A Station AR Guidance Method Based on Graphical Encoding and Potential Field Evolution

Abstract

This invention relates to the field of station navigation technology, and particularly to a station AR navigation method based on graphic encoding and potential field evolution. The method includes: a fixed station terminal receiving navigation destination information input by a passenger and acquiring real-time passenger flow heatmap data and equipment operating status data; generating machine-readable graphic encoding based on the above information using a path planning model; a mobile terminal scanning the machine-readable graphic encoding to obtain a dynamically optimized navigation path and dynamic AR guidance information; the mobile terminal performing real-time six-degree-of-freedom pose calculation and environmental perception to obtain the current pose; converting the dynamically optimized navigation path to the mobile terminal's local coordinate system based on the current pose to generate an updated navigation path; and generating an AR navigation interface based on the dynamic AR guidance information and the updated navigation path, combined with the user's movement status and scene spatial relationships. This invention achieves accurate and intuitive station navigation guidance.

Description

Technical Field

[0001] This invention relates to the field of station guidance technology, and in particular to a station AR guidance method based on graphic encoding and potential field evolution. Background Technology

[0002] In the operation of rail transit stations, especially large stations such as transfer stations, passenger guidance and navigation is a core service requirement. Currently, existing station guidance systems mainly use static signs, text guidance, or voice Q&A based on keyword matching to guide passengers. Specifically, the system usually sets up fixed directional signs and electronic screens in places such as platforms and passageways in advance, or triggers preset text responses through simple voice commands to provide passengers with route information.

[0003] Furthermore, traditional navigation guidance is typically presented in a static or fixed-refresh format (such as providing a complete path text or fixed arrows all at once). Passengers are highly susceptible to missing turning points or deviating from their intended direction while walking, leading to inefficient navigation, frequent getting lost, and severely impacting the travel experience. While some research has focused on intelligent optimization methods in the rail transit field—for example, patent application CN112381298A (classification G06Q) provides an intelligent navigation and self-service system for airport passengers, patent application CN112163701A (classification G06Q) provides a station hub transfer management method and device, and patent application CN119783935A (classification G06Q) provides an AI-based adaptive navigation optimization method and system for scenic areas—these methods primarily focus on macro-level traffic prediction, operational adjustments, or offline route planning.

[0004] Therefore, there is an urgent need for a station AR guidance method based on graphic encoding and potential field evolution. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] In view of the above-mentioned shortcomings and deficiencies of the existing technology, the present invention provides a station AR navigation method based on graphic encoding and potential field evolution. It solves the technical problems of low efficiency and poor accuracy in passenger navigation at rail transit stations in the existing technology.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0009] This invention provides a station AR (Augmented Reality) guidance method based on graphic encoding and potential field evolution, executed by a station fixed terminal and a mobile terminal, specifically including:

[0010] The station's fixed terminal receives navigation destination information input by passengers and acquires real-time passenger flow heat data and equipment operation status data. Based on the navigation destination information, passenger flow heat data, and equipment operation status data, the station's fixed terminal uses a path planning model to generate machine-readable graphic codes. The machine-readable graphic codes include dynamically optimized passenger guidance routes and dynamic AR guidance information containing congestion avoidance prompts and equipment status prompts.

[0011] The mobile terminal scans the machine-readable graphic code to obtain dynamically optimized passenger guidance routes and dynamic AR guidance information including congestion avoidance prompts and equipment status prompts;

[0012] The mobile terminal performs six-degree-of-freedom pose calculation and environmental perception in real time to obtain the current pose; based on the current pose, the dynamically optimized guide path is converted to the local coordinate system of the mobile terminal to generate an updated guide path; and the mobile terminal generates an AR navigation interface that is stably updated with the user's movement based on the dynamic AR guidance information and the updated guide path, combined with the user's movement status and the spatial relationship of the scene.

[0013] Optionally, in some embodiments of this application, the navigation destination information includes:

[0014] The destination identifier, described in natural language, is entered by the passenger through the interactive interface of the station's fixed terminal;

[0015] And / or the destination option selected by the passenger from the list of station facilities and locations provided by the station's fixed terminal; the destination identifier or destination option corresponds to the coordinates of a specific functional area or facility location predefined in the station's high-precision map.

[0016] Optionally, in some embodiments of this application, the station fixed terminal generates machine-readable graphic code based on the navigation destination information, passenger flow heat data, and equipment operating status data, using a path planning model, including:

[0017] The navigation destination information, pedestrian flow heat map data, and equipment operation status data are processed to construct a multimodal cost tensor.

[0018] The multimodal cost tensor is input into the path planning model, and the output is a sequence of guide route points that have avoided congestion and faulty equipment, as well as AR prompt labels bound to each route point;

[0019] The route path sequence and AR prompt labels are jointly encoded according to the minimum information redundancy rule, and a machine-readable graphic code is generated through graphic encoding.

[0020] Optionally, in some embodiments of this application, the data processing of the navigation destination information, pedestrian flow heat map data, and equipment operating status data to construct a multimodal cost tensor specifically includes:

[0021] Obtain a high-precision map of the station where the station fixed terminal is located, divide the high-precision map into multiple grid units, and assign a crowd density value and equipment status label to each grid unit based on pedestrian flow heat data and equipment operation status data;

[0022] A nonlinear saturation mapping method is used to convert the population density value of each grid cell into a dynamic congestion cost, thereby generating a dynamic congestion cost field.

[0023] The passage cost of the grid cell containing the faulty device in the device operation status data is set to infinity, and the passage cost of the grid cell containing the normal device is set to zero, thereby generating a passable semantic mask.

[0024] Based on the navigation destination information, each grid cell is assigned a gravitational potential energy value that decreases with distance from the destination. The closer the distance, the greater the gravitational potential energy, thus generating a gravitational potential field centered on the destination.

[0025] The dynamic congestion cost field, the passable semantic mask, and the gravitational potential field are superimposed grid by grid to obtain the multimodal cost tensor.

[0026] Optionally, in some embodiments of this application, a nonlinear saturation mapping method is used to convert the crowd density value of each grid cell into a dynamic congestion cost, and the generation of the dynamic congestion cost field specifically includes:

[0027] Based on the population density value of each grid cell and the preset maximum congestion cost, obtain the basic congestion cost for each grid cell;

[0028] Obtain the crowd density value of each grid cell across consecutive frames, and obtain the density change rate of each grid cell based on the crowd density value of each grid cell across consecutive frames;

[0029] Based on the density change rate, the dynamic adjustment amount of the half-saturation threshold is obtained. Based on the dynamic adjustment amount of the half-saturation threshold, the basic congestion cost of each grid cell is adjusted, the final dynamic congestion cost of each grid cell is obtained, and the dynamic congestion cost field is obtained.

[0030] Optionally, in some embodiments of this application, the path planning model includes:

[0031] An initial potential field construction layer is used to initialize the value of each grid cell in the multimodal cost tensor to the initial potential energy of that grid cell, thereby constructing an initial potential field.

[0032] The potential field evolution layer is used to simulate the spatiotemporal interaction between the dynamic congestion cost field, the passable semantic mask, and the gravitational potential field. The initial potential field is updated iteratively to obtain the converged potential field. In each iteration, the potential energy is redistributed according to the potential energy difference between adjacent grid cells, and the weight of the local congestion cost is dynamically adjusted according to the population density change rate of each grid cell, so that the potential field gradually converges to a smooth distribution.

[0033] The path backtracking layer is used to obtain the navigation destination grid cell and the grid cell where the fixed terminal is located, and to backtrack from the navigation destination grid cell along the direction of the fastest potential energy decrease to the grid cell where the fixed terminal is located, and extract the grid cells passed through in sequence as the guide path point sequence.

[0034] A label attachment layer is used to generate an AR prompt label bound to each path point in the route path point sequence, based on the final dynamic congestion cost and the passable semantic mask.

[0035] Optionally, in some embodiments of this application, the mobile terminal performs six-degree-of-freedom pose calculation and environmental awareness in real time, and obtains the current pose by including:

[0036] The mobile terminal continuously captures image frames through its own camera and obtains inertial measurement unit data; when the machine-readable graphic code appears in the image frame, it decodes and obtains the absolute coordinates of the machine-readable graphic code in the high-precision map of the station, as well as the relative pose of the machine-readable graphic code relative to the mobile terminal.

[0037] The mobile terminal extracts feature points and performs inter-frame matching on image frames, and combines the data from the inertial measurement unit to perform visual-inertial tight coupling calculation to obtain continuous pose changes.

[0038] The mobile terminal uses the relative pose as an absolute reference to integrate and calibrate the continuous pose changes, eliminate cumulative drift, and obtain the current pose.

[0039] Optionally, in some embodiments of this application, the process of converting the dynamically optimized travel path to the local coordinate system of the mobile terminal based on the current pose and generating the updated travel path specifically includes:

[0040] Obtain the global coordinates of each path point in the dynamic optimized passenger guidance path in the station's high-precision map; based on the current pose, transform the global coordinates of each path point to the local coordinate system with the mobile terminal as the origin to obtain the local coordinate sequence;

[0041] Based on the distance of each coordinate point in the local coordinate sequence to the origin, the retention density of path points is adjusted, and the connection between adjacent path points is smoothed to generate an updated guide path.

[0042] Optionally, in some embodiments of this application, the mobile terminal generates an AR navigation interface that is stably updated as the user moves, based on the dynamic AR guidance information and the updated navigation path, combined with the user's movement status and the spatial relationship of the scene. This specifically includes:

[0043] Based on the updated navigation route, the mobile terminal overlays a 3D navigation guide line onto the real-time image of its own camera, and generates auxiliary markers at the corresponding positions of the 3D navigation guide line according to the congestion avoidance prompts and device status prompts in the dynamic AR guidance information.

[0044] In addition, the system acquires the user's movement speed and the spatial depth map of the current scene, dynamically adjusts the display length and visual saliency of the 3D navigation guide line according to the user's movement speed, performs occlusion rendering on the 3D navigation guide line according to the spatial depth map, and acquires the updated 3D navigation guide line.

[0045] Based on the updated 3D navigation guide lines and auxiliary markers, an AR navigation interface is generated that updates stably as the user moves.

[0046] Optionally, in some embodiments of this application, the mobile terminal is a smartphone with a camera, an inertial measurement unit and a depth sensing module, or augmented reality glasses with see-through display function and environmental depth perception capability.

[0047] (III) Beneficial Effects

[0048] The beneficial effects of this invention are as follows: This invention provides a station AR navigation method based on graphic encoding and potential field evolution. It generates graphic encoding containing dynamically optimized paths and congestion / equipment prompts by fusing passenger flow heatmaps and equipment status data through a fixed station terminal. A mobile terminal scans and decodes this encoding, performs six-degree-of-freedom pose calculation, and transforms the path to the local coordinate system, generating a stable and followable AR navigation interface. Compared to existing technologies, it does not rely on a continuous network connection; a single scan is sufficient to obtain a dynamically optimized path. Through congestion avoidance and equipment status prompts, it can proactively guide passengers away from high-density areas and faulty facilities, alleviating station congestion and improving traffic efficiency. Simultaneously, the six-degree-of-freedom pose calculation and local coordinate system transformation significantly improve navigation accuracy, solving the problems of virtual information jitter and offset in traditional AR navigation. This achieves precise alignment and stable following of AR guidance with the real-world scene, thereby enhancing the robustness of the navigation system and the user experience. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating a station AR guidance method based on graph encoding and potential field evolution according to an embodiment of this application;

[0050] Figure 2This is a schematic diagram illustrating the process of obtaining the final graphic code by a station fixed terminal according to one embodiment of this application;

[0051] Figure 3 This is a schematic diagram illustrating the process of constructing a multimodal cost tensor according to an embodiment of this application;

[0052] Figure 4 This is a schematic diagram illustrating the process of a mobile terminal obtaining its current pose according to an embodiment of this application. Detailed Implementation

[0053] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0054] In related technologies, AR guidance at stations usually relies on mobile terminals to continuously access wireless networks or cloud servers to obtain navigation routes in real time, or uses fixed QR codes to store static guidance information, or relies only on GPS and Bluetooth beacons for rough positioning, without fully considering the dynamic changes in passenger flow and equipment operating status within the station.

[0055] However, this existing solution has the following technical problems in practical applications: First, it is highly dependent on real-time networks, and navigation interruptions or delays are likely to occur when there are dense crowds at stations or when network signals fluctuate; second, static coding cannot dynamically adjust the path according to real-time congestion or equipment failure, making it difficult to effectively avoid high-risk areas; third, the virtual guidance information in traditional AR navigation is not accurate enough in matching the real scene, and it is prone to jitter and deviation as the user's posture changes, making it difficult to follow the path.

[0056] Therefore, the core technical problem this application aims to solve is: how to achieve dynamic adaptive path optimization within a station without relying on continuous network communication, while simultaneously ensuring high accuracy and stable alignment of AR navigation guidance. To address this core technical problem, this application adopts the following technical solution: A fixed terminal at the station collects real-time passenger flow heatmap and equipment status data, combines this with the passenger's input destination, and uses a path planning model to generate a machine-readable graphic code containing dynamically optimized paths, congestion avoidance, and equipment status prompts; after scanning this code, the mobile terminal obtains its current pose through six-degree-of-freedom pose calculation and environmental perception, transforms the dynamic path to the local coordinate system, and then generates an AR navigation interface that stably updates as the user moves. Through this solution, the following technical effects can be achieved: dynamically optimized guidance paths can be obtained without continuous network connection, significantly improving navigation accuracy, achieving precise alignment and stable following of AR virtual guidance with the real scene, and effectively guiding passengers to avoid congestion and malfunctioning equipment, thereby improving station throughput and user experience.

[0057] Additionally, it should be noted that this application's embodiments are applied to the typical scenario of subway stations: subway stations typically experience dense crowds with significant tidal changes, fluctuating network signals, complex transfer channels, and a lack of GPS signals. By deploying fixed terminals at subway station entrances / exits, transfer nodes, and platforms, real-time thermal data and the operational status of equipment such as escalators and turnstiles are collected. Combined with the passenger's input destination, a graphical code containing congestion avoidance and fault indication is dynamically generated. After scanning the code with a mobile device such as a smartphone, passengers can obtain high-precision, stable AR navigation guidance without a continuous internet connection. This effectively addresses the needs of managing large passenger flows during peak hours, detours for temporary equipment malfunctions, and guidance for complex transfer routes, thereby significantly improving the operational safety of subway stations and passenger travel efficiency.

[0058] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0059] Figure 1 This is a flowchart illustrating a station AR guidance method based on graph encoding and potential field evolution according to an embodiment of this application. Figure 1 As shown, the AR passenger guidance system at this station is implemented by both fixed and mobile terminals, specifically including:

[0060] The station's fixed terminal receives navigation destination information input by passengers and acquires real-time passenger flow heat data and equipment operation status data. Based on the navigation destination information, passenger flow heat data, and equipment operation status data, the station's fixed terminal uses a path planning model to generate machine-readable graphic codes. The machine-readable graphic codes include dynamically optimized passenger guidance routes and dynamic AR guidance information containing congestion avoidance prompts and equipment status prompts.

[0061] In the specific implementation process, station fixed terminals are deployed at key nodes such as subway station entrances and exits, transfer nodes, and next to automatic ticket vending machines. Their physical form is a standing information kiosk, a wall-mounted screen, or a robot.

[0062] Specifically, the station's fixed terminal includes at least a human-computer interaction module, a data communication module, and a data processing module. The human-computer interaction module is used to receive navigation destination information input by passengers via a touch screen or voice recognition. The data communication module is used to access the station's IoT sensing system in real time to obtain passenger flow heat data generated based on video analysis, WiFi / Bluetooth probes, and turnstile access data, as well as real-time operating status data of equipment such as elevators, turnstiles, and platform screen doors. The data processing module is used to generate robot-readable graphic codes.

[0063] With the above structure, the station's fixed terminal can provide passengers with real-time, dynamic, and low-latency passenger guidance code generation services without relying on the cloud.

[0064] The navigation destination information includes:

[0065] The destination identifier, described in natural language, is entered by the passenger through the interactive interface of the station's fixed terminal;

[0066] And / or the destination option selected by the passenger from the list of station facilities and locations provided by the station's fixed terminal; the destination identifier or destination option corresponds to the coordinates of a specific functional area or facility location predefined in the station's high-precision map.

[0067] In other words, navigation destination information includes two input methods. The first is for passengers to input their destination identifier in natural language through the interactive interface of a station-based terminal, such as "I want to go to the accessible elevator," "Go to platform 3," "Go to Exit B," or "Find a power bank rental shop." The station's built-in natural language processing module extracts key intents from the statement and semantically matches them with predefined functional areas or facility locations in the station's high-precision map. The second method is for passengers to directly select their destination from a list of station facilities and locations provided by the terminal. The list categorizes facilities such as entrances, platforms, transfer passages, restrooms, mother-and-baby rooms, customer service centers, automatic ticket machines, and shops. Passengers can then click on a specific option. Both the destination identifier derived from natural language parsing and the selected destination option from the list correspond to the coordinates of a specific functional area or facility location predefined in the station's high-precision map.

[0068] For example, suppose a high-precision map of a subway station predefines the following locations: the boarding area for the Line 2 train heading towards Exit A is located in the middle of the first basement level, the accessible elevator near Exit B is located in the northeast corner of the first floor, and the customer service center is located in the central lobby of the first floor. Scenario 1: A passenger says "Take me to the Line 2 train heading towards Exit A" in front of a standing information kiosk. The terminal, after natural language parsing, recognizes "Line 2," "towards Exit A," and "train," automatically matching the coordinates of the middle of the first basement level. Scenario 2: A passenger clicks on "Accessible Facilities" in the list on the robot terminal and then selects "Accessible Elevator (near Exit B)." The terminal directly obtains the coordinates of the northeast corner of the first floor. All the above destination coordinates are passed to the path planning model as target points for dynamically optimizing the passenger guidance path. These two input methods can adapt to different passenger expression habits and operational preferences, improving the usability of the passenger guidance system.

[0069] Optionally, navigation destination information can also be obtained through voice input. To adapt to the complex environment of subway stations, a sound processing model based on Mel spectrum and convolutional neural networks is adopted. The specific process is as follows: First, the voice signal input by the passenger is preprocessed by pre-emphasis, framing, and windowing. Then, a set of Mel filters converts the preprocessed voice feature sequence into a Mel spectrogram, which has a fixed size of 128×128 pixels. The Mel frequency conversion is based on the characteristics of human hearing, performing a non-linear mapping on the original voice frequencies to achieve higher resolution in the low-frequency part. The converted Mel spectrogram is then fed into a three-layer convolutional neural network for processing. The first and second layers use 3×3 convolutional kernels with a stride of one, and the third layer uses 2×2 convolutional kernels with the same stride of one. Through these three convolutional operations, audio semantic features related to navigation intent are gradually extracted. After the convolutional layers, a fully connected layer is connected, finally outputting a fixed-dimensional audio feature vector with a dimension of 512, which can effectively represent the semantics of the passenger's navigation request. To further improve the model's recognition accuracy in noisy subway station environments, a noise suppression module was added to the model input. This module dynamically filters out background noise such as announcements, crowd noise, and train sounds, ensuring the robustness of subsequent feature extraction. Through the aforementioned structure combining Mel-frequency spectrum and CNN, this embodiment can accurately identify natural voice navigation requests such as "I want to go to the accessible elevator" or "Go to platform 3," and automatically map them to the corresponding target coordinates on the station's high-precision map, thereby providing passengers with a more convenient and touchless interactive experience.

[0070] Furthermore, the station fixed terminal, based on the navigation destination information, passenger flow heat map data, and equipment operating status data, utilizes a path planning model to generate machine-readable graphical encoding, including (e.g.) Figure 2 (as shown)

[0071] S10. Process the navigation destination information, pedestrian flow heat data and equipment operation status data to construct a multimodal cost tensor;

[0072] In the specific implementation process, see Figure 3 S10 includes:

[0073] S11. Obtain a high-precision map of the station where the station fixed terminal is located, divide the high-precision map into multiple grid units, and assign a crowd density value and an equipment status label to each grid unit based on the pedestrian flow heat data and the equipment operation status data.

[0074] The high-precision map of the station is pre-stored in the station's fixed terminal. This map contains information on all the physical structures inside the station, such as the location coordinates and connectivity of entrances and exits, platforms, transfer passages, stairs, escalators, elevators, turnstiles, customer service centers, restrooms, shops, and other facilities.

[0075] In addition, when dividing the high-precision map of the station, an adaptive grid division strategy is adopted: smaller grid side lengths (e.g., 0.5 meters) are used in areas where crowds tend to gather (such as turnstiles and platform screen doors) to improve local density resolution; larger grid side lengths (e.g., 2 meters) are used in long, straight passageways with simple passage attributes to reduce computational overhead. Each grid cell is assigned a unique index number and its geometric center coordinates in the global coordinate system, as well as its floor information, are recorded.

[0076] Furthermore, in the specific implementation process, the pedestrian density value is the estimated number of people in each grid cell at the current moment, divided by the area of ​​that grid cell. Equipment operation status data is obtained from the station's IoT system and identified as normal or faulty; areas without equipment are labeled as empty, thus obtaining the equipment status label for each grid cell.

[0077] S12. A nonlinear saturation mapping method is adopted to convert the crowd density value of each grid cell into a dynamic congestion cost, generating a dynamic congestion cost field. This mapping is not a simple linear proportion, but considers the nonlinear increase in the difficulty of passage caused by density itself. That is, when the density is low, the impact of congestion cost is small, and when the density exceeds a certain threshold, the congestion cost rises sharply. At the same time, the density change rate in the time dimension is introduced to dynamically adjust the half-saturation threshold, so that the congestion cost can reflect the crowd gathering trend in real time. For example, the cost is increased in advance in areas that are becoming crowded, and the cost is appropriately reduced in areas that are being evacuated.

[0078] Specifically, S12 includes:

[0079] Based on the population density value of each grid cell and the preset maximum congestion cost, the basic congestion cost of each grid cell is obtained. This transformation adopts a non-linear saturation mapping in the form of Sigmoid: when the density is much less than the half-saturation threshold, the basic congestion cost is close to zero; when the density is much greater than the half-saturation threshold, the basic congestion cost is close to the preset maximum value.

[0080] Secondly, the station's fixed terminal updates the pedestrian flow heat map data at regular intervals to obtain the crowd density value of each grid unit in consecutive frames. Based on the crowd density value of each grid unit in consecutive frames, the density change rate of each grid unit is obtained, which is the density difference between the current frame and the previous frame divided by the time interval. A positive value indicates that the density is increasing and tending to be congested, while a negative value indicates that the density is decreasing and tending to be dispersed.

[0081] Finally, based on the density change rate, a dynamic adjustment amount for the half-saturation threshold (proportional to the density change rate) is obtained. Based on this dynamic adjustment, the base congestion cost of each grid cell is adjusted to obtain the final dynamic congestion cost of each grid cell, and a dynamic congestion cost field is obtained. In other words, when density increases, the half-saturation threshold decreases, meaning that the congestion cost at the same current density will be pushed up in advance, guiding path planning to avoid the area as early as possible. When density decreases, the half-saturation threshold increases, and the congestion cost at the same current density is reduced because the population is dispersing, and the actual passage difficulty may be lower than the level implied by the static density. This adjustment of the base congestion cost of each grid cell yields the final dynamic congestion cost of each grid cell. The final dynamic congestion costs of all grid cells together constitute the dynamic congestion cost field.

[0082] In the specific implementation process, the final formula for obtaining the dynamic congestion cost is:

[0083] ;

[0084] Where C represents the final dynamic congestion cost, C max The maximum congestion cost is preset, k is the steepness coefficient, and ρ is the maximum cost of congestion. half The half-saturation threshold is ρ, where ρ is the population density value. α represents the density change rate, and α is a preset adjustment coefficient.

[0085] For example, suppose there is a grid cell with a side length of 1 meter and an area of ​​1 square meter in a subway station transfer hall. The preset maximum congestion cost is 100, and the initial half-saturation threshold is 2 people / m². 2 The density change rate adjustment factor is set to 0.5, and the update interval is five seconds. Initially, the current density of this grid is 1.5 people / m². 2 The calculated basic congestion cost is approximately 26.9. If the area is becoming congested, the previous frame density was 1 person / m².2 The density change rate is 0.1 people / m³. 2 The half-saturation threshold decreased, and the recalculated dynamic congestion cost was approximately 28.9, about 2.0 higher than the base cost, indicating that the system increased the passage cost in advance. Conversely, if the area was being evacuated, the density in the previous frame was 2 people / m². 2 The density change rate is -0.1 people / m³. 2 As the half-saturation threshold increases, the final dynamic congestion cost is approximately 25.0, a decrease of about 1.9 from the base cost, indicating that the system perceives a reduction in congestion risk. Through the aforementioned nonlinear saturation mapping and dynamic adjustment of the density change rate, the resulting dynamic congestion cost field can more accurately and sensitively reflect the actual passage difficulty of people within the station, providing robust cost input for subsequent route planning.

[0086] S13. Set the passage cost of the grid cell containing the faulty equipment in the equipment operation status data to infinity, indicating that the grid is impassable in path planning, thus forcing a detour around the faulty facility; set the passage cost of the grid cell containing the normal equipment to zero, indicating that these grids are not subject to additional penalties in terms of traversability semantics, thereby generating a traversability semantic mask; the generated traversability semantic mask is essentially a binary cost layer of the same size as the map grid: grids corresponding to infinity are obstacles, and grids corresponding to zero are free passage areas. The function of this mask is to physically exclude areas that cannot be used due to equipment failure or maintenance in subsequent path calculations, avoiding the planned path from passing through these impassable locations.

[0087] S14. Based on the navigation destination information, assign a gravitational potential energy value to each grid cell according to the distance to the destination, which decreases with distance. The closer the distance, the greater the gravitational potential energy, and generate a gravitational potential field centered on the destination.

[0088] Specifically, firstly, the navigation destination information is mapped to specific coordinates in the high-precision map of the station, and the grid cell to which these coordinates belong is determined as the target grid. Then, for each grid cell in the map, the distance from its geometric center to the center of the target grid is calculated. The distance can be calculated using Euclidean distance (straight-line distance) or Manhattan distance (a broken line distance along the grid axis), with Euclidean distance being more commonly used in practice to reflect the approximation of a straight line in space. Based on this, each grid is assigned a gravitational potential energy value that decreases with distance. The basic rule is that the closer the distance, the greater the gravitational potential energy, and the farther the distance, the smaller the gravitational potential energy. Commonly used decay function forms include inverse proportional decay (potential energy is inversely proportional to distance) or exponential decay (potential energy decreases exponentially with distance). This embodiment uses inverse proportional decay. To effectively guide the search direction in path planning, the change in gravitational potential energy is usually made relatively smooth, ensuring that the potential energy tends to zero or a small value when far from the destination, and increases rapidly when approaching the destination. Ultimately, the gravitational potential energy values ​​of all grids constitute a potential field that gradually weakens outward from the destination. This field acts similarly to the attraction field in the artificial potential field method, guiding the optimization path toward the endpoint.

[0089] S15. The dynamic congestion cost field, the passable semantic mask, and the gravitational potential field are superimposed grid by grid to obtain the multimodal cost tensor.

[0090] In the specific implementation process, for each grid cell in the map, its dynamic congestion cost in the dynamic congestion cost field, its passage cost (infinity or zero) in the passable semantic mask, and its gravitational potential energy value in the gravitational potential field are superimposed according to preset weight coefficients. Since the dynamic congestion cost and gravitational potential energy have different dimensions, a balance factor is introduced here for normalization or weighted summation. A typical superposition method is: total cost = dynamic congestion cost × w1 + gravitational potential energy × w2. If the cost of the grid in the passable semantic mask is infinite, the total cost is directly set to infinity (not passable). Among them, w1 and w2 are adjustable hyperparameters, determined according to the actual operational needs of the station. For example, a higher weight can be given to the congestion cost to prioritize avoiding congested areas, or a higher weight can be given to the gravitational potential energy to shorten the path length. After the grid-by-grid calculation is completed, the total cost of all grids constitutes a three-dimensional or two-dimensional tensor (depending on whether the map includes floor dimensions), which is the multimodal cost tensor. It integrates real-time crowd dynamics (congestion cost), equipment availability (semantic mask), and destination attractiveness (gravitational potential field), providing comprehensive input features for subsequent route planning models, enabling the generated guiding routes to avoid congestion and failures while quickly reaching the destination.

[0091] S20. Input the multimodal cost tensor into the path planning model, and output the sequence of guide path points that have avoided congestion and faulty equipment, as well as AR prompt labels bound to each path point;

[0092] The path planning model includes:

[0093] The initial potential field construction layer is used to initialize the value of each grid cell in the multimodal cost tensor to the initial potential energy of that grid cell, thus constructing the initial potential field. In the initial potential field, the total cost of each grid cell is the initial potential energy of that grid. For grids with infinite costs (i.e., grids containing faulty or under-maintenance equipment), their initial potential energy is also set to infinite, indicating that they are impassable. For normal grids, the initial potential energy is equal to the total cost after the aforementioned weighted summation, and its value is usually normalized to between 0 and 1. After the initial potential field is constructed, the entire station map is mapped to a two-dimensional or three-dimensional potential energy distribution map, where low potential energy areas correspond to locations with low passage costs and close to the destination, and high potential energy areas correspond to locations with severe congestion or far from the destination.

[0094] The potential field evolution layer is used to simulate the spatiotemporal interaction between the dynamic congestion cost field, the passability semantic mask, and the gravitational potential field. It updates the initial potential field iteratively to obtain the converged potential field. This layer borrows ideas from thermal diffusion or anisotropic diffusion equations. In each iteration, potential energy is redistributed based on the potential energy difference between adjacent grid cells, and the weights of local congestion costs are dynamically adjusted according to the rate of change of population density in each grid cell, causing the potential field to gradually converge to a smooth distribution.

[0095] The redistribution rule is as follows: if the potential energy of the current grid is higher than the average potential energy of its neighboring grids, then some potential energy is transferred to the neighboring grids; otherwise, potential energy is absorbed from the neighboring grids. This diffusion process gradually smooths the potential field, eliminating local abrupt changes, thus generating a natural and smooth low-potential energy channel from the starting point to the ending point. Specifically, the amount of potential energy transferred is determined by both the diffusion coefficient and the potential energy difference, i.e.:

[0096] ;

[0097] Among them, P new Let P be the potential energy after the current grid update. old β represents the potential energy of the current grid, and β is the diffusion coefficient. This represents the average potential energy of adjacent grid cells.

[0098] For each adjacent grid, the potential energy gained or lost from the current grid is distributed accordingly.

[0099] Furthermore, before the potential energy redistribution, the weight of congestion cost in the potential energy is adjusted based on the change in population density of each grid cell. In this way, areas with rapid density increases maintain higher potential energy during the diffusion process, guiding paths to actively detour; areas with rapid density decreases have lower potential energy, encouraging paths to pass through.

[0100] After each full-grid potential energy update, the root mean square of the potential energy change for all grids is calculated. If this change is less than a preset threshold (e.g., the relative change in potential energy for all grids is less than one-thousandth, or the absolute change is less than 10), then the value is considered valid. -6 When the potential field converges to a smooth distribution, the iteration stops.

[0101] Through the above quantified rules, the potential field evolution layer can explicitly perform potential energy redistribution and dynamic weight adjustment to ensure the generation of smooth, adaptive crowd avoidance paths.

[0102] The path backtracking layer is used to obtain the navigation destination grid cell and the grid cell where the fixed terminal is located, and to backtrack from the navigation destination grid cell along the direction of the fastest potential energy decrease to the grid cell where the fixed terminal is located, and to extract the grid cells passed through in sequence as the route point sequence.

[0103] Specifically, let the grid where the station's fixed terminal is located be the starting grid, and the grid where the navigation destination is located be the target grid. In the converged potential field, the potential energy of the target grid is the global minimum, while the potential energy of the starting grid is relatively high. The backtracking algorithm starts from the target grid, takes it as the current grid, and adds it to the path sequence. Then, it iteratively performs the following operations: obtain all neighboring grids of the current grid, including the vertical directions (up, down, left, right, and diagonal), while considering vertical connections between floors (such as grids containing stairs and escalators). From these neighboring grids, select grids whose potential energy is less than the current grid's potential energy and whose potential energy is not infinite. If multiple grids exist, select the one with the smallest potential energy as the next grid; if multiple grids have the same minimum potential energy, prioritize the grid with the smallest angle to the starting direction. Insert the selected grid into the head of the path sequence and set it as the current grid. Repeat the above process until the current grid equals the starting grid. If a neighbor with lower potential energy cannot be found during the backtracking process, trigger backtracking failure handling, such as appropriately relaxing the conditions or re-evolving the potential field. After backtracking is completed, a grid sequence from the starting point to the destination is obtained. To further smooth the path, collinear redundant intermediate points can be deleted, and only necessary turning points can be retained. Finally, the guide path point sequence is output, and each point contains global coordinates, the floor it belongs to, and the direction of travel.

[0104] A label attachment layer is used to generate an AR prompt label bound to each path point in the route path point sequence, based on the final dynamic congestion cost and the passable semantic mask.

[0105] In the specific implementation process, each path point in the path point sequence is traversed to obtain its dynamic congestion cost, equipment status label, directional change angle relative to adjacent path points, and whether it contains special facilities (such as escalators, stairs, elevators, turnstiles, entrances / exits). Labels are generated according to preset rules: if the dynamic congestion cost exceeds the high congestion threshold (e.g., 60% of the maximum cost), a message such as "Congested area ahead, please slow down" or "Detour recommended" is generated; if it exceeds the medium congestion threshold but is below the high threshold, a message such as "High passenger flow, please give way" is generated. If the equipment status is "Congested," a message such as "This equipment is busy, expected waiting time is long" is generated; if the equipment status is "Faulty" or "Under maintenance" and the path point is adjacent to a faulty device, a message such as "Equipment faulty ahead, please use other channels" is generated. When the directional change angle reaches 15 degrees or more, directional indicator labels such as "Turn left ahead," "Turn right ahead," "Go upstairs," or "Go downstairs" are generated based on the turning direction, and a warning label is generated several meters before the turning point (e.g., two grids in advance). For the destination grid, a message "Destination reached" is generated, along with a prompt based on the specific facility type. If the same waypoint meets multiple prompt conditions simultaneously, the prompts are displayed in the following order: fault prompts take precedence over congestion prompts, congestion prompts take precedence over directional indicators, and directional indicators take precedence over destination prompts. They can also be displayed together. All AR prompt labels, along with their corresponding coordinates, floors, display timing, and other information, are packaged and output for real-time overlay display on the AR interface of mobile terminals, thus providing passengers with dynamic, accurate, and semantically rich navigation guidance.

[0106] S30. The route point sequence and AR prompt labels are jointly encoded according to the minimum information redundancy rule, and machine-readable graphic codes are generated through graphic encoding.

[0107] The minimum information redundancy rule is specifically as follows:

[0108] First, differential compression is performed on the pathpoint sequence. This involves recording the absolute value of the starting point's coordinates, and storing only the coordinate differences between subsequent pathpoints and the previous point (such as row offset, column offset, and floor change indicators). Since the distance between adjacent pathpoints within a station is usually small, differential encoding can significantly reduce the data volume. For direction changes, only the angle change value of the turning point is recorded, rather than the direction of each point. Second, AR prompt labels are stored in association according to their bound pathpoint indices. For multiple prompt labels on the same pathpoint, they are merged into a composite label according to a preset priority (e.g., "Turn right ahead, area congested") to avoid duplicate header information from multiple independent labels. In addition, for the same prompt content on consecutive pathpoints (e.g., multiple consecutive grids marked "high passenger flow"), run-length encoding is used, recording the starting position and continuous length instead of storing it repeatedly point by point. After completing the above differential and run-length compression, a checksum and version identifier are added to the compressed binary data to form the data stream to be encoded. Finally, a high-fault-tolerant graphic encoding standard is used to convert this data stream into machine-readable graphic encoding. The size and module density of the graphic encoding can be dynamically adjusted according to the screen resolution of the station's fixed terminals and the scanning distance of the passenger's mobile terminals, ensuring fast and accurate recognition under common lighting and distance conditions. Through the minimum information redundancy rule, the data volume after joint encoding is significantly reduced, allowing a single graphic encoding to accommodate complete dynamic guidance paths and rich AR prompts, while maintaining high decoding robustness.

[0109] This embodiment presents a station AR passenger guidance method based on graphic coding and potential field evolution. A fixed station terminal acquires real-time passenger flow heatmap and equipment operating status data through an IoT system. Combined with the passenger's input destination, it dynamically generates machine-readable graphic codes containing congestion avoidance and fault indications using a locally deployed path planning model. This terminal can independently complete multi-source data fusion, potential field evolution calculation, and code generation without relying on a continuous network connection or cloud server, significantly reducing dependence on external communication conditions and improving the system's robustness and availability in complex station environments. Simultaneously, through adaptive grid partitioning, nonlinear congestion cost mapping, and passability semantic masks, the terminal can quickly respond to changes in passenger flow and equipment status within the station, providing passengers with instant, low-latency, and dynamically optimized passenger guidance codes. This effectively alleviates local congestion and avoids faulty facilities, thereby improving station operational safety and passenger guidance service efficiency.

[0110] Next, the station AR guidance method based on graph encoding and potential field evolution in this embodiment also includes:

[0111] The mobile terminal scans the machine-readable graphic code to obtain dynamically optimized passenger guidance routes and dynamic AR guidance information including congestion avoidance prompts and equipment status prompts.

[0112] It is important to understand that the mobile terminal here refers to a smartphone with a camera, inertial measurement unit, and depth sensing module, or augmented reality glasses with see-through display function and environmental depth perception capability.

[0113] In the specific implementation process, the mobile terminal activates its camera and points it at the machine-readable graphic code displayed on the station's fixed terminal. It automatically identifies the encoded area, completes image acquisition, binarization, positioning, and error correction decoding, and obtains a compressed binary data stream. Then, it decompresses the data following the reverse process of the minimum information redundancy rule: first, it parses the version identifier and checksum to verify data integrity; then, it performs run-length decoding to restore continuously repeating prompt labels; next, it performs differential decoding to reconstruct a complete sequence of guide route points based on the difference between the starting point's absolute coordinates and subsequent coordinates. Each route point includes global coordinates, its floor, and direction of travel; finally, it reconstructs the AR prompt label list according to the route point index, including congestion avoidance prompts, equipment status prompts, direction indicators, and destination prompts. At this point, the mobile terminal obtains complete dynamically optimized guide routes and dynamic AR guidance information. Simultaneously, the mobile terminal stores the acquired dynamically optimized guide routes and dynamic AR guidance information in its local memory, allowing for independent AR navigation without subsequent scanning or continuous network connection.

[0114] Through the above detailed steps, the mobile terminal can quickly and accurately obtain all the navigation data from a single graphic code, providing reliable input for subsequent pose calculation and AR rendering.

[0115] Furthermore, the mobile terminal performs six-degree-of-freedom pose calculation and environmental perception in real time to obtain the current pose; based on the current pose, the dynamically optimized guidance path is converted to the local coordinate system of the mobile terminal to generate an updated guidance path; and the mobile terminal generates an AR navigation interface that is stably updated as the user moves, based on the dynamic AR guidance information and the updated guidance path, combined with the user's movement status and the spatial relationship of the scene.

[0116] Specifically, see Figure 4 The mobile terminal performs real-time six-degree-of-freedom pose calculation and environmental perception to obtain the current pose, including:

[0117] S41. The mobile terminal continuously captures image frames through its own camera and obtains inertial measurement unit data; when the machine-readable graphic code appears in the image frame, it decodes and obtains the absolute coordinates of the machine-readable graphic code in the high-precision map of the station, and obtains the relative pose of the machine-readable graphic code relative to the mobile terminal.

[0118] S42. The mobile terminal extracts feature points and performs inter-frame matching on the image frames, and combines the data from the inertial measurement unit to perform visual-inertial tight coupling calculation to obtain continuous pose changes.

[0119] S43. The mobile terminal uses the relative pose as an absolute reference to integrate and calibrate the continuous pose change, eliminate cumulative drift, and obtain the current pose.

[0120] Specifically, step S42 includes the following:

[0121] S421. The mobile terminal uses the ORB algorithm to quickly detect feature points in the currently acquired image frame and generates a "descriptor" for each feature point. This descriptor can be understood as a short binary code used to uniquely identify the local texture information of the feature point. In subsequent matching, the similarity of the descriptors can be compared to determine whether two feature points correspond to the same point in space.

[0122] S422. Perform pairwise matching between feature points from the previous frame and feature points from the current frame, and calculate the difference between descriptors (measured by Hamming distance between descriptors). The two points with the smallest difference are considered a candidate match. Additionally, compare the ratio of the "difference of the nearest neighbor" to the "difference of the second nearest neighbor". If this ratio is less than 0.7, it indicates that the nearest neighbor is significantly better than the second nearest neighbor, and the match can be considered reliable; if the ratio is greater than or equal to 0.7, it indicates high texture repetition and a high risk of mismatch, and the match is discarded. Subsequently, the RANSAC algorithm is used for further filtering and elimination, finally obtaining all visual feature matching results.

[0123] S423. Perform pre-integration on the inertial measurement unit data to obtain the IMU integration results, namely the predicted values ​​of relative rotation, relative velocity, and relative displacement between adjacent frames.

[0124] S424. The visual feature matching results are fused with the IMU integration results by using a sliding window-based nonlinear optimization or extended Kalman filter to calculate the continuous pose changes between adjacent frames in real time, including the rotation matrix and translation vector.

[0125] Taking nonlinear optimization as an example: The system maintains a sliding window containing several consecutive image frames. For each pair of adjacent frames within the window, two types of residual terms are constructed. The first type is the visual reprojection residual, which is obtained by projecting the 3D points onto the image plane using the feature point pairs matched between the two frames and the 3D map points obtained through triangulation, based on the currently estimated camera pose, and subtracting the actual pixel coordinates from the projected values ​​to obtain the visual residual. The second type is the IMU pre-integration residual, which is obtained by pre-integrating the IMU angular velocity and acceleration data between two frames based on the IMU integration results to obtain the predicted values ​​of relative rotation, velocity, and displacement, and then subtracting these values ​​from the theoretical values ​​of the pose, velocity, etc., to be estimated in the optimization variables to obtain the IMU residual.

[0126] Furthermore, the visual residual and IMU residual are weighted and summed to construct a joint cost function. The weights are typically determined based on the observation noise of the visual feature points and the noise covariance matrix of the IMU sensor. Through iterative solving, the cost function is minimized, thereby simultaneously optimizing the camera pose (rotation matrix and translation vector), velocity, and IMU bias for each frame within the window. After optimization, the rotation matrix and translation vector between adjacent frames represent the required continuous pose changes.

[0127] In the implementation of this application, on the one hand, IMU pre-integration provides high-frequency motion prediction, maintaining the continuity of pose estimation even when visual features are briefly lost (such as when facing a white wall or a sudden change in lighting); on the other hand, visual reprojection residuals can effectively correct the drift caused by zero bias and noise in IMU integration, avoiding the rapid divergence of pose error over time. The resulting continuous pose change has the characteristics of high local accuracy, fast response to rapid motion, and small long-term drift, providing a solid foundation for subsequent integration calibration based on graphic coding as the absolute reference.

[0128] Furthermore, based on the current pose, the mobile terminal transforms the dynamically optimized route to its local coordinate system, generating an updated route that specifically includes:

[0129] S51. Obtain the global coordinates of each path point in the dynamic optimized passenger guidance path in the station's high-precision map; based on the current pose, transform the global coordinates of each path point to the local coordinate system with the mobile terminal as the origin, and obtain the local coordinate sequence; that is, calculate the offset vector relative to the mobile terminal for each path point in the passenger guidance path, and then rotate this vector from the global coordinate system to the local coordinate system of the mobile terminal, thereby obtaining the coordinate sequence of each path point in the local coordinate system.

[0130] S52. Based on the distance from each coordinate point in the local coordinate sequence to the origin, adjust the retention density of path points and smooth the connections between adjacent path points to generate an updated guide path.

[0131] Specifically, step S52 includes:

[0132] Using the mobile terminal's current location (the origin of the local coordinate system) as a reference, the Euclidean distance from each path point to the origin is calculated. Path points closer to the origin occupy a larger visual area in the AR view, and users are more sensitive to details; therefore, more points need to be retained to ensure path continuity and smooth transitions at turns. Path points farther away project a smaller area on the screen, and users have relatively lower requirements for path accuracy; therefore, they can be appropriately sparsified to reduce rendering load. The specific adjustment rules adopt a segmented strategy: setting a first distance threshold (e.g., 5 meters) and a second distance threshold (e.g., 15 meters). For path points less than the first threshold, all are retained to ensure detailed rendering of nearby paths. For path points between the first and second thresholds, they are retained at fixed intervals, such as retaining one every other point, or randomly downsampling according to a retention ratio (e.g., 60%). For path points greater than the second threshold, the sparsity is further increased, such as retaining one every two points, or sampling according to a retention ratio (e.g., 30%). In addition, regardless of distance, the following key points must be retained: turning points in the path (points with a change in direction of 15 degrees or more), points bound to AR prompt tags (such as congestion prompts, equipment failure prompts, directional indicators, etc.), and points associated with special facilities such as stairs, escalators, and turnstiles. Through this adaptive retention density adjustment, redundant path points in distant locations can be effectively reduced while ensuring the accuracy of nearby path guidance and preventing the loss of key information. This reduces the computational and rendering overhead of mobile terminals and improves the smoothness of AR navigation. The adjusted path point sequence is then combined with smoothing of adjacent point connections (such as Bézier curve interpolation or moving average filtering) to finally generate an updated navigation path.

[0133] Furthermore, the mobile terminal generates an AR navigation interface that is stably updated as the user moves, based on the dynamic AR guidance information and the updated navigation path, combined with the user's movement status and the spatial relationships of the scene. Specifically, this includes:

[0134] Based on the updated navigation route, the mobile terminal overlays a 3D navigation guide line onto the real-time image of its own camera, and generates auxiliary markers at the corresponding positions of the 3D navigation guide line according to the congestion avoidance prompts and device status prompts in the dynamic AR guidance information.

[0135] Specifically, for points along the path that are associated with congestion avoidance prompts (such as "crowded area ahead"), a yellow or red warning icon (such as a densely packed people icon or an exclamation mark) is generated above or to the side of the point, along with a floating text label. For equipment status prompts (such as escalator malfunction or turnstile congestion), corresponding equipment icons (such as an escalator cross or turnstile queue icon) and text descriptions are generated at the location of the corresponding facility. All auxiliary signs are updated in real time to their screen projection position according to the terminal's pose, ensuring that they are always accurately attached to their corresponding positions in the real scene.

[0136] In addition, the system acquires the user's movement speed and the spatial depth map of the current scene, dynamically adjusts the display length and visual salience of the 3D navigation guide line according to the user's movement speed, performs occlusion rendering on the 3D navigation guide line according to the spatial depth map, and obtains the updated 3D navigation guide line; when the user's movement speed is fast (e.g., greater than 1.5 meters per second), the guide line extends forward a longer distance (e.g., 30 meters ahead), and the lines are thicker and the colors are more vibrant, so that the user can see the direction in advance when walking quickly; when the user's movement speed is slow (e.g., less than 0.5 meters per second) or stationary, the guide line is only displayed for the nearest 10 to 15 meters, and the lines become thinner and the colors lighter to reduce visual interference.

[0137] Simultaneously, occlusion rendering is applied to the 3D navigation guide lines based on the spatial depth map: for each line segment or strip-shaped portion of the guide line, the depth map is used to determine whether it is occluded by real objects (such as pillars, walls, or other passengers). If the depth value of a segment of the guide line is greater than the depth map value of the corresponding pixel (i.e., the guide line is behind an object), that segment is cropped or made semi-transparent and blurred, displaying only the unoccluded portion; if the guide line is in front of an object, it is displayed normally. In this way, the navigation guide lines can naturally blend into the real scene without "penetrating" walls or people, thus enhancing realism and credibility.

[0138] The spatial depth map is generated in real time using the mobile terminal's own camera.

[0139] Based on the updated 3D navigation guide lines and auxiliary markers, an AR navigation interface is generated that stably updates as the user moves. Navigation ends on the mobile device once the user reaches their destination.

[0140] The station AR navigation method based on graphic encoding and potential field evolution in this embodiment integrates passenger flow heat data and equipment operating status in real time through station fixed terminals, and dynamically generates machine-readable graphic codes that include congestion avoidance and fault prompts by combining path planning models. This achieves low-latency navigation code generation without continuous network connection. After scanning and decoding by the mobile terminal, the path is accurately converted to the local coordinate system by using six-degree-of-freedom pose calculation and visual-inertial tight coupling fusion. The display length, visual saliency, and occlusion relationship of the AR guide line are adaptively adjusted according to the user's movement speed and scene depth map, generating a high-precision AR navigation interface that is stably updated as the user moves. Compared to existing technologies, this invention effectively reduces reliance on real-time network communication and can operate independently in crowded and signal-fluctuating station environments. By using dynamic congestion cost fields and passable semantic masks, it proactively guides passengers to avoid congested areas and faulty equipment, alleviating localized station congestion and improving facility utilization and traffic efficiency. At the same time, pose calculation and depth-aware rendering based on graphic encoding significantly improve navigation accuracy and realism, solving the problems of virtual information jitter, offset, and obstacle penetration in traditional AR navigation, thereby comprehensively improving the robustness, safety, and user experience of the station guidance system.

[0141] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0142] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0143] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "over," or "on top" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," or "beneath" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0144] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0145] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A station AR guiding and riding method based on graph coding and potential field evolution, characterized in that, The method, executed by a station fixed terminal and a mobile terminal, includes: The station's fixed terminal receives navigation destination information input by passengers and acquires real-time passenger flow heat data and equipment operation status data. Based on the navigation destination information, passenger flow heat data, and equipment operation status data, the station's fixed terminal uses a path planning model to generate machine-readable graphic codes. The machine-readable graphic codes include dynamically optimized passenger guidance routes and dynamic AR guidance information containing congestion avoidance prompts and equipment status prompts. The mobile terminal scans the machine-readable graphic code to obtain dynamically optimized passenger guidance routes and dynamic AR guidance information including congestion avoidance prompts and equipment status prompts; The mobile terminal performs six-degree-of-freedom pose calculation and environmental perception in real time to obtain the current pose; based on the current pose, the dynamically optimized guide path is converted to the local coordinate system of the mobile terminal to generate an updated guide path; and the mobile terminal generates an AR navigation interface that is stably updated with the user's movement based on the dynamic AR guidance information and the updated guide path, combined with the user's movement status and the spatial relationship of the scene.

2. The station AR guiding and riding method based on graph coding and potential field evolution according to claim 1, characterized in that, The navigation destination information includes: The destination identifier, described in natural language, is entered by the passenger through the interactive interface of the station's fixed terminal; And / or the destination option selected by the passenger from the list of station facilities and locations provided by the station's fixed terminal; the destination identifier or destination option corresponds to the coordinates of a specific functional area or facility location predefined in the station's high-precision map.

3. The station AR guiding and riding method based on graph coding and potential field evolution according to claim 2, characterized in that, The station's fixed terminal generates machine-readable graphical encoding based on the navigation destination information, passenger flow heat map data, and equipment operating status data, using a path planning model. The navigation destination information, pedestrian flow heat map data, and equipment operation status data are processed to construct a multimodal cost tensor. The multimodal cost tensor is input into the path planning model, and the output is a sequence of guide route points that have avoided congestion and faulty equipment, as well as AR prompt labels bound to each route point; The route path sequence and AR prompt labels are jointly encoded according to the minimum information redundancy rule, and a machine-readable graphic code is generated through graphic encoding.

4. The station AR guiding and riding method based on graph coding and potential field evolution according to claim 3, characterized in that, The process of processing the navigation destination information, pedestrian flow heatmap data, and equipment operating status data to construct a multimodal cost tensor specifically includes: Obtain a high-precision map of the station where the station fixed terminal is located, divide the high-precision map into multiple grid units, and assign a crowd density value and equipment status label to each grid unit based on pedestrian flow heat data and equipment operation status data; A nonlinear saturation mapping method is used to convert the population density value of each grid cell into a dynamic congestion cost, thereby generating a dynamic congestion cost field. The passage cost of the grid cell containing the faulty device in the device operation status data is set to infinity, and the passage cost of the grid cell containing the normal device is set to zero, thereby generating a passable semantic mask. Based on the navigation destination information, each grid cell is assigned a gravitational potential energy value that decreases with distance from the destination. The closer the distance, the greater the gravitational potential energy, thus generating a gravitational potential field centered on the destination. The dynamic congestion cost field, the passable semantic mask, and the gravitational potential field are superimposed grid by grid to obtain the multimodal cost tensor.

5. The station AR guidance method based on graphic encoding and potential field evolution according to claim 4, characterized in that, A nonlinear saturation mapping method is used to convert the population density values ​​of each grid cell into dynamic congestion costs, generating a dynamic congestion cost field that specifically includes: Based on the population density value of each grid cell and the preset maximum congestion cost, obtain the basic congestion cost for each grid cell; Obtain the crowd density value of each grid cell across consecutive frames, and obtain the density change rate of each grid cell based on the crowd density value of each grid cell across consecutive frames; Based on the density change rate, the dynamic adjustment amount of the half-saturation threshold is obtained. Based on the dynamic adjustment amount of the half-saturation threshold, the basic congestion cost of each grid cell is adjusted, the final dynamic congestion cost of each grid cell is obtained, and the dynamic congestion cost field is obtained.

6. The station AR guidance method based on graphic encoding and potential field evolution according to claim 5, characterized in that, The path planning model includes: An initial potential field construction layer is used to initialize the value of each grid cell in the multimodal cost tensor to the initial potential energy of that grid cell, thereby constructing an initial potential field. The potential field evolution layer is used to simulate the spatiotemporal interaction between the dynamic congestion cost field, the passable semantic mask, and the gravitational potential field. The initial potential field is updated iteratively to obtain the converged potential field. In each iteration, the potential energy is redistributed according to the potential energy difference between adjacent grid cells, and the weight of the local congestion cost is dynamically adjusted according to the population density change rate of each grid cell, so that the potential field gradually converges to a smooth distribution. The path backtracking layer is used to obtain the navigation destination grid cell and the grid cell where the fixed terminal is located, and to backtrack from the navigation destination grid cell along the direction of the fastest potential energy decrease to the grid cell where the fixed terminal is located, and extract the grid cells passed through in sequence as the guide path point sequence. A label attachment layer is used to generate an AR prompt label bound to each path point in the route path point sequence, based on the final dynamic congestion cost and the passable semantic mask.

7. The station AR guidance method based on graphic encoding and potential field evolution according to claim 1, characterized in that, The mobile terminal performs real-time six-degree-of-freedom pose calculation and environmental awareness to obtain the current pose, including: The mobile terminal continuously captures image frames through its own camera and obtains inertial measurement unit data; when the machine-readable graphic code appears in the image frame, it decodes and obtains the absolute coordinates of the machine-readable graphic code in the high-precision map of the station, as well as the relative pose of the machine-readable graphic code relative to the mobile terminal. The mobile terminal extracts feature points and performs inter-frame matching on image frames, and combines the data from the inertial measurement unit to perform visual-inertial tight coupling calculation to obtain continuous pose changes. The mobile terminal uses the relative pose as an absolute reference to integrate and calibrate the continuous pose changes, eliminate cumulative drift, and obtain the current pose.

8. The station AR guidance method based on graphic encoding and potential field evolution according to claim 1, characterized in that, Based on the current pose, the dynamically optimized travel path is transformed to the local coordinate system of the mobile terminal, and the updated travel path is generated, specifically including: Obtain the global coordinates of each path point in the dynamic optimized passenger guidance path in the station's high-precision map; based on the current pose, transform the global coordinates of each path point to the local coordinate system with the mobile terminal as the origin to obtain the local coordinate sequence; Based on the distance of each coordinate point in the local coordinate sequence to the origin, the retention density of path points is adjusted, and the connection between adjacent path points is smoothed to generate an updated guide path.

9. The station AR guidance method based on graphic encoding and potential field evolution according to claim 1, characterized in that, Based on the dynamic AR guidance information and the updated navigation path, combined with the user's movement status and the spatial relationships of the scene, the mobile terminal generates an AR navigation interface that is stably updated as the user moves. Specifically, this includes: Based on the updated navigation route, the mobile terminal overlays a 3D navigation guide line onto the real-time image of its own camera, and generates auxiliary markers at the corresponding positions of the 3D navigation guide line according to the congestion avoidance prompts and device status prompts in the dynamic AR guidance information. In addition, the system acquires the user's movement speed and the spatial depth map of the current scene, dynamically adjusts the display length and visual saliency of the 3D navigation guide line according to the user's movement speed, performs occlusion rendering on the 3D navigation guide line according to the spatial depth map, and acquires the updated 3D navigation guide line. Based on the updated 3D navigation guide lines and auxiliary markers, an AR navigation interface is generated that updates stably as the user moves.

10. The station AR guidance method based on graphic encoding and potential field evolution according to claim 1, characterized in that, The mobile terminal is a smartphone equipped with a camera, an inertial measurement unit, and a depth sensing module, or augmented reality glasses with see-through display function and environmental depth perception capability.

Images