Digital human navigation and surrounding service fusion method based on augmented reality
By constructing a high-dimensional spatiotemporal grid and a dynamic directed graph, real-time dynamic profiles are generated, solving the problems of data heterogeneity and information silos in vehicle navigation, enabling personalized service push, and improving user experience and security.
Patent Information
- Application Number
- CN202611318039.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-28
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, the service push of in-vehicle or mobile phone navigation lacks accuracy and personalization, and there are data heterogeneity and information silos, which leads to the distraction of users and increases driving safety hazards.
By acquiring users' historical service data, constructing a high-dimensional spatiotemporal grid, performing cross-domain correlation and feature completion, generating a dynamic directed graph, mining data causal logic chains and resolving conflicts, generating real-time dynamic profiles, and combining environmental and cognitive load to push personalized services.
It enables precise and personalized service delivery while ensuring driving safety, improving user experience and security, and solving the problem of multi-source data fusion.
Smart Images

Figure CN123118980A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, and in particular to a method for integrating augmented reality-based digital human navigation with surrounding services. Background Technology
[0002] With the rapid development of vehicle-to-everything (V2X) and navigation technologies, providing users with various third-party services such as car washing, refueling, charging, entertainment, and dining has become a crucial aspect of enhancing the user navigation experience. The user experience of in-car or mobile navigation systems has a significant impact on the efficiency and satisfaction of users' driving trips.
[0003] In existing technologies, user-facing service push notifications are typically handled independently by various third-party service applications, or mechanically pushed based solely on a single user's vehicle location information. Traditional push methods suffer from several drawbacks. First, severe data silos and information barriers exist between different third-party platforms, resulting in highly heterogeneous and spatiotemporally fragmented data. This hinders deep correlation and integration, preventing the system from accurately understanding the user's true behavioral intentions and even leading to logical misjudgments when multiple data sources conflict, resulting in inaccurate service content pushed to users. Second, existing push mechanisms are rigid, lacking the ability to perceive the vehicle's real-time physical environment and the user's current cognitive load while driving. Regardless of the user's driving status, fixed text and image pop-ups on the in-vehicle or mobile phone screen can easily cause information overload and push fatigue, severely distracting the user and posing significant driving safety hazards. Therefore, how to integrate navigation to provide timely and personalized service push notifications while ensuring driving safety has become an urgent problem to solve. Summary of the Invention
[0004] The purpose of this invention is to provide a method for integrating digital human navigation and surrounding services based on augmented reality, so as to solve the problems mentioned in the background art.
[0005] This application provides a method for integrating augmented reality-based digital human navigation with surrounding services, the method comprising: Obtain the user's historical service data and map the historical service data to a preset spatiotemporal high-dimensional grid to obtain a standardized spatiotemporal behavior vector; Extract the time decay coefficient and data source reputation score from the historical service data, perform cross-domain association and feature completion on the spatiotemporal behavior vector to obtain the completed behavior vector; Based on the completed behavior vector, a dynamic directed graph is constructed, and the data causal logic chain in the dynamic directed graph is mined to generate a scene behavior intent graph. The scenario behavior intent graph is subjected to conflict self-resolution processing based on the reputation score of the data source to obtain the target behavior intent graph, and a real-time dynamic profile of the user is generated based on the target behavior intent graph. Based on the real-time dynamic profile, the environmental status, vehicle status, and cognitive load around the user are collected in real time. When the environmental status, vehicle status, and cognitive load all meet the preset thresholds, a push wake-up command is triggered. Based on the push wake-up command, the optimal solution for the target service is obtained by optimizing and sorting the multi-source services around the user. The interaction mode is determined by combining the optimal solution for the target service and the cognitive load, and personalized service is pushed according to the interaction mode. The implicit feedback data after the personalized service is pushed is collected, and behavioral intent evolution parameters are generated based on the implicit feedback data. The generation process of the scene behavioral intent map is then subjected to long-term intervention and evolution based on the behavioral intent evolution parameters.
[0006] Preferably, the step of acquiring the user's historical service data and mapping the historical service data to a preset spatiotemporal high-dimensional grid to obtain a standardized spatiotemporal behavior vector is as follows: Acquire users' historical transaction records, behavior trajectories, and underlying perception logs from in-vehicle devices on multiple different third-party service platforms, and aggregate them to generate historical service data; Based on the historical service data, extract the data occurrence time information and the platform attribution tag from the historical service data, mark the data occurrence time information as a timestamp feature, and mark the platform attribution tag as a data source; Construct a virtual three-dimensional coordinate system that includes a time dimension axis, a spatial dimension axis, and a service type dimension axis, and divide the virtual three-dimensional coordinate system into multiple uniform spatiotemporal high-dimensional grids according to a preset granularity; The historical service data is projected into the spatiotemporal high-dimensional grid according to the timestamp feature and the data source, respectively. Extract all service feature information aggregated in each of the spatiotemporal high-dimensional grids, and obtain a standardized spatiotemporal behavior vector based on the service feature information.
[0007] Preferably, the steps of extracting the time decay coefficient and data source reputation score from the historical service data, performing cross-domain correlation and feature completion on the spatiotemporal behavior vector to obtain the completed behavior vector are as follows: Extract the time difference between the timestamp feature and the current physical time, and substitute the time difference into a preset exponential decay model to obtain the time decay coefficient; The data source is matched and compared with a preset platform reputation mapping rating table to obtain the original reputation weight corresponding to the data source, and the original reputation weight is used as the data source reputation score. Cross-comparison of the spatiotemporal behavior vectors from different data source platforms yields cross-domain association features between the different spatiotemporal behavior vectors. Based on the time decay coefficient, the cross-domain association features are subjected to time-sensitivity downweighting, and based on the data source reputation score, the cross-domain association features are subjected to authenticity upweighting, resulting in calibrated association features. The associated features are used as compensation data to complete the spatiotemporal behavior vector, resulting in a completed behavior vector.
[0008] Preferably, the steps of constructing a dynamic directed graph based on the completed behavior vector, mining the data causal logic chain in the dynamic directed graph, and generating a scene behavior intent graph are as follows: Extract the geographic location transition trajectory and consumption behavior sequence from the completed behavior vector, and use the completed behavior vector as the basic node of the graph structure, and the geographic location transition trajectory and the consumption behavior sequence as directed edges between the nodes; Based on the interconnections between the basic nodes and the directed edges, an initial structure that can characterize the evolution of user behavior over a continuous time period is constructed, resulting in a dynamic directed graph. Identify the connected subgraphs in the dynamic directed graph, obtain the occurrence frequency of the connected subgraphs, and determine whether the occurrence frequency exceeds a preset frequency threshold; If it is determined that the occurrence frequency exceeds the frequency threshold, then the pseudo-correlated edges in the connected subgraph are removed, and the target directed edges with sequential causal relationships are extracted. Identify the first and last basic nodes of the target directed edge, and encapsulate the target directed edge and the first and last basic nodes to obtain the data causal logic chain; By acquiring different scenario dimensions of users' lives, classifying and reorganizing the causal logic chain of the data according to the different scenario dimensions, and merging them to generate a scenario behavior intent map of users with potential service needs in different scenarios.
[0009] Preferably, the step of performing conflict self-resolution processing on the scene behavior intent graph based on the data source reputation score to obtain the target behavior intent graph, and generating a real-time dynamic profile of the user based on the target behavior intent graph, specifically includes: A global traversal scan of the scene behavior intent graph is performed to identify and extract two diametrically opposed behavioral logic branches that have contradictory derivations or point to the same spatiotemporal node, and these behavioral logic branches are marked as conflict nodes. Extract the source data reputation score corresponding to the source data of the conflict node, and evaluate the credibility probability of the behavioral logic branch based on the source data reputation score to obtain the confidence level of the behavioral logic branch. Based on the confidence level, the pseudo-logic branches with lower confidence levels are identified and discarded, while the true logic branches with higher confidence levels are retained. The target behavior intent map is obtained based on the true logic branches. Extract user preference features, consumption level features, and price sensitivity features from the target behavior intent map, combine them into a multi-dimensional feature set, and generate a real-time dynamic profile of the user.
[0010] Preferably, the step of real-time collection of the user's surrounding environmental status, vehicle status, and cognitive load, and triggering a push wake-up command when the environmental status, vehicle status, and cognitive load all meet preset thresholds, specifically includes: The system collects real-time data on the external weather conditions at the user's current location and the traffic congestion index at the vehicle's current location, and combines the external weather conditions and the traffic congestion index to generate an environmental status. The system reads the controller bus data, current driving range, and operational health status of key components inside the vehicle in real time, and generates the vehicle status by combining the controller bus data, the current driving range, and the operational health status. The system acquires the vehicle's current speed, steering wheel angle change rate, and user eye-tracking data to assess the user's mental workload and attention distraction level while processing additional information. The system then combines the mental workload and attention distraction level to obtain the user's cognitive load. Based on the environmental state and the vehicle state, the user's current preference reference value is obtained, and it is determined whether the current preference reference value exceeds the preference tolerance pre-stored in the real-time dynamic profile. If it is determined that the current preference reference value exceeds the preference tolerance, then it is determined whether the cognitive load is lower than the preset safety interference threshold; If it is determined that the cognitive load is lower than the safety interference threshold, a push wake-up command is generated and sent out.
[0011] Preferably, the steps of optimizing and ranking multiple source services around the user to obtain the optimal solution for the target service, determining the interaction pattern by combining the optimal solution for the target service and the cognitive load, and then pushing personalized services based on the interaction pattern are as follows: Based on the push wake-up command, the current user location is obtained, and based on the current user location, a set of potential third-party service providers within a preset range that match the real-time dynamic profile is retrieved. Obtain the service price, route distance, and user history ratings for each potential third-party service provider in the set of potential third-party service providers. Combine the service price, route distance, and user history ratings to generate a comprehensive rating for each potential third-party service provider. The potential third-party service provider with the highest comprehensive score is selected as the target service provider, and the service provided by the target service provider is taken as the optimal solution for the target service. Based on the cognitive load, the interaction mode with the least interference to driving operation is obtained by matching among the preset pure voice broadcast mode, minimal floating window mode and rich text and image pop-up mode. The data content of the optimal solution of the target service is loaded into the display template of the interactive form and distributed to the user through the in-vehicle multimedia system.
[0012] Preferably, the steps of collecting implicit feedback data after the personalized service push, generating behavioral intent evolution parameters based on the implicit feedback data, and performing long-term intervention and evolution on the generation process of the scene behavioral intent map based on the behavioral intent evolution parameters are as follows: After completing the personalized push, monitor the user's interaction with the interaction pattern and mark the interaction as preliminary feedback information; The vehicle navigation trajectory is monitored, the final parking point location is obtained, and the final parking point location is marked as the behavior closed-loop verification result; By combining the initial feedback information and the behavioral closed-loop verification results, implicit feedback data is generated to characterize the user's true level of acceptance. Based on the implicit feedback data, the deduction deviation in the intention deduction process is obtained, and behavioral intention evolution parameters are generated according to the deduction deviation. The generation process of the scene behavioral intention map is adjusted according to the behavioral intention evolution parameters to intervene in the evolution in a long-term manner.
[0013] In summary, this application includes at least one of the following beneficial technical effects: By acquiring users' historical service data, extracting timestamp features and data source representations, and mapping them to a pre-defined high-dimensional spatiotemporal grid, standardized spatiotemporal behavior vectors are obtained. Then, time decay coefficients and data source reputation scores are extracted, and cross-domain correlation and feature completion are performed on the spatiotemporal behavior vectors to obtain completed behavior vectors. Next, a dynamic directed graph is constructed based on the completed behavior vectors, mining the causal logic chains within it to generate a scene behavior intent graph. Conflict self-resolution processing is performed on the graph based on the data source reputation score to eliminate contradictory logic from multiple data sources, thereby generating a highly accurate real-time dynamic user profile. Subsequently, the system collects real-time environmental conditions, vehicle status, and cognitive load around the user. Only when all three meet preset thresholds is a push notification triggered. After the notification is triggered, multi-objective optimization ranking is performed on the user's surrounding multi-source services to obtain the optimal solution for the target service. The system then adaptively determines the least disruptive interaction pattern based on the user's cognitive load to execute personalized service pushes. Finally, the system continuously collects implicit feedback data after push notifications to generate behavior intent evolution parameters, providing long-term intervention and evolution for the generation process of the scene behavior intent graph. It effectively breaks down information silos between third-party platforms, solves the problem of integrating heterogeneous and conflicting multi-source data, and improves the accuracy, interactive security, and intelligent foresight of personalized service pushes to users while ensuring user driving safety and avoiding push fatigue. Attached Figure Description
[0014] Figure 1 This is a flowchart of the method for integrating augmented reality-based digital human navigation and surrounding services provided in the embodiments of this application. Detailed Implementation
[0015] The following combination Figure 1 This application will be described in further detail, but the embodiments of the present invention are not limited thereto.
[0016] This application discloses a method for integrating augmented reality-based digital human navigation with surrounding services.
[0017] In this embodiment, the method for integrating augmented reality-based digital human navigation with surrounding services includes: S100: Obtain the user's historical service data and map the historical service data to a preset spatiotemporal high-dimensional grid to obtain a standardized spatiotemporal behavior vector; S200: Extract the time decay coefficient and data source reputation score from historical service data, perform cross-domain correlation and feature completion on the spatiotemporal behavior vector, and obtain the completed behavior vector; S300: Construct a dynamic directed graph based on the completed behavior vector, mine the data causal logic chain in the dynamic directed graph, and generate a scene behavior intent graph; S400: Based on the reputation score of the data source, the scenario behavior intent graph is processed by conflict resolution to obtain the target behavior intent graph, and a real-time dynamic profile of the user is generated based on the target behavior intent graph. S500: Based on real-time dynamic profiles, it collects the environmental status, vehicle status, and cognitive load around the user in real time. When the environmental status, vehicle status, and cognitive load all meet the preset thresholds, it triggers a push wake-up command. S600: Based on the push wake-up command, it optimizes and sorts the multiple source services around the user to obtain the optimal solution of the target service. It combines the optimal solution of the target service and the cognitive load to determine the interaction pattern and pushes personalized services according to the interaction pattern. S700: Collects implicit feedback data after personalized service push, generates behavioral intent evolution parameters based on implicit feedback data, and performs long-term intervention and evolution on the generation process of scene behavioral intent map based on behavioral intent evolution parameters.
[0018] The steps for acquiring users' historical service data and mapping it to a predefined spatiotemporal high-dimensional grid to obtain standardized spatiotemporal behavior vectors are as follows: Acquire users' historical transaction records, behavior trajectories, and underlying perception logs from in-vehicle devices on multiple different third-party service platforms, and aggregate them to generate historical service data; Based on historical service data, extract the data occurrence time information and the platform attribution label from the historical service data, mark the data occurrence time information as a timestamp feature, and mark the platform attribution label as a data source; Construct a virtual three-dimensional coordinate system that includes a time dimension axis, a spatial dimension axis, and a service type dimension axis, and divide the virtual three-dimensional coordinate system into multiple uniform spatiotemporal high-dimensional grids according to a preset granularity; Historical service data is projected into a spatiotemporal high-dimensional grid according to timestamp characteristics and data sources; Extract all service feature information aggregated in each spatiotemporal high-dimensional grid, and obtain a standardized spatiotemporal behavior vector based on the service feature information.
[0019] In practice, let's take a user as an example. The user has usage records on a group-buying app, a navigation app, and a car-related app. The system retrieves the user's food delivery and group-buying orders from the group-buying app over the past three months, their navigation history from the navigation app, and their vehicle charging logs from the car-related app. This data is aggregated to form historical service data. The system then identifies the time of each record, such as 12:00 PM on March 15, 2025, and the platform from which the record originated, such as the group-buying app. The system then establishes a virtual three-dimensional coordinate system, with time as one axis, space as another, and service type as a third axis. This coordinate system is divided into many small grids, each representing a time period, a location range, and a service type. For example, time is divided into hours, space into 500-meter grids, and service types are categorized into four types: refueling, charging, dining, and shopping. The system then places the user's historical service data into the corresponding small grids according to time and platform. For example, if a user navigates to Chaoyang Joy City on a navigation app at 12:00 noon on March 15, 2025, this data would be placed into the time cell "2025-03-15 12:00~13:00", the spatial cell "500 meters from Chaoyang Joy City", and the service type cell "navigation". Finally, the system extracts the characteristics of all the data in each cell and transforms them into a standardized numerical vector. This vector is the spatiotemporal behavior vector.
[0020] The steps for extracting the time decay coefficient and data source reputation score from historical service data, performing cross-domain correlation and feature completion on the spatiotemporal behavior vector, and obtaining the completed behavior vector are as follows: Extract the time difference between the timestamp feature and the current physical time, and substitute the time difference into the preset exponential decay model to obtain the time decay coefficient; The data source is matched and compared with the preset platform reputation mapping rating table to obtain the original reputation weight corresponding to the data source, and the original reputation weight is used as the data source reputation score. Cross-comparison of spatiotemporal behavior vectors from different data source platforms yields cross-domain correlation features between different spatiotemporal behavior vectors; The time-related features are weighted based on the time decay coefficient, and the cross-domain association features are weighted based on the authenticity based on the data source reputation score, so as to obtain the calibrated association features. Using the associated features as compensation data, the spatiotemporal behavior vector is completed to obtain the completed behavior vector.
[0021] In practice, let's take a user as an example. The system obtains their spatiotemporal behavior vector. Now, it needs to calculate the time decay coefficient and the data source reputation score. The system first looks at the timestamp of each data point. For example, the user has a food delivery record on a group-buying app from March 1, 2025. The current time is March 20, 2025, a time difference of 19 days. The system substitutes this time difference into the exponential decay model and calculates the time decay coefficient to be 0.85. This model is pre-built during the system initialization phase, and its core expression is... Where t represents the time difference (in days), and λ is a preset decay factor. Based on the differences in the contribution of historical data across different time intervals to the prediction of current behavioral intent, the system sets λ to 0.008 (this value is an empirical value pre-calibrated based on the average rate of decline in the importance of historical data over time in service recommendation scenarios; a larger λ indicates faster data decay, and a smaller λ indicates slower data decay; the system can adaptively adjust this value according to actual application scenarios). Substituting the time difference of 19 days into this formula, we get... The calculated result is approximately 0.859, which the system rounds to two decimal places and records as a time decay coefficient of 0.85. The exponential decay model calculates the coefficient according to a preset formula based on the magnitude of the time difference; the older the data, the smaller this coefficient. The system then checks the platform's reputation table: a group-buying app has a reputation score of 0.9, Gaode Maps has 0.8, and a vehicle app has 0.95. These are the data source reputation scores. The pre-set platform reputation mapping rating table is not generated temporarily, but is a static mapping relationship table established in advance based on a comprehensive evaluation of factors such as the historical data quality, data update frequency, user complaint rate, and industry-recognized authority of various mainstream third-party service platforms (such as a group-buying software, Gaode Map, and a car software). This table is stored in the system backend, and each common platform is pre-assigned a fixed initial reputation weight benchmark value. During real-time processing, the system directly extracts the marked "data source" (i.e., the platform tag) from the historical service data, uses this tag as an index to perform a matching query in the mapping rating table, and quickly reads the pre-stored weight (i.e., 0.9, 0.8, or 0.95) corresponding to the platform, using it as the "data source reputation score". After obtaining this score, the system uses it in the subsequent authenticity weighting processing step—that is, according to the score, it differentiates the weighting of cross-domain association features from different platforms (e.g., using "data source reputation score"). The system uses a score of 0.9 to increase the weight of this association feature, ensuring that data sources with higher credibility have a greater positive correction weight when generating the "calibrated association feature" and subsequent behavior vector completion. This ensures that the completed data is more aligned with the logic of highly credible sources. Next, the system compares spatiotemporal behavior vectors from different platforms. For example, if a user frequently orders takeout at lunchtime on a group-buying app and also frequently navigates to restaurants at lunchtime on a navigation app, these two behaviors are related, and the system identifies this cross-domain association feature. Then, the system uses a time decay coefficient of 0.85 to reduce the weight of this association feature because older data is less important. It then uses a data source credibility score of 0.9 to increase the weight of this association feature because the data from the group-buying app is reliable. This calibration yields an accurate association feature. Finally, the system adds this association feature as supplementary information to the original spatiotemporal behavior vector, filling in the missing parts to obtain the completed behavior vector.
[0022] The steps for constructing a dynamic directed graph based on completed behavior vectors, mining the causal logic chain of data in the dynamic directed graph, and generating a scene behavior intent graph are as follows: Extract the geographic location transition trajectory and consumption behavior sequence from the completion behavior vector, and use the completion behavior vector as the basic node of the graph structure, and the geographic location transition trajectory and consumption behavior sequence as directed edges between the nodes; Based on the interconnections between basic nodes and directed edges, an initial structure is constructed that can represent the evolution of user behavior over a continuous time period, resulting in a dynamic directed graph. Identify connected subgraphs in a dynamic directed graph, obtain the occurrence frequency of the connected subgraphs, and determine whether the occurrence frequency exceeds a preset frequency threshold. If the frequency of occurrence exceeds the frequency threshold, then the pseudo-correlated edges in the connected subgraph are removed, and the target directed edges with sequential causal relationships are extracted. Identify the first and last basic nodes of the target directed edge, and encapsulate the target directed edge and the first and last basic nodes to obtain the data causal logic chain. By acquiring different scenarios and dimensions of users' lives, the data causal logic chain is classified and reorganized according to different scenario dimensions, and then integrated to generate a scenario behavior intent map of users with potential service needs in different scenarios.
[0023] In practice, let's take a user as an example. The system already has completed behavior vectors. From these vectors, the system identifies the user's geographic location change trajectory and consumption behavior sequence. For example, the user starts from home, goes to the office, then has lunch at a shopping mall, returns to the office in the afternoon, and goes to the gym in the evening. These location changes are the geographic location transition trajectory. The consumption behavior sequence is: buy coffee in the morning, eat lunch at a restaurant, and buy a gym class in the evening. The system treats these completed behavior vectors as points in a graph, and geographic location changes and consumption sequences as directed arrows between these points. For example, the point "home" points to the point "office," and the arrows represent the direction of movement. Connecting all the points and arrows yields a dynamic directed graph. The system identifies recurring connection patterns in this graph, such as the user going to the gym and then to the convenience store every Friday evening; this pattern occurs 10 times. The system checks if this number exceeds a preset threshold, for example, 5 times. If it exceeds 10 times, the system considers this pattern a true causal chain. The system removes accidental erroneous connections, retaining only the arrow with a causal relationship and its endpoints, then packages them into a data causal logic chain. The system also recognizes different scenarios in a user's life, such as weekdays, weekends, and business trips. Based on these scenarios, similar causal chains are grouped together; for example, all weekday evening behavior chains are categorized together. Ultimately, a scenario-based behavioral intent graph is generated, containing the services a user might need in different scenarios.
[0024] The steps involved in performing conflict resolution on the scenario behavior intent graph based on the data source reputation score to obtain the target behavior intent graph, and then generating a real-time dynamic user profile based on the target behavior intent graph, are as follows: A global traversal scan of the scene behavior intent graph is performed to identify and extract two completely opposite behavioral logic branches that have contradictory derivations or point to the same spatiotemporal node, and these behavioral logic branches are marked as conflict nodes. Extract the source data reputation score corresponding to the source data of the conflict node, and evaluate the credibility probability of the behavioral logic branch based on the source data reputation score to obtain the confidence level of the behavioral logic branch. Based on the confidence level, identify and discard pseudo-logic branches with low confidence, retain true logic branches with high confidence, and obtain the target behavior intent map based on the true logic branches; Extract user preference features, consumption level features, and price sensitivity features from the target behavior intent map, combine multi-dimensional features, and generate a real-time dynamic profile of the user.
[0025] In practice, taking a specific user as an example, after obtaining the scenario behavior intent graph, the system discovered some contradictions. For instance, one chain in the graph shows that the user prefers spicy hot pot for lunch, while another chain shows that the user prefers light meals for lunch. These two conclusions are contradictory and point to the same time point (weekday lunchtime). The system marks these two branches as conflict nodes. Then, the system identifies which platform each piece of data comes from. The spicy hot pot data comes from a group-buying app with a credibility score of 0.9; the light meal data comes from another group-buying app with a credibility score of 0.7. The system evaluates the credibility of each branch based on the credibility score: the data from the first group-buying app is more credible, so the confidence level for spicy hot pot is 90%, and for light meals it is 70%. The system decides to retain the more credible spicy hot pot branch and remove the less credible light meal branch. This resolves the conflict, resulting in the target behavior intent graph. The system extracts the user's main characteristics from this profile: they like spicy food, spend an average of 30 yuan per meal, and are price-sensitive (hesitating if the price exceeds 40 yuan). The system combines these characteristics to generate a real-time dynamic user profile. This profile updates over time; for example, if the user is working overtime, the profile might indicate they want a faster meal.
[0026] The system collects real-time data on the user's surrounding environment, vehicle status, and cognitive load. When all three conditions meet preset thresholds, a wake-up command is triggered. Specifically: Real-time data collection of external weather conditions at the user's current location and traffic congestion index at the vehicle's current location; combined with external weather conditions and traffic congestion index, an environmental status is generated. The system reads the controller bus data, current driving range, and operational health status of key components inside the vehicle in real time, and generates the vehicle status by combining the controller bus data, current driving range, and operational health status. The system acquires data on the vehicle's current speed, steering wheel angle change rate, and the user's eye-tracking to assess the user's mental workload and attention distraction level when processing additional information. The system then combines the mental workload and attention distraction level to obtain the user's cognitive load. By combining the environmental and vehicle status, the user's current preference reference value is obtained, and it is determined whether the current preference reference value exceeds the preference tolerance pre-stored in the real-time dynamic profile. If it is determined that the current preference reference value exceeds the preference tolerance, then it is determined whether the cognitive load is lower than the preset safety interference threshold. If the cognitive load is determined to be below the safe interference threshold, a push wake-up command is generated and sent out.
[0027] In practice, taking a user as an example, after the system obtains a real-time dynamic profile, it begins to collect information about the surrounding environment in real time. The system uses a weather API to determine that it is currently raining outside, and the road condition shows a congestion index of 8 (out of 10). Combining these two data points, the system determines the environmental state as "rainy day + severe congestion." The system reads internal vehicle data: the car battery has 30% charge remaining, the brake pad wear is moderate, and the engine temperature is normal. Combining these data, the system determines the vehicle status as "low battery + normal braking." The system also obtains the vehicle's current speed of 20 km / h, the steering wheel angle change rate of 5 degrees per second, and the camera's observation that the user's gaze is on the road rather than on their phone. The system calculates the user's mental workload as 60 points (out of 100), the attention distraction score as 20 points (low distraction), and the vehicle speed of 20 km / h falling within the [0, 30) range. The baseline mental workload is set at 40 points. The system presets a base mental workload of 40 points (out of 100) for a vehicle speed range of [0, 30 km / h). A turning angle change rate of 5 degrees per second is incremented by (5 / 10) × 50 = 25 points, which is then reduced by 0.8 at low speeds to 20 points. The reduction factor is preset to 0.8 (the extra mental resources required for small steering wheel movements at low speeds are lower than at medium and high speeds). The base value plus the increment equals a mental workload of 60 points (out of 100). Current camera data indicates a user's gaze deviation angle of 3 degrees, lasting less than 0.5 seconds. The base dispersion value is 10 points, and the deviation angle penalty is 3 × 5 = 15 points. The system calculates the deviation angle between the user's gaze focus and the preset safe driving field of vision center area in real time and records the duration of each deviation. The system's preset distraction scoring function is: Distraction score = Base distraction score of 10 points + Offset angle penalty (5 points for every 1 degree increase in offset angle) + Persistent deviation penalty (10 points for every 1 second of persistent deviation). After a time-based reduction of 0.67, the score is 10 points. However, the system further reduces the penalty based on the duration of the deviation (less than 0.5 seconds after timing), with a preset reduction coefficient of 0.67. The actual offset angle penalty is rounded to approximately 15 × 0.67 ≈ 10 points with no persistent deviation penalty. The total distraction score is 20 points (out of 100). Combining these two factors, the cognitive load is rated as "moderate to low". After obtaining a mental load score of 60 points (moderate level) and a distraction score of 20 points (low level), the system uses a weighted fusion rule to calculate the overall cognitive load. In the driving safety scenario, the system presets that mental load contributes 70% and distraction contributes 30% (because the complexity of driving operations themselves has a greater impact on receiving additional information than on the impact of brief visual attention shifts). Therefore, the calculated cognitive load is 48 points. Based on the preset grading thresholds (0~40 points is low load, 40~70 points is medium-low load, 70~85 points is medium-high load, and 85~100 points is high load), the system classifies 48 points as medium-low load.The system, combining environmental and vehicle status, determines that the user is likely to need charging (due to low battery, traffic congestion, or rain). This preference reference value exceeds the preference tolerance pre-stored in the real-time dynamic profile (e.g., the user generally wants to charge when the battery is below 40%). The system then checks if the cognitive load is below a safe interference threshold (e.g., 70 points). The current cognitive load of 48 points is below 70 points, indicating that the user is currently available to receive push notifications. Therefore, the system generates a push notification wake-up command and sends it to the backend, preparing to recommend nearby charging stations to the user.
[0028] The steps involve optimizing and ranking the multiple source services around the user to obtain the optimal solution for the target service, combining the optimal solution of the target service with cognitive load to determine the interaction pattern, and then pushing personalized services based on the interaction pattern. Specifically: Based on the push wake-up command, the current user location is obtained, and a set of potential third-party service providers within a preset range that match the real-time dynamic profile are retrieved based on the current user location. Obtain the service price, route distance, and user history reviews for each potential third-party service provider in the potential third-party service provider set. Combine the service price, route distance, and user history reviews to generate a comprehensive rating for each potential third-party service provider. Select the potential third-party service provider with the highest comprehensive score as the target service provider, and take the service provided by the target service provider as the optimal solution for the target service; Based on cognitive load, the interaction mode with the least interference to driving operation is obtained by matching among the preset pure voice broadcast mode, minimal floating window mode and rich text and image pop-up mode. The data content of the optimal solution for the target service is loaded into the interactive display template and distributed to the user through the in-vehicle multimedia system.
[0029] In practice, let's take a user as an example. After receiving the push notification wake-up command, the system first obtains the user's current location, for example, near the Third Ring Road in Beijing. Based on the user's real-time dynamic profile (preferring affordable and well-rated charging stations), the system searches for all charging service providers within a 3-kilometer radius, finding three: Charging Brand 1, Charging Brand 2, and Charging Brand 3. The system obtains the prices from these three providers: Charging Brand 1 at 1.2 yuan per kWh, Charging Brand 2 at 1.0 yuan per kWh, and Charging Brand 3 at 0.8 yuan per kWh. The distances are: Charging Brand 1 at 500 meters, Charging Brand 2 at 1 kWh, and Charging Brand 3 at 1.5 kilometers. The user's historical ratings are: Charging Brand 1 at 4.5 stars, Charging Brand 2 at 4.0 stars, and Charging Brand 3 at 4.8 stars. The system calculates a weighted overall score based on these three factors: Charging Brand 3 receives the highest score (low price, high rating, and acceptable distance despite its distance). The system then assigns a comprehensive score to the three charging brands. The preset weights for price, distance, and rating are 50%, 20%, and 30%, respectively (price is prioritized for price-sensitive users). Price scores are calculated as "lowest price / unit price of the brand". Charging brand 3, with a unit price of 0.8 yuan, has the lowest score of 1.0; charging brand 2, with a unit price of 1.0, scores 0.8; and charging brand 1, with a unit price of 1.2 yuan, scores 0.67. Distance scores are calculated as "shortest distance / distance of the brand". Charging brand 1 scores 1.0, charging brand 2 scores 0.5, and charging brand 3 scores 0.33. Rating scores are directly normalized to a maximum of 5 points: charging brand 3 scores 0.96, charging brand 2 scores 0.8, and charging brand 1 scores 0.9. Overall rating = price × 50% + distance × 20% + rating × 30%. Calculation results: Charging brand 1: 0.67×50%+1.0×20%+0.9×30%=0.335+0.2+0.27=0.805; Charging brand 2: 0.8×50%+0.5×20%+0.8×30%=0.4+0.1+0.24=0.74; Charging brand 3: 1.0×50%+0.33×20%+0.96×30%=0.5+0.066+0.288=0.854. Charging brand 3 has the highest score, therefore it is selected as the optimal solution for the target service. The system selects charging brand 3 as the target service provider, and its service is the optimal solution for the target service. The system then considers the user's cognitive load, which is 48 points, which is considered low to medium. The system selects from three interaction modes: pure voice broadcast (least interference), minimalist floating window (medium), and rich graphic pop-up window (most interference). Because the user's cognitive load is low, but for safety reasons, the system chooses a minimalist floating window format. The system fills in the name, distance, price, and other information of the charging station of Brand 3 into the template of the minimalist floating window, and then displays it on the in-vehicle multimedia screen, while simultaneously announcing in a simple voice, "There is a charging station of Brand 3 1.5 kilometers ahead, with cheap prices."
[0030] The process of collecting implicit feedback data after personalized service pushes, generating behavioral intent evolution parameters based on this data, and then using these parameters to intervene and evolve the generation process of the scene behavioral intent graph over the long term is as follows: After completing the personalized push notification, monitor the user's interaction with the interactive pattern and mark the interaction as preliminary feedback information; Monitor the vehicle's navigation trajectory, obtain the final parking point location, and mark the final parking point location as the behavior loop verification result; By combining preliminary feedback information and behavioral closed-loop verification results, implicit feedback data is generated to characterize the user's true level of acceptance. Based on implicit feedback data, the derivation deviation in the intention derivation process is obtained, and behavioral intention evolution parameters are generated according to the derivation deviation. The generation process of the scene behavioral intention map is adjusted according to the behavioral intention evolution parameters to intervene in the evolution in a long-term manner.
[0031] In practice, taking a user as an example, after the system pushes information about charging station #3, it observes the user's reaction. After seeing the floating window, the user taps the "Navigation" button on the screen, and the system records this tap as initial feedback (indicating acceptance). The system then continues to monitor the vehicle's navigation trajectory, discovering that the user ultimately parks at charging station #3 and begins charging. This final parking location is the behavioral loop verification result. The system combines the tap and the actual parking information to generate implicit feedback data, indicating that the user truly accepted and used the recommendation. The system analyzes the entire deduction process: the system previously predicted the user wanted to charge and recommended charging station #3, and the user actually went there. This deduction bias is very small, only 5%. Based on this bias, the system generates behavioral intent evolution parameters, such as increasing the weight of the rule "recommend low-priced charging stations when battery is low." The system then applies these parameters to the generation of the scenario behavioral intent graph. In similar situations in the future, the system will prioritize recommending low-priced charging stations. In this way, the system continuously improves its predictive capabilities through the user's actual behavior, achieving long-term intervention and evolution.
[0032] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.
Claims
1. A method for integrating digital human navigation with surrounding services based on augmented reality, characterized in that, include: Obtain the user's historical service data and map the historical service data to a preset spatiotemporal high-dimensional grid to obtain a standardized spatiotemporal behavior vector; Extract the time decay coefficient and data source reputation score from the historical service data, perform cross-domain association and feature completion on the spatiotemporal behavior vector to obtain the completed behavior vector; Based on the completed behavior vector, a dynamic directed graph is constructed, and the data causal logic chain in the dynamic directed graph is mined to generate a scene behavior intent graph. The scenario behavior intent graph is subjected to conflict self-resolution processing based on the reputation score of the data source to obtain the target behavior intent graph, and a real-time dynamic profile of the user is generated based on the target behavior intent graph. Based on the real-time dynamic profile, the environmental status, vehicle status, and cognitive load around the user are collected in real time. When the environmental status, vehicle status, and cognitive load all meet the preset thresholds, a push wake-up command is triggered. Based on the push wake-up command, the optimal solution for the target service is obtained by optimizing and sorting the multi-source services around the user. The interaction mode is determined by combining the optimal solution for the target service and the cognitive load, and personalized service is pushed according to the interaction mode. The implicit feedback data after the personalized service is pushed is collected, and behavioral intent evolution parameters are generated based on the implicit feedback data. The generation process of the scene behavioral intent map is then subjected to long-term intervention and evolution based on the behavioral intent evolution parameters. 2.The augmented reality-based digital human navigation and surrounding service fusion method of claim 1, wherein, The steps of acquiring users' historical service data and mapping the historical service data to a preset spatiotemporal high-dimensional grid to obtain standardized spatiotemporal behavior vectors are as follows: Acquire users' historical transaction records, behavior trajectories, and underlying perception logs from in-vehicle devices on multiple different third-party service platforms, and aggregate them to generate historical service data; Based on the historical service data, extract the data occurrence time information and the platform attribution tag from the historical service data, mark the data occurrence time information as a timestamp feature, and mark the platform attribution tag as a data source; Construct a virtual three-dimensional coordinate system that includes a time dimension axis, a spatial dimension axis, and a service type dimension axis, and divide the virtual three-dimensional coordinate system into multiple uniform spatiotemporal high-dimensional grids according to a preset granularity; The historical service data is projected into the spatiotemporal high-dimensional grid according to the timestamp feature and the data source, respectively. Extract all service feature information aggregated in each of the spatiotemporal high-dimensional grids, and obtain a standardized spatiotemporal behavior vector based on the service feature information. 3.The augmented reality-based digital human navigation and surrounding service fusion method of claim 2, wherein, The steps for extracting the time decay coefficient and data source reputation score from the historical service data, performing cross-domain correlation and feature completion on the spatiotemporal behavior vector to obtain the completed behavior vector are as follows: Extract the time difference between the timestamp feature and the current physical time, and substitute the time difference into a preset exponential decay model to obtain the time decay coefficient; The data source is matched and compared with a preset platform reputation mapping rating table to obtain the original reputation weight corresponding to the data source, and the original reputation weight is used as the data source reputation score. Cross-comparison of the spatiotemporal behavior vectors from different data source platforms yields cross-domain association features between the different spatiotemporal behavior vectors. Based on the time decay coefficient, the cross-domain association features are subjected to time-sensitivity downweighting, and based on the data source reputation score, the cross-domain association features are subjected to authenticity upweighting, resulting in calibrated association features. The associated features are used as compensation data to complete the spatiotemporal behavior vector, resulting in a completed behavior vector.
4. The augmented reality-based digital human navigation and surrounding service fusion method of claim 3, wherein, The steps for constructing a dynamic directed graph based on the completed behavior vector, mining the causal logic chain of data in the dynamic directed graph, and generating a scene behavior intent graph are as follows: Extract the geographic location transition trajectory and consumption behavior sequence from the completed behavior vector, and use the completed behavior vector as the basic node of the graph structure, and the geographic location transition trajectory and the consumption behavior sequence as directed edges between the nodes; Based on the interconnections between the basic nodes and the directed edges, an initial structure that can characterize the evolution of user behavior over a continuous time period is constructed, resulting in a dynamic directed graph. Identify the connected subgraphs in the dynamic directed graph, obtain the occurrence frequency of the connected subgraphs, and determine whether the occurrence frequency exceeds a preset frequency threshold; If it is determined that the occurrence frequency exceeds the frequency threshold, then the pseudo-correlated edges in the connected subgraph are removed, and the target directed edges with sequential causal relationships are extracted. Identify the first and last basic nodes of the target directed edge, and encapsulate the target directed edge and the first and last basic nodes to obtain the data causal logic chain; By acquiring different scenario dimensions of users' lives, classifying and reorganizing the causal logic chain of the data according to the different scenario dimensions, and merging them to generate a scenario behavior intent map of users with potential service needs in different scenarios.
5. The augmented reality-based digital human navigation and surrounding service fusion method of claim 4, wherein, The steps of performing conflict self-resolution processing on the scenario behavior intent graph based on the reputation score of the data source to obtain the target behavior intent graph, and generating a real-time dynamic profile of the user based on the target behavior intent graph, are as follows: A global traversal scan of the scene behavior intent graph is performed to identify and extract two diametrically opposed behavioral logic branches that have contradictory derivations or point to the same spatiotemporal node, and these behavioral logic branches are marked as conflict nodes. Extract the source data reputation score corresponding to the source data of the conflict node, and evaluate the credibility probability of the behavioral logic branch based on the source data reputation score to obtain the confidence level of the behavioral logic branch. Based on the confidence level, the pseudo-logic branches with lower confidence levels are identified and discarded, while the true logic branches with higher confidence levels are retained. The target behavior intent map is obtained based on the true logic branches. Extract user preference features, consumption level features, and price sensitivity features from the target behavior intent map, combine them into a multi-dimensional feature set, and generate a real-time dynamic profile of the user.
6. The augmented reality-based digital human navigation and surrounding service fusion method of claim 5, wherein, The system collects real-time data on the user's surrounding environment, vehicle status, and cognitive load. When all three conditions (environment, vehicle, and cognitive load) meet preset thresholds, a wake-up command is triggered. Specifically: The system collects real-time data on the external weather conditions at the user's current location and the traffic congestion index at the vehicle's current location, and combines the external weather conditions and the traffic congestion index to generate an environmental status. The system reads the controller bus data, current driving range, and operational health status of key components inside the vehicle in real time, and generates the vehicle status by combining the controller bus data, the current driving range, and the operational health status. The system acquires the vehicle's current speed, steering wheel angle change rate, and user eye tracking data to assess the user's mental workload and attention distraction level while processing additional information. The system then combines the mental workload and attention distraction level to obtain the user's cognitive load. Based on the environmental state and the vehicle state, the user's current preference reference value is obtained, and it is determined whether the current preference reference value exceeds the preference tolerance pre-stored in the real-time dynamic profile. If it is determined that the current preference reference value exceeds the preference tolerance, then it is determined whether the cognitive load is lower than the preset safety interference threshold; If it is determined that the cognitive load is lower than the safety interference threshold, a push wake-up command is generated and sent out.
7. The augmented reality-based digital human navigation and surrounding service fusion method of claim 6, wherein, The steps of optimizing and ranking multiple source services around the user to obtain the optimal solution for the target service, combining the optimal solution for the target service with the cognitive load to determine the interaction pattern, and then pushing personalized services based on the interaction pattern are as follows: Based on the push wake-up command, the current user location is obtained, and based on the current user location, a set of potential third-party service providers within a preset range that match the real-time dynamic profile is retrieved. Obtain the service price, route distance, and user history ratings for each potential third-party service provider in the set of potential third-party service providers. Combine the service price, route distance, and user history ratings to generate a comprehensive rating for each potential third-party service provider. The potential third-party service provider with the highest comprehensive score is selected as the target service provider, and the service provided by the target service provider is taken as the optimal solution for the target service. Based on the cognitive load, the interaction mode with the least interference to driving operation is obtained by matching among the preset pure voice broadcast mode, minimal floating window mode and rich text and image pop-up mode. The data content of the optimal solution of the target service is loaded into the display template of the interactive form and distributed to the user through the in-vehicle multimedia system.
8. The augmented reality-based digital human navigation and surrounding service fusion method of claim 7, wherein, The steps of collecting implicit feedback data after the personalized service push, generating behavioral intent evolution parameters based on the implicit feedback data, and performing long-term intervention and evolution on the generation process of the scene behavioral intent map based on the behavioral intent evolution parameters are as follows: After completing the personalized push, monitor the user's interaction with the interaction pattern and mark the interaction as preliminary feedback information; The vehicle navigation trajectory is monitored, the final parking point location is obtained, and the final parking point location is marked as the behavior closed-loop verification result; By combining the initial feedback information and the behavioral closed-loop verification results, implicit feedback data is generated to characterize the user's true level of acceptance. Based on the implicit feedback data, the deduction deviation in the intention deduction process is obtained, and behavioral intention evolution parameters are generated according to the deduction deviation. The generation process of the scene behavioral intention map is adjusted according to the behavioral intention evolution parameters to intervene in the evolution in a long-term manner.
