technology, deep learning-based positioning models have been introduced to learn complex signal patterns and enable more accurate location estimation.

[0005] However, most conventional hybrid positioning technologies had a server-based operating structure. The method involved transmitting signal information collected from a terminal to a server, where the server calculated the location using a large-scale AI model and then sent the result back to the terminal. This server-centric approach had the following problems.

[0006] First, there was a problem with a lack of real-time capability. Network delays occurred due to data transmission between the terminal and the server, and it took a significant amount of time to receive positioning results, particularly in environments with unstable network conditions. In situations requiring rapid location identification, such as emergency rescue operations, this delay could be a critical issue.

[0007] Second, there was a problem with high network dependency. Positioning itself was impossible in situations where communication with the server was not possible, and services could not be provided in environments with limited network infrastructure, such as underground spaces or disaster situations. In particular, in emergency rescue services such as 119 or 112, failures to determine location due to communication failures occurred frequently.

[0008] Third, there was the issue of personal information protection. Since users' location information and signal pattern data were continuously transmitted to and stored on the server, there was a constant risk of personal information leakage. In addition, issues regarding user consent and privacy protection were intricately intertwined when performing additional training using data accumulated on the server.

[0009] Fourth, there was the problem of high server operating costs. Processing large-scale AI models in real time required high-performance server infrastructure, and there were limitations to scalability as server load increased rapidly with the number of users. In addition, maintenance costs to provide stable 24-hour service were substantial.

[0010] Fifth, there was a problem with the difficulty of providing personalized positioning services. Since the location of all users was estimated using a single server model, it was difficult to provide customized services that reflected individual movement patterns or the characteristics of frequently visited areas. Even in the areas where users were primarily active, only a general level of accuracy could be provided. The problem to be solved

[0011] An objective according to one aspect of the present invention is to provide an on-device composite positioning method and system that can determine location in real time without network delay by directly executing a fragmented AI positioning model received from a server at a terminal, enable positioning even in situations where communication is impossible, protect privacy by processing personal information only within the terminal, and provide customized positioning accuracy for each user through self-learning. means of solving the problem

[0012] According to one embodiment of the present invention for achieving the above-mentioned purpose, an on-device composite positioning method using a plurality of wireless signals at a terminal comprises: a step of collecting a composite wireless signal including at least two of a cellular base station signal, a WiFi AP signal, and a BLE signal; a step of extracting cellular cell identification information (cellular cell ID) from the collected composite wireless signal; a step of transmitting the extracted cellular cell identification information to a server; a step of receiving from the server a distilled on-device AI positioning model corresponding to a region related to the current location of the terminal among the AI ​​positioning models possessed by the server; and a step of determining the current location of the terminal based on the collected composite wireless signal using the received on-device AI positioning model, wherein the distilled on-device AI positioning model may be provided to the terminal by analyzing the cellular cell identification information at the server.

[0013] The fragmented on-device AI positioning model may be a total positioning model of the server divided into grid-shaped regional units through at least one of clustering and spatial indexing algorithms.

[0014] The above grid shape can be adaptively determined depending on the region.

[0015] The above fragmented on-device AI positioning model can be composed of the collected composite wireless signal and sensor information.

[0016] The above method may further include the step of self-training the fragmented on-device AI positioning model using the positioning result through the fragmented on-device AI positioning model and the collected composite wireless signal.

[0017] The above fragmented on-device AI positioning model can have different positioning precision by region through self-learning.

[0018] The above method further includes the step of receiving movement information of the terminal from a sensor, and based on the movement information, predicting the direction of movement of the terminal to receive the fragmented on-device AI positioning model.

[0019] The above method may further include the steps of generating a WiFi AP signal in WiFi Direct mode, detecting the generated WiFi AP signal at another terminal, measuring the signal strength of the detected WiFi AP signal, determining the proximity distance between the terminal and the other terminal based on the measured signal strength, and outputting a signal based on the determined proximity distance.

[0020] The above method further includes the step of detecting a BLE advertising signal generated from the other terminal and the step of measuring the signal strength of the detected BLE advertising signal, wherein the step of determining the proximity distance may include the step of determining the proximity distance between the terminal and the other terminal by combining the signal strength of the WiFi AP signal and the signal strength of the BLE advertising signal.

[0021] The above method may further include the step of tracking the location of the terminal in real time in an application equipped with the fragmented on-device AI positioning model, and the step of the application automatically transmitting the real-time tracked current location information to a pre-designated contact in response to detecting the occurrence of an emergency.

[0022] According to another aspect of the present invention for achieving the above-mentioned purpose, an on-device composite positioning terminal using a plurality of wireless signals comprises: a signal collection unit that collects a composite wireless signal including at least two of a cellular base station signal, a WiFi AP signal, and a BLE signal; a signal processing unit that extracts cellular cell identification information (cellular cell ID) from the collected composite wireless signal; a communication unit that transmits the extracted cellular cell identification information to a server and receives a distilled on-device AI positioning model from the server; a storage unit that stores the received on-device AI positioning model; and a positioning processing unit that determines the current location of the terminal based on the collected composite wireless signal using the on-device AI positioning model, wherein the distilled on-device AI positioning model may be provided as a model corresponding to a region related to the current location of the terminal by analyzing the cellular cell identification information at the server.

[0023] According to another aspect of the present invention for achieving the above-mentioned purpose, a positioning server supporting on-device composite positioning of a terminal comprises: a model storage unit that stores an entire AI positioning model learned based on composite wireless signal data including at least two of a cellular base station signal, a WiFi AP signal, and a BLE signal, by dividing the data into grid-shaped regional units and storing the data in the form of fragmented on-device AI positioning models; a communication unit that receives cellular cell identification information (cellular cell ID) from a terminal; and a model transmission unit that analyzes the received cellular cell identification information and transmits a fragmented on-device AI positioning model related to the current location of the terminal to the terminal, wherein the fragmented on-device AI positioning model may be configured to enable positioning processing based on composite wireless signals in an on-device manner at the terminal. Effects of the invention

[0024] According to the on-device composite positioning method and system of the present invention, real-time positioning without network delay is possible by directly executing a fragmented AI positioning model on the terminal, location determination is possible even in situations where communication is impossible, privacy is protected as personal information is processed only within the terminal, positioning accuracy for areas frequently visited by the user is continuously improved through self-learning, and system scalability is improved by reducing server load. Brief explanation of the drawing

[0025] FIG. 1 is a conceptual diagram illustrating the concept of composite positioning according to one embodiment of the present invention, FIG. 2 is a block diagram showing the overall configuration of an on-device composite positioning system according to an embodiment of the present invention, FIG. 3 is a flowchart illustrating the process of collecting and processing complex signals of a terminal according to an embodiment of the present invention, FIG. 4 is a conceptual diagram for explaining the process of partitioning the entire model and creating fragment models of a server according to an embodiment of the present invention. FIG. 5 is a conceptual diagram showing the structure of an AI positioning model according to an embodiment of the present invention, FIG. 6 is a flowchart illustrating a self-learning process according to an embodiment of the present invention, FIG. 7 is a conceptual diagram showing grid-type region partitioning and adaptive grid size according to an embodiment of the present invention, FIG. 8 is an exemplary diagram illustrating a WiFi Direct-based proximity tracking function according to an embodiment of the present invention. FIG. 9 is an exemplary diagram showing a personal protection application linkage system according to an embodiment of the present invention. FIG. 10 is a conceptual diagram showing a pipeline for building and operating a fragment positioning model for on-device according to one embodiment of the present invention. Specific details for implementing the invention

[0026] The present invention is capable of various modifications and may have various embodiments, and specific embodiments are illustrated in the drawings and described in detail.

[0027] However, this is not intended to limit the invention to specific embodiments, and it should be understood that it includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention.

[0028] Terms such as "first," "second," etc., may be used to describe various components, but said components should not be limited by said terms. Such terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component. The term "and / or" includes a combination of a plurality of related described items or any of a plurality of related described items.

[0029] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.

[0030] The terms used in this application are used merely to describe specific embodiments and are not intended to limit the invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "having" are intended to specify the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0031] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0032] Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. In order to facilitate an overall understanding of the present invention, the same reference numerals are used for identical components in the drawings, and redundant descriptions of identical components are omitted.

[0034] FIG. 1 is a conceptual diagram illustrating the concept of composite positioning according to one embodiment of the present invention.

[0035] Referring to FIG. 1, composite positioning is a technology that determines location by utilizing multiple wireless signals rather than a single signal source. According to one embodiment of the present invention, at least two of cellular base station signals (e.g., LTE base station signals), WiFi AP signals, and BLE signals are utilized comprehensively. In this case, a combination of signals collected at each point acts as a unique fingerprint of the location, and this signal fingerprint is stored in a database and trained with an AI model to estimate the location when a new signal pattern is input.

[0036] Meanwhile, the hybrid positioning server acquires at least two of the cellular base station signals, WiFi AP signals, and BLE signals collected through multiple signal collectors, and trains a positioning AI model by integrating them. At this stage, it is desirable to utilize base station signals from all carriers to enable accurate positioning regardless of the carrier. The collected signal data is processed by the hybrid positioning server, which consists of a base station matching model and a positioning AI model, to ultimately determine the positioned location.

[0037] To determine the position, the terminal transmits cellular cell identification information (e.g., LTE Cell ID) to the server. The server, which possesses a trained composite positioning AI model covering the entire area, transmits a distilled composite positioning AI model for the area corresponding to the terminal's location to the terminal based on the Cell ID received from the terminal. Upon receiving the distilled composite positioning AI model, the terminal performs on-device positioning using the received model to enable precise location measurement.

[0038] FIG. 2 is a block diagram showing the overall configuration of an on-device composite positioning system according to an embodiment of the present invention. As shown in FIG. 2, the composite positioning system according to an embodiment of the present invention may include a positioning optimization server (210), a positioning model server (220), and a user terminal (230).

[0039] Referring to FIG. 2, the positioning optimization server (210) performs the function of processing signal data collected from various regions and training the entire positioning AI model. This server performs the processing of dividing the entire model by region through clustering and spatial indexing algorithms.

[0040] The positioning model server (220) performs the function of storing and managing fragmented AI models generated by the positioning optimization server (210). When a cellular cell ID is received from a user terminal (230), a fragmented model of the region matching the ID is selected and transmitted to the terminal. Through this, the terminal can receive and use only the lightweight model specialized for that region, rather than the entire large-capacity model.

[0041] The user terminal (230) may include a signal collection unit (231), a model storage unit (233), a positioning processing unit (235), a self-learning unit (237), and a communication unit (239). Here, the signal collection unit (231) and the communication unit (239) may be implemented as a communication processor or an antenna, etc. The model storage unit (233) may be implemented as a memory, and the positioning processing unit (235) and the self-learning unit (237) may be implemented as processors.

[0042] The signal collection unit (231) collects signals from surrounding cellular base stations, WiFi APs, and BLE beacons. This can be performed periodically. At this time, the signal processing unit (not shown) extracts a cell ID from the collected signal and transmits it to the positioning model server (220) via the communication unit (239). Based on this cell ID, the positioning model server (220) transmits a fragmented AI positioning model corresponding to the current location of the user terminal (230) to the communication unit (239) of the user terminal (230). The model storage unit (233) stores the fragmented AI positioning model received from the server. The positioning processing unit (235) calculates the current location based on the signal collected by the signal collection unit (231) using the stored fragmented AI positioning model. The self-learning unit (237) can perform the function of continuously self-learning the fragmented AI positioning model stored in the model storage unit (233) using the positioning results and collected signal information to improve positioning accuracy. The communication unit (239) is responsible for transmitting and receiving data with a server or other terminal.

[0043] FIG. 3 is a flowchart illustrating the process of collecting and processing complex signals of a terminal according to one embodiment of the present invention.

[0044] Referring to FIG. 3, a composite signal collection and processing process according to one embodiment of the present invention begins with a composite signal collection step (S310). First, the terminal collects at least two of a cellular base station signal, a WiFi AP signal, and a BLE signal. The collection may be performed periodically or non-periodically.

[0045] Then, the terminal extracts information such as signal strength (RSSI, RSRP, RSRQ) and identifiers (Cell ID, MAC address, UUID) from each signal source (S320). In particular, the terminal extracts the cell ID of the base station currently connected to the terminal from the cellular signal among the signals collected in the above step (S310). This cell ID can be used as an indicator representing the approximate location of the terminal. In particular, it can serve as a criterion for the server to select an appropriate composite positioning-related fragment model.

[0046] Next, the terminal transmits the extracted cell ID to the positioning model server (S330). The data transmitted at this time is very small, consisting only of cell ID information, and does not include the terminal user's exact location information. This ensures the protection of personal information.

[0047] Then, the terminal receives a fragmented AI positioning model specialized for the region from the server (S340). This model is a part of the entire AI positioning model held by the positioning server and is a lightweight model that has learned only the signal patterns of the region. It is desirable that the model size be 1 / 100 to 1 / 1000 of the original model, a size that can be sufficiently processed by the terminal.

[0048] Next, the terminal calculates the location directly within the terminal using the received fragmented AI positioning model (S350). When the composite signal collected in step (S310) is input into the fragmented AI positioning model, the fragmented model can estimate the current location based on learned patterns. This entire process is performed in an on-device environment within the terminal. Accordingly, there is no network delay, and real-time processing is possible even when GPS is unavailable.

[0049] Additionally, the terminal can perform self-learning on one or more fragmented AI positioning models it possesses (S360). This is an optional step in which the fragmented models can be continuously updated using positioning results and collected signal information. In particular, by continuously learning signal patterns in areas frequently visited by the user, the positioning accuracy in those areas can be gradually improved. Since this learning process takes place only within the terminal, personal information is protected. In other words, the fact that there is no risk of personal information being leaked externally is an additional advantage.

[0050] FIG. 4 is a conceptual diagram illustrating the process of splitting the entire model and creating fragment models of a server according to one embodiment of the present invention.

[0051] Referring to FIG. 4, the positioning model server (220 in FIG. 2) may possess a global positioning AI model. The global positioning AI model is a large-scale model trained on vast amounts of nationwide data and has a size that can only be executed on a server. This model includes signal patterns from all regions, so it has high accuracy, but it is too large and complex to be executed directly on a terminal.

[0052] Meanwhile, the server performs preprocessing to partition the entire model by region through clustering and spatial indexing. The clustering algorithm involves grouping regions with similar signal patterns. Spatial indexing generates a grid structure based on geographical location. Efficient regional partitioning is possible by combining these two techniques.

[0053] More specifically, in grid-based regional partitioning, the server divides the entire region into a grid and assigns a unique identifier to each grid. The grid size is adaptively determined based on regional characteristics, and it can be configured so that urban areas are divided into small grids and rural areas into large grids. However, it is not necessarily limited to this. The grid form can be configured in various other ways.

[0054] Meanwhile, each grid possesses an independent model that reflects the signal characteristics of the corresponding region. When generating fragmented models, the server extracts partial models corresponding to each grid area. At this stage, model lightweighting techniques are applied to remove unnecessary parameters and preserve only the features important to that region. Consequently, each fragmented model is significantly smaller than the original model while providing sufficient accuracy for the respective region.

[0055] The server selects an appropriate fragment model based on the cell ID received from the terminal. Each cell ID is mapped to a specific region, and the server can quickly search for and match a fragment model for the region where the terminal is currently located using a region identifier or other index. The fragment model selected through this matching process is immediately transmitted to the terminal and used for on-device positioning.

[0056] Meanwhile, according to another embodiment of the present invention, the server includes a function to predict the movement pattern and direction of the terminal and to pre-transmit the necessary fragment model at the next location. This is intended to minimize the model replacement delay that occurs when the terminal moves across grid boundaries and to provide a seamless, continuous positioning service.

[0057] A movement pattern prediction module (not shown) comprehensively analyzes the terminal's past movement trajectory, current speed and direction, and movement characteristics by time period. It collects acceleration and angular velocity data from an inertial measurement unit (IMU) mounted in the terminal to track the terminal's movement in real time. These sensor data are processed through a Kalman filter or a particle filter to remove noise and extract the accurate direction of movement.

[0058] The movement direction prediction algorithm calculates the movement vector at the current location and, based on this, identifies grids that the terminal is likely to enter in the future. For example, if the terminal is currently located at grid A and is moving northeast at a speed of 5 km / h, northern grid B, eastern grid C, and northeast grid D are identified as potential entry grids. The entry probability for each grid is calculated by considering the angle difference with respect to the direction of movement, the distance to the grid, past movement patterns, etc.

[0059] Learning user-specific movement patterns improves prediction accuracy by learning an individual's daily movement paths. Commuting routes, frequently visited places, and movement patterns by day of the week are modeled through time series analysis. For example, if a pattern of leaving home at 8 a.m. on a weekday and traveling to work via a specific route is repeatedly observed, a model for the grids along that route during that time period is prepared in advance.

[0060] The multi-model caching strategy pre-downloads multiple fragment models along the predicted movement path and caches them on the terminal. Priorities are assigned in order of highest entry probability, and an appropriate number of models is maintained considering the terminal's storage space. Generally, 9 models within a 3×3 area around the current grid are cached by default, and this can be expanded to a 5×5 area if the movement speed is fast.

[0061] The adaptive prefetching algorithm dynamically adjusts the timing of model transmission by considering network conditions and movement speed. It pre-prepares a wider range of models during high-speed movement and reduces the caching scope during low-speed movement or stationary states to efficiently manage storage space. When network quality is good, it pre-downloads more models, while prioritizing the securing of only core models when the network is unstable.

[0062] The grid boundary proximity detection mechanism detects in real time that the terminal is approaching the boundary of the current grid. It calculates the estimated time to reach the boundary by comprehensively considering the distance from the grid center, the direction of movement, and the speed. When the estimated time to reach the boundary approaches within a threshold (e.g., 30 seconds), it immediately requests the adjacent grid model in that direction.

[0063] Model switching scheduling manages seamless model transitions when entering a new grid. By applying a double buffering technique, the next model is loaded into memory simultaneously while the current model is in use. The system switches to the new model immediately upon crossing a grid boundary, ensuring the continuity of the positioning service.

[0064] Contextual information is also utilized to improve the accuracy of travel path prediction. Unrealistic routes are excluded from predictions by considering geographical constraints such as time-based traffic patterns, road network information, and building layouts. For example, if driving on a highway, higher weights are assigned to grids oriented along the road, as there is a high probability of moving along the road.

[0065] History-based prediction correction continuously improves the algorithm by analyzing the accuracy of past predictions. It calculates the prediction error by comparing the predicted grid with the actual entered grid and adjusts the prediction model parameters based on this. Machine learning techniques are applied to learn individual, regional, and time-based characteristics to enhance prediction accuracy.

[0066] Bandwidth optimization strategies efficiently utilize network resources when transmitting multi-piece models. By applying delta encoding, the amount of transmitted data is reduced by excluding common parts between adjacent grids and transmitting only the differences. Additionally, compression algorithms are applied to further reduce model size, and low-priority models are downloaded gradually in the background.

[0067] The cache management policy applies LRU (Least Recently Used) or LFU (Least Frequently Used) algorithms to efficiently utilize limited storage space. Older or infrequently used models are automatically deleted, while models for frequently visited areas are stored in the persistent cache. By analyzing the range of users' daily activities, models for key areas are always maintained.

[0068] The real-time path recalculation feature responds rapidly to unexpected path changes. If a device deviates from the predicted path, it immediately performs a new prediction, deletes unnecessary cached models, and requests a model for the new path. This process runs in the background, ensuring that users do not experience any service interruptions.

[0069] Multipath prediction considers multiple possible paths simultaneously, rather than a single path. Since there is a possibility of moving in multiple directions at intersections or junctions, the probability for each direction is calculated, and models for the top N paths with the highest probabilities are prepared. This allows for an immediate response regardless of which direction is traveled.

[0070] Through this prediction-based advance transmission function, the present invention provides a smoother user experience. Model loading delays occurring when crossing grid boundaries are eliminated, enabling continuous positioning, and services can continue to be provided using cached models even if the network is temporarily interrupted. The effectiveness of this function is maximized, particularly during high-speed movement or in environments with unstable networks.

[0071] FIG. 5 is a conceptual diagram showing the structure of an AI positioning model according to an embodiment of the present invention. This may be the structure of an entire AI positioning model at a server, or the structure of a fragmented AI positioning model for on-device processing at a terminal. Both may have the same or similar structure.

[0072] Referring to Fig. 5, the fragmented on-device AI positioning model is largely composed of an input layer, a signal pattern learning layer, a location estimation layer, and an output layer. The input layer receives cellular signal characteristics, WiFi signal characteristics, BLE signal characteristics, and / or sensor information. Cellular signal characteristics include RSRP (Reference Signal Received Power), RSRQ (Reference Signal Received Quality), and Cell ID. WiFi signal characteristics may include RSSI (Received Signal Strength Indicator), MAC address, and SSID. BLE signal characteristics may include RSSI, UUID, and Major / Minor values. Sensor information may include accelerometer, gyroscope, and geomagnetic sensor data.

[0073] The signal pattern learning layer is composed of a deep learning-based neural network that recognizes patterns of input complex signals and extracts features. This layer consists of multiple hidden layers, each of which learns features at different levels. Lower layers recognize basic patterns of individual signals, while higher layers grasp the meaning of complex signal combinations.

[0074] The position estimation layer calculates the position probability distribution based on learned features. This layer calculates the probabilities for possible location candidates and determines the location with the highest probability as the final location. It also evaluates the reliability of the estimated location and provides information on the uncertainty of the positioning results.

[0075] The output layer finally outputs the estimated location coordinates (X, Y) and the confidence index. The location coordinates can be converted into actual map coordinates and provided to the user. The confidence index indicates the reliability of the estimated location. This information can be used by applications to assess the quality of the location data.

[0076] Meanwhile, due to the nature of the fragment model, its size is reduced to between 1 / 100 and 1 / 1000 of that of the original model. This is achieved by preserving only the parameters necessary for a specific region and removing unnecessary information from other regions. As such, the region-specific model is optimized for the signal patterns of that area, providing high accuracy despite its small size. Furthermore, the model structure is optimized for real-time processing, enabling rapid inference even with the limited computational power of the terminal. Accuracy can be continuously improved by learning user patterns through self-learning capabilities.

[0077] FIG. 6 is a flowchart illustrating a self-learning process according to one embodiment of the present invention.

[0078] Referring to Figure 6, the initial fragment model is a basic model received from a server that has learned the general signal patterns of the area. This model provides basic positioning accuracy, but it is not optimized for an individual's specific movement paths or frequently visited places.

[0079] Meanwhile, as the user moves, the terminal continuously collects complex signals and estimates the location using the current model. At this time, the collected signal information and the estimated location information can be utilized as data for self-learning.

[0080] Subsequently, the terminal verifies the accuracy of the estimated location through positioning result feedback. In environments where GPS is available, the error can be calculated by comparing it with the GPS location. Alternatively, the accuracy of the estimated location at the terminal can be verified by receiving error information regarding the on-device positioning results from the server. In particular, the validity of the location is evaluated by analyzing the user's movement patterns or duration of stay. This feedback information serves as a standard for model improvement.

[0081] Meanwhile, the terminal updates the fragment model by adjusting the model's weights based on collected data and feedback information. The terminal can gradually improve the fragment model using optimization algorithms such as gradient descent. Additionally, normalization techniques to prevent overfitting may be applied. As mentioned above, since this entire process takes place within the terminal, personal information is not leaked externally.

[0082] The fragment model improved through the above process provides enhanced accuracy through self-learning. In particular, it can demonstrate significantly higher accuracy in places frequently visited by the user or along regular travel routes. This model is then used for signal collection and positioning, and is continuously improved through successive learning cycles.

[0083] The main features of the self-learning according to the embodiment of FIG. 6 include privacy protection, adaptive learning, incremental improvement, regional optimization, and real-time updates. Since all learning is performed only within the terminal, there is no risk of the user's location information or movement patterns being leaked externally, and there is an advantage in building a personalized model that reflects the user's actual movement paths and visit patterns. In addition, stability is maintained by gradually improving the model without making large changes all at once. Practical accuracy improvement is achieved by prioritizing the learning of frequently visited areas. New data is reflected in the model as soon as it is collected, allowing for rapid adaptation to environmental changes.

[0084] FIG. 7 is a conceptual diagram showing grid-type region partitioning and adaptive grid size according to one embodiment of the present invention.

[0085] Referring to FIG. 7, as described in the embodiment of FIG. 4, the entire service area is divided into a grid, and each grid has an independent piece model. The size of the grid is adaptively determined according to the characteristics of the region, which is the result of considering both positioning accuracy requirements and model efficiency.

[0086] For example, urban areas (M1-1 to M1-9) (high-density areas) can be divided into small grids of 100m × 100m. Urban areas require precise positioning due to their dense buildings, complex signal environments, and high user density. By using small grids, the detailed signal characteristics of each area can be accurately modeled. Additionally, since many WiFi APs and BLE beacons are installed in urban areas, precise positioning utilizing rich signal information is possible.

[0087] In another example, suburban areas (M2-1 to M2-9) (medium-density areas) are divided into medium-sized grids of 500m × 500m. Suburbs have lower building density than urban centers and relatively simpler signal environments. Using medium-sized grids allows for efficient management of model size while providing an appropriate level of accuracy. The grid size can also be adjusted according to use, such as residential or commercial areas.

[0088] In another example, rural areas (M3-1 to M3-6) (low-density areas) are divided into large grids of 1 km × 1 km. Rural areas have sparse buildings and rely primarily on cellular base station signals for positioning. Even with large grids, sufficient accuracy can be provided, and storage space and transmission bandwidth are saved by minimizing the model size. In special terrains such as mountainous or coastal areas, the grid shape is adjusted to reflect terrain characteristics.

[0089] Each grid has a unique model ID and is connected to a cell ID through a mapping table. Alternatively, a region ID, cell ID, and grid ID may be connected to each other through a mapping table.

[0090] When a terminal transmits a specific cell ID, the server refers to the mapping table to find the model ID of the corresponding grid and sends the corresponding fragment model to the terminal. This process is carried out very quickly, allowing the terminal to receive a new fragment model whenever it moves to a different region.

[0091] Factors considered when determining grid size include base station density, population density, terrain complexity, and service requirements. In areas with a high density of base stations, smaller grids are used to perform precise positioning. Smaller grids are applied in areas with a large number of users because higher accuracy is required. Areas with many buildings or complex terrain require detailed modeling due to complex signal patterns. Grid size is adjusted according to service characteristics, such as emergency rescue or indoor navigation.

[0092] FIG. 8 is an exemplary diagram illustrating a WiFi Direct-based proximity tracking function according to one embodiment of the present invention.

[0093] Referring to FIG. 8, the above-described WiFi Direct-based proximity tracking function is a technology that determines the proximity distance using direct WiFi signals between two terminals, and enables distance measurement between terminals without separate infrastructure.

[0094] The tracked terminal (810) periodically generates a signal by activating WiFi Direct mode. WiFi Direct is a technology that enables the terminal to act as an AP and communicate directly with other terminals. The tracked terminal broadcasts a beacon signal containing a unique identifier, and this signal reaches a range of about 30m around it.

[0095] The tracker terminal (820) scans surrounding WiFi Direct signals to detect the signal of the target terminal. When a signal is detected, the RSSI value is measured to determine the signal strength. RSSI is an indicator of the strength of the received signal, and has a higher value the closer the distance.

[0096] At this point, the distance determination module converts the measured RSSI value into distance. The relationship between RSSI and distance is modeled through a path loss model, and accuracy is improved by applying correction factors based on the environment. Generally, the range of -30dBm to -40dBm is estimated to be 0-5m, -40dBm to -50dBm to 5-10m, -50dBm to -60dBm to 10-20m, and -60dBm to -70dBm to 20-30m.

[0097] The proximity notification feature provides a notification to the user when the estimated distance reaches a set threshold. The threshold can be adjusted according to the purpose, and multi-level notifications can also be set. For example, it can be configured with a first notification when entering within 30m, a second notification when approaching within 10m, and an emergency notification when approaching within 5m.

[0098] Meanwhile, in one embodiment of the present invention, the accuracy of proximity tracking can be improved by combining a WiFi Direct signal with a BLE advertising signal. While WiFi Direct signals provide a stable signal over a relatively long range (approximately 30m), they have limitations in precise distance measurement at short distances. To compensate for this, it is desirable to utilize BLE advertising signals simultaneously. BLE signals operate at low power while enabling more precise distance measurement at short distances within 50 meters. In particular, in the closest proximity situation within 10m, changes in the RSSI of the BLE signal respond more sensitively than those of the WiFi Direct signal, allowing for more accurate distance estimation. In such a dual-signal method, WiFi Direct and BLE advertising signals are measured simultaneously, and the final distance can be calculated by applying different weights based on distance. For example, hybrid distance estimation is performed by assigning a high weight to the WiFi Direct signal when the measured distance is 10m or more, and a high weight to the BLE signal when it is within 10m. Through this, consistent and accurate distance measurement is possible across the entire tracking range.

[0099] The key advantages of this feature are infrastructure independence, compatibility with both indoor and outdoor environments, bidirectional tracking, and low-power operation. It operates through direct communication between devices without the need for a separate access point (AP) or server. It can be used indoors where GPS is unavailable and functions identically outdoors. Since both devices can track each other, mutual location verification is possible. WiFi Direct supports a low-power mode to minimize battery consumption.

[0100] By utilizing the above-mentioned functions, it is possible to perform tasks such as preventing children from getting lost, protecting dementia patients, tracking pets, finding tourists who have strayed during group tours, and issuing warnings about approaching dangerous areas.

[0101] FIG. 9 is an exemplary diagram showing a personal protection application linkage system according to one embodiment of the present invention.

[0102] Referring to Fig. 9, this system provides the function of rapidly identifying the user's location in emergency situations and transmitting it to rescue agencies. The personal protection application consists of a location tracking module and an emergency detection module. The location tracking module tracks the user's location in real time through an on-device positioning SDK. Even in environments without GPS signals, the accurate location can be determined through composite signal-based positioning. The emergency detection module detects emergency situations through manual activation by the user or automatic pattern analysis.

[0103] The on-device positioning SDK is a software development kit that provides positioning capabilities to applications. It incorporates fragmented AI models, enabling location calculation without a server connection. The SDK offers simple APIs, allowing application developers to easily integrate positioning functions.

[0104] The location information transmission module automatically transmits location data in the event of an emergency. It enables a rapid response by simultaneously sending data to pre-configured emergency contacts and rescue agencies. The transmitted information includes the current location, time, and type of emergency.

[0105] The emergency contact database stores contact information for family members, guardians, designated persons, and others. Notifications are automatically sent to these contacts, which the user has registered in advance, in emergency situations. Notification priority and transmission methods can be set for each contact.

[0106] The emergency rescue agency server is integrated with public rescue systems such as 112 and 119. It transmits location information via standardized protocols and supports the rapid dispatch of rescue agencies. Designed to be compatible with existing emergency rescue systems, it does not require the construction of separate infrastructure.

[0107] The operation process is as follows. The on-device positioning SDK tracks the location in real time even without GPS. When the application detects an emergency, it recognizes the situation by allowing the user to manually press an emergency button or by automatically detecting abnormal patterns. The location information transmission module automatically sends location data to emergency contacts and rescue agencies. It enables rapid rescue by providing an accurate location even in areas where GPS is unavailable, such as indoors or underground.

[0108] According to this system, the ability to operate independently of GPS is effective. In particular, it provides accurate location even in places where GPS signals do not reach, such as indoors, underground, and tunnels. Furthermore, it can determine location through on-device positioning even in situations where network connections are unstable. In terms of privacy protection, since location calculations are performed on the device, location information is not normally stored on a server.

[0109] Its applications are highly diverse. In child protection services, it tracks the location of children and provides warnings when they approach dangerous areas. In elderly care services, it monitors the safety of the elderly living alone or patients with dementia. In industrial safety management, it identifies the location of workers in hazardous workplaces and supports rapid rescue in the event of an accident.

[0110] FIG. 10 is a conceptual diagram showing a pipeline for building and operating a fragment positioning model for on-device according to one embodiment of the present invention.

[0111] Referring to Fig. 10, this pipeline is a system that systematically manages the entire process from signal acquisition to model deployment.

[0112] The composite signal collector performs the role of collecting composite signal data from across the country. This device operates using dedicated collection equipment or a crowdsourcing method and can simultaneously collect at least two of cellular base station signals, WiFi AP signals, and BLE signals. The collected data is recorded with timestamps along with location information, allowing for the tracking of signal changes over time.

[0113] The positioning optimization server processes collected raw data and trains AI models. During the raw data parsing and cleaning process, errors in the collected data are removed and converted into a standard format. In the clustering and spatial indexing process, data is classified by region and spatial indexes are generated. In the multi-source positioning model training process, a deep learning model is trained using complex signals as input.

[0114] The positioning model server stores and manages trained models and provides appropriate fragment models based on requests from terminals. The on-device model transmission unit selects and transmits the fragment model for the corresponding region based on the cell ID received from the terminal. Through the model version management function, the update history of models for each region is managed, and a rollback to a previous version is possible if necessary.

[0115] Each stage of the data pipeline is automated, so new data is automatically processed and the model is updated as soon as it is collected. This enables a rapid response to changes in the signaling environment and allows for continuous improvement of service quality. During the model optimization process, various techniques are applied to optimize model size and performance. Model size is reduced by removing low-importance neurons using pruning and decreasing weight precision through quantization. Furthermore, knowledge distillation is used to transfer knowledge from large models to smaller ones, maintaining high performance even in lightweight models.

[0116] The pipeline monitoring system monitors the processing status of each stage in real time. It evaluates system performance by tracking metrics such as data collection rate, model training progress, and service response time. When abnormal signs are detected, it automatically generates alerts to enable a rapid response.

[0117] An on-device hybrid positioning system according to one embodiment of the present invention overcomes the limitations of existing server-based positioning systems and offers various advantages. It enables real-time positioning without network delay, allowing for immediate location identification even in emergency situations. It operates even in environments where communication is impossible, enabling the provision of services in disaster situations or underground spaces. Personal information is processed exclusively within the terminal, ensuring thorough protection of privacy. It provides customized accuracy for each user through self-learning, and performance improves with continued use. Server load is significantly reduced, allowing for efficient support of a large number of users.

[0118] Furthermore, as previously explained, the on-device composite positioning according to one embodiment of the present invention can be applied to various application fields. In emergency rescue services, rapid location identification is possible even in environments where GPS is not operational, thereby increasing the probability of rescue within the golden hour.

[0120] The system or device described above may be implemented as a hardware component, a software component, and / or a combination of a hardware component and a software component. For example, the system, device, and component described in the embodiments may be implemented using one or more general-purpose or special-purpose computers, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing unit may execute an operating system (OS) and one or more software applications executed on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing unit may be described as being used as a single unit, but those skilled in the art will understand that the processing unit may include a plurality of processing elements and / or a plurality of types of processing elements. For example, the processing unit may include multiple processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are also possible.

[0121] Software may include computer programs, code, instructions, or a combination of one or more of these, and may configure a processing unit to operate as desired or command the processing unit independently or collectively. Software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium or device, or transmitted signal wave so as to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over networked computer systems and may be stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

[0122] The method according to the embodiments may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either alone or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the embodiments, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operation of the embodiments, and vice versa.

[0123] Although the embodiments have been described above with reference to limited examples and drawings, those skilled in the art can make various modifications and variations from the description above. For example, suitable results can be achieved even if the described techniques are performed in a different order than described, and / or the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0124] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below.

Claims

Claim 1 An on-device composite positioning method using multiple wireless signals at a terminal, comprising: a step of collecting a composite wireless signal including at least two of a cellular base station signal, a WiFi AP signal, and a BLE signal; a step of extracting cellular cell identification information (cellular cell ID) from the collected composite wireless signal; a step of transmitting the extracted cellular cell identification information to a server; a step of receiving from the server a distilled on-device AI positioning model corresponding to a region related to the terminal's current location, which is generated by the server from a whole AI positioning model learned based on the composite wireless signal data; and a step of determining the current location directly at the terminal based on the collected composite wireless signal using the received on-device AI positioning model; wherein the distilled on-device AI positioning model is provided to the terminal by the server analyzing the cellular cell identification information. Claim 2 In claim 1, the fragmented on-device AI positioning model is a method for on-device composite positioning in which the entire positioning model of the server is divided into grid-shaped regional units through at least one of clustering and spatial indexing algorithms. Claim 3 In claim 2, the on-device composite positioning method wherein the grid shape is adaptively determined according to the region. Claim 4 In claim 1, the fragmented on-device AI positioning model is composed of the collected composite wireless signal and sensor information, an on-device composite positioning method. Claim 5 An on-device composite positioning method according to claim 1, further comprising the step of self-learning the fragmented on-device AI positioning model using the positioning result through the fragmented on-device AI positioning model and the collected composite wireless signal. Claim 6 In claim 5, the fragmented on-device AI positioning model has different positioning precision by region through self-learning, an on-device composite positioning method. Claim 7 An on-device composite positioning method according to claim 1, further comprising the step of receiving movement information of the terminal from a sensor, wherein, based on the movement information, the direction of movement of the terminal is predicted and a fragmented on-device AI positioning model of an adjacent grid corresponding to the direction of movement is received in advance. Claim 8 An on-device composite positioning method according to claim 1, further comprising: a step of generating a WiFi AP signal in WiFi Direct mode; a step of detecting the generated WiFi AP signal at another terminal; a step of measuring the signal strength of the detected WiFi AP signal; a step of determining the proximity distance between the terminal and the other terminal based on the measured signal strength; and a step of outputting a signal based on the determined proximity distance. Claim 9 In claim 8, the on-device composite positioning method further comprises the step of detecting a BLE advertising signal generated from the other terminal; and the step of measuring the signal strength of the detected BLE advertising signal, wherein the step of determining the proximity distance includes the step of determining the proximity distance between the terminal and the other terminal by combining the signal strength of the WiFi AP signal and the signal strength of the BLE advertising signal. Claim 10 An on-device composite positioning method according to claim 1, further comprising: a step of tracking the location of the terminal in real time in an application equipped with the fragmented on-device AI positioning model; and a step of the application automatically transmitting the real-time tracked current location information to a pre-designated contact in response to detecting the occurrence of an emergency. Claim 11 An on-device composite positioning terminal using multiple wireless signals, comprising: a signal collection unit for collecting a composite wireless signal including at least two of a cellular base station signal, a WiFi AP signal, and a BLE signal; a signal processing unit for extracting cellular cell identification information (cellular cell ID) from the collected composite wireless signal; a communication unit for transmitting the extracted cellular cell identification information to a server and receiving from the server a distilled on-device AI positioning model corresponding to a region related to the terminal's current location, which is generated by the server from a whole AI positioning model learned based on the composite wireless signal data; a storage unit for storing the received on-device AI positioning model; and a positioning processing unit for determining the current location directly at the terminal based on the collected composite wireless signal using the on-device AI positioning model, wherein the distilled on-device AI positioning model is provided as a model corresponding to a region related to the terminal's current location by analyzing the cellular cell identification information at the server. Claim 12 A positioning server that supports on-device composite positioning of a terminal, comprising: a model storage unit that stores an entire AI positioning model learned based on composite wireless signal data including at least two of a cellular base station signal, a WiFi AP signal, and a BLE signal, by dividing the data into grid-shaped regional units and storing the data in the form of fragmented on-device AI positioning models; a communication unit that receives cellular cell identification information (cellular cell ID) from a terminal; and a model transmission unit that analyzes the received cellular cell identification information and transmits a fragmented on-device AI positioning model related to the current location of the terminal to the terminal, wherein the fragmented on-device AI positioning model is configured to enable positioning processing based on composite wireless signals directly on-device at the terminal.

Images