A dynamic planning method for a travel route based on multi-source positioning data fusion

Abstract

This invention relates to the field of smart tourism technology, specifically to a dynamic planning method for tourist routes based on multi-source positioning data fusion. The method includes: S100: acquiring real-time environmental physical data and multi-source positioning data collected by the terminal's built-in sensors, and a pre-constructed environmental feature map, wherein the environmental feature map contains topological nodes associated with physical reference features; S200: comparing the real-time environmental physical data with the physical reference features at the corresponding topological nodes. This invention solves the problem of positioning divergence for tourists in complex terrain transition areas through an adaptive factor graph optimization model with strong coupling of physical environmental features. Simultaneously, it constructs an event-driven silent estimation mechanism based on spatial vector deviation and location confidence decay, reducing the terminal's communication state from global polling to on-demand triggering of simplified state features, thereby exponentially reducing the operating power consumption of smart devices and the concurrent load of scenic area base stations.

Description

Technical Field

[0001] This invention relates to the field of smart tourism technology, specifically to a dynamic planning method for tourism routes based on multi-source positioning data fusion. Background Technology

[0002] With the development of smart tourism, location-based scenic area navigation systems have become an important tool for enhancing the tourist experience. Existing navigation systems typically rely on Global Navigation Satellite Systems (GNSS), wireless communication base stations, and built-in maps on smartphones for route guidance. However, faced with the extremely complex physical spaces and ever-changing network environments of real-world tourism scenarios, existing technological solutions have revealed the following deep-seated technical shortcomings: First, in areas with complex physical terrain transitions, the lack of environmental awareness in the localization algorithm leads to severe drift.

[0003] Natural scenic areas typically encompass complex terrains such as canyons, dense forests, semi-enclosed caves, and indoor exhibition halls. When tourists traverse between outdoor and extreme indoor environments, GNSS signals experience severe multipath effects or sudden drops. Existing multi-source fusion positioning schemes (such as the traditional extended Kalman filter algorithm) mostly rely solely on the attenuation of the receiver's signal-to-noise ratio to linearly adjust the weights of the positioning sources. Because the system completely lacks the ability to perceive abrupt changes in the "physical environment" in which tourists are located, this mathematical adjustment mechanism, which purely depends on the strength of the communication signal, suffers from severe hysteresis, causing the terminal to easily experience trajectory divergence and random jumps of tens of meters in terrain transition areas.

[0004] Second, the high-frequency polling architecture leads to a sharp drop in terminal battery life and high-concurrency network congestion.

[0005] To correct the aforementioned complex positioning drift, existing scenic area navigation applications generally adopt a polling strategy of "high-frequency continuous online coordinate reporting" to request the server to continuously verify the route. This "heavy reliance on the cloud" mechanism faces fatal flaws during peak tourist seasons: on the one hand, the high-frequency wake-up of the radio frequency module can cause tourists' terminal batteries to run out within hours, rendering it impractical for long-term outdoor navigation; on the other hand, the massive concurrent coordinate reporting can cause severe congestion on the scenic area's local area network. When tourists actually "go the wrong way" or get lost at remote intersections, they often cannot receive timely alerts due to data packet loss in weak network environments.

[0006] Third, cloud-based route reconstruction is detached from real physical constraints and lacks an emergency blocking mechanism under weak network conditions.

[0007] In terms of dynamic route replanning, most existing cloud navigation engines only use "shortest two-dimensional geometric distance" as the sole optimization objective. Existing graph neural network algorithms, when aggregating routes in space, completely disregard the strong constraints of the real physical world. The system neither considers the non-linear physical exhaustion caused by mountainous terrain (such as accumulated elevation gain) nor incorporates the dynamic opening time windows of individual candidate attractions. This often results in the system planning a route that is "short in distance but extremely physically demanding," or even guiding tourists to closed attractions. Furthermore, in extreme emergencies such as when tourists are trapped in deep signal blind spots, the existing architecture rigidly attempts time-consuming route synchronization with the cloud, completely lacking a rapid response mechanism that relies on the local edge to directly block and generate distress routes, posing a significant safety hazard in preventing tourists from getting lost.

[0008] Therefore, there is an urgent need for a novel dynamic planning method for tourism routes based on multi-source positioning data fusion to solve the aforementioned technical bottlenecks. Summary of the Invention

[0009] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a dynamic planning method for tourist routes based on multi-source positioning data fusion, including the following steps: S100: acquiring real-time environmental physical data and multi-source positioning data collected by the terminal's built-in sensors, and a pre-constructed environmental feature map, wherein the environmental feature map contains topological nodes associated with physical reference features; S200: Compare the real-time environmental physical data with the physical reference features at the corresponding topological node, dynamically adjust the observation weights of different positioning data sources in the factor graph optimization model according to the degree of difference in the comparison, and output the real-time location coordinates and real-time positioning reliability of the terminal through the factor graph optimization model. S300. Monitor the real-time location reliability and the route deviation status of the terminal. When the real-time location reliability is lower than a preset safety threshold or the route deviation status meets the triggering condition, wake up the terminal's sleep communication module and send a status feature packet containing the real-time location coordinates to the edge server. S400. The edge server responds to the state feature packet, blocks the conventional global planning based on the state of the real-time location confidence, and forcibly generates a local anti-lost emergency route, or collaborates with the cloud server to generate a global alternative route based on the current scenic area congestion, and sends the generated dynamic route to the terminal.

[0010] Furthermore, the specific methods for acquiring the real-time environmental physical data and the multi-source positioning data include: calling the ambient light sensor, barometer, and magnetometer built into the terminal to collect the ambient light illuminance, air pressure fluctuation frequency, and magnetic field strength variance of the current location in real time as the real-time environmental physical data; and calling the wireless communication module and inertial measurement unit built into the terminal to collect the signal-to-noise ratio of the global navigation satellite system, the low-power Bluetooth received signal strength indication, the terminal acceleration, and angular velocity in real time as the multi-source positioning data.

[0011] Furthermore, the environmental feature map is a directed graph data structure containing multiple topological nodes and connecting edges, wherein: the topological nodes contain three-dimensional spatial coordinates and the physical reference features, the physical reference features are composed of pre-collected historical ambient light illuminance reference values, historical magnetic field variance reference values ​​and low-power Bluetooth fingerprint reference vectors; the connecting edges contain the road segment connectivity attributes, physical walking distance and historical average network signal coverage strength between two adjacent topological nodes.

[0012] Furthermore, the step of dynamically adjusting the observation weights of different location data sources in the factor graph optimization model based on the degree of difference in the comparison includes: calculating the environmental abrupt change index between the real-time environmental physical data and the physical reference features at the corresponding topological node; and dynamically updating the observation error covariance matrix of the multi-source location data in the factor graph optimization model based on the environmental abrupt change index using a set decay function, with the update formula being:

[0013] in, Indicates the first Location data source The observation error covariance matrix updated at each time step. This shows the initial observation error covariance matrix. Indicates the first The environmental sensitivity coefficient of the location data source The environmental mutation index represents the sudden drop in ambient light intensity exceeding a set threshold and the frequency of air pressure fluctuations meeting indoor characteristics. When the environmental mutation index represents the sudden drop in ambient light intensity exceeding a set threshold and the frequency of air pressure fluctuations meeting indoor characteristics, the system automatically increases the environmental sensitivity coefficient corresponding to the global navigation satellite system, causing the corresponding observation error covariance matrix to be exponentially amplified, thereby reducing its observation weight in the factor graph optimization model.

[0014] Furthermore, the factor graph optimization model includes state nodes and factor nodes connected to the state nodes, wherein: the state nodes include the physical position, velocity, and attitude of the terminal at various historical moments and the current moment; the factor nodes include kinematic prior factor nodes and multi-source observation factor nodes; the kinematic prior factor nodes are constructed based on the terminal acceleration and the angular velocity, and the multi-source observation factor nodes are constructed based on the signal-to-noise ratio of the Global Navigation Satellite System and the Bluetooth Low Energy received signal strength indication; the information matrix of the multi-source observation factor nodes is the inverse matrix of the observation error covariance matrix corresponding to the positioning data source.

[0015] Further, the step of calculating and outputting the real-time location coordinates and real-time positioning reliability of the terminal through the factor graph optimization model includes: The optimal state node sequence is solved by minimizing the sum of Mahalanobis distances of all the factor nodes, and the optimal state at the current moment is extracted as the real-time position coordinates. The marginal covariance matrix corresponding to the real-time position coordinates when solving the factor graph optimization model is extracted, and the real-time position confidence is calculated using the following confidence evaluation formula:

[0016] in, express The real-time location reliability at that moment. The set scale adjustment constant, Represents the edge covariance matrix The smaller the value of the trace, the higher the reliability of the real-time positioning.

[0017] Furthermore, monitoring the real-time location reliability and the route deviation status of the terminal includes: If the triggering condition is not met and the real-time positioning confidence is greater than or equal to the preset safety threshold, the communication module of the terminal is controlled to remain in a sleep state, and dead reckoning is performed using the local road network topology cached locally by the terminal and the pose increment output by the inertial measurement unit, and the real-time position coordinates are updated. The vertical projection distance from the real-time position coordinates to the preset planned route is calculated in real time, and the heading angle between the current motion speed vector of the terminal and the tangential vector of the preset planned route is calculated in advance. These are combined as the route deviation state.

[0018] Further, the step of waking up the terminal's sleep communication module when the real-time positioning confidence is lower than a preset safety threshold or when the route deviation state meets the triggering condition includes: When any of the following event-driven conditions are detected, the system generates a hardware interrupt signal to wake up the communication module: Condition 1: The time span during which the real-time location confidence is continuously lower than the preset safety threshold exceeds the set time tolerance window. ; Condition 2: The vertical projection distance is greater than the set spatial deviation tolerance. Furthermore, the cosine value of the heading angle is consistently less than zero; Sending a status feature packet containing the real-time location coordinates to the edge server includes: The real-time position coordinates, the event type encoding that triggered the wake-up, and the current motion velocity vector are serialized and encapsulated to generate a lightweight state feature package; The data payload of the state feature packet excludes the original sampling sequence of the real-time environmental physical data and the original observation message of the multi-source positioning data, and transmits the state feature packet uplink once only through the wireless radio frequency channel to the edge service that is closest to the terminal in physical distance.

[0019] Furthermore, the step of blocking conventional global planning and forcibly generating local anti-wandering emergency routes based on the state of the real-time location confidence includes: When the event type encoding in the status feature packet indicates that the real-time location confidence is lower than the preset security threshold, the edge server immediately disconnects the route synchronization link between the current terminal and the cloud server. In the micro-security topology map stored locally on the edge server, using the real-time location coordinates and the current movement speed vector in the state feature packet as the initial state, the shortest connecting path to the nearest open physical safety area or the nearest manual assistance base station is searched and prioritized as the local anti-lost emergency route to block cross-regional network delays.

[0020] Furthermore, the collaborative cloud server generates a global alternative route based on the current congestion level of the scenic area, including: When the state feature package indicates that the route deviation state meets the triggering condition and the real-time positioning confidence is at a normal level, the edge server determines whether the local micro-human flow density within a preset physical radius around the current deviation position is greater than the congestion threshold. If the congestion threshold is exceeded, the edge server directly generates an instant turning instruction to avoid the current congested intersection based on the local micro-road network and sends it to the terminal. After issuing the instant turn command, the edge server transmits the real-time location coordinates to the cloud server; the cloud server calls the spatiotemporal graph convolutional neural network model, inputs the real-time human flow heat map features, opening time windows and tourist physical exertion index of each candidate point of interest in the whole domain, and outputs the optimal global alternative route to avoid high-risk congestion areas. When the cloud server calls the spatiotemporal graph convolutional neural network model to output the optimal global alternative route, it performs the following spatial-temporal joint physical constraint evaluation: Calculate the estimated physical walking time from the real-time location coordinates to the candidate point of interest, compare the estimated physical walking time with the remaining open time window of the candidate point of interest, and filter out nodes that are unreachable by time. Extract the cumulative elevation difference of the terrain undulations from the real-time location coordinates to the connecting road segment where the candidate point of interest is located, and calculate the physical exertion penalty weight by combining it with the tourist physical exertion index. The physical exertion penalty weight and the real-time human flow thermal features are used as dynamic adjustment parameters for the edge weights in the spatiotemporal graph convolutional neural network model. The connected sequence with the minimum total algebraic value is selected through graph space aggregation operation to generate the optimal global alternative route.

[0021] Beneficial effects This invention solves the problem of location divergence for tourists in complex terrain transition areas by using an adaptive factor graph optimization model strongly coupled with physical environment features. Simultaneously, it constructs an event-driven silent estimation mechanism based on spatial vector deviation and location reliability decay, reducing the terminal's communication state from global polling to on-demand triggering based on simplified state features. This exponentially reduces the power consumption of smart devices and the concurrent load of scenic area base stations. Furthermore, relying on edge-side physical blocking response and cloud-based spatiotemporal graph convolutional neural network with injected elevation and physical exertion constraints, it can not only provide millisecond-level local escape guidance in extremely weak network and lost scenarios, but also plan global alternative routes that fully conform to tourists' physiological limits and the opening time of attractions during periods of crowd congestion. Attached Figure Description

[0022] Figure 1 This is a flowchart of a dynamic planning method for tourism routes based on multi-source positioning data fusion, according to the present invention. Detailed Implementation

[0023] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but includes other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] The present invention will now be described in further detail with reference to the accompanying drawings: Example: like Figure 1 As shown, a dynamic planning method for tourism routes based on multi-source positioning data fusion includes the following steps: S100: Acquire real-time environmental physical data and multi-source positioning data collected by the terminal's built-in sensors, and a pre-constructed environmental feature map, wherein the environmental feature map contains topological nodes associated with physical reference features; S200: Compare the real-time environmental physical data with the physical reference features at the corresponding topological node, dynamically adjust the observation weights of different positioning data sources in the factor graph optimization model according to the degree of difference in the comparison, and output the real-time location coordinates and real-time positioning reliability of the terminal through the factor graph optimization model. S300. Monitor the real-time location reliability and the route deviation status of the terminal. When the real-time location reliability is lower than a preset safety threshold or the route deviation status meets the triggering condition, wake up the terminal's sleep communication module and send a status feature packet containing the real-time location coordinates to the edge server. S400. The edge server responds to the state feature packet, blocks the conventional global planning based on the state of the real-time location confidence, and forcibly generates a local anti-lost emergency route, or collaborates with the cloud server to generate a global alternative route based on the current scenic area congestion, and sends the generated dynamic route to the terminal.

[0026] Furthermore, the specific implementation process of step S100 is as follows: In this embodiment, step S100 is not a simple data retrieval, but a multimodal data preparation process that encompasses cloud-based prior knowledge loading and real-time physical environment perception at the terminal. Traditional route planning methods often rely solely on single latitude and longitude coordinates, while this invention decouples and collects "physical environment features" and "location data" in parallel.

[0027] Regarding the acquisition of real-time data from the terminal: As tourists move around the scenic area with their smart devices, the system continuously calls the heterogeneous sensor network built into the device in the background via a low-power application programming interface (API). Specifically, the system strictly divides the acquired data into two categories: "real-time environmental physical data" and "multi-source positioning data," for mutual verification.

[0028] For real-time environmental physics data, the system uses an ambient light sensor to collect the ambient light intensity at the current location, a barometer to obtain air pressure readings, and a magnetometer to obtain triaxial magnetic field strength. To eliminate sensor noise interference and extract features with physical significance, the system performs feature extraction on the collected raw physical quantities within a sliding time window. For example, for magnetic field data, the system does not use absolute magnetic field values ​​but instead calculates and sets the time window... The variance of magnetic field strength within (e.g., 2 seconds) The calculation formula is as follows:

[0029] in, For time window The total number of sampling points within, The first The three-axis magnetic field components of each sampling point The variance characteristic represents the average magnetic field strength within a time window. This variance characteristic can extremely sensitively reflect whether tourists have approached indoor steel-framed corridors or underground artificial tunnels containing a large amount of metal structures, which is something that conventional navigation cannot detect.

[0030] Simultaneously, for multi-source positioning data, the system calls the wireless communication module and the inertial measurement unit to extract the average signal-to-noise ratio of all visible satellites in the global navigation satellite system and the received signal strength indication set of surrounding low-power Bluetooth beacons, in conjunction with the terminal's acceleration and angular velocity data.

[0031] Regarding pre-constructed environmental feature maps: When a terminal enters a scenic area or connects to its local area network, it pre-downloads or dynamically requests an "environmental feature map" stored on the edge server. This map is not a traditional two-dimensional planar map in its data structure, but rather a carefully designed directed graph with multi-dimensional physical attributes. .

[0032] Topological nodes in the directed graph Not only anchored three-dimensional spatial coordinates Each node is deeply bound to a "physical baseline feature vector". This vector is composed of historical ambient light intensity reference values, historical magnetic field variance reference values, and low-power Bluetooth fingerprint reference vectors pre-collected during the scenic area survey phase. Directed edges connecting adjacent nodes... In addition to including standard road connectivity attributes and physical walking distance, this embodiment associates them with "historical average network signal coverage strength." This gives the terminal the ability to predict network blind spots ahead. Experimental scenario and comparative analysis: To conclusively demonstrate the significant advancements of the aforementioned data acquisition method, this embodiment provides a comparative experiment of a typical tourism scenario.

[0033] Experimental scenario: Tourists are walking along an open canyon boardwalk and are about to enter a semi-enclosed karst cave with complex terrain.

[0034] Comparative Example (Existing Technical Solution): When entering a cave, traditional navigation terminals, lacking prior knowledge of the "environment," still primarily rely on GPS data. When GPS signals experience multipath effects and sudden drops at the cave entrance due to obstruction, traditional systems only exhibit "significant random drift of the positioning point," frequently triggering route replanning requests based on erroneous drift coordinates. This leads to a rapid depletion of the terminal's battery and provides tourists with incorrect directions.

[0035] In this embodiment of the invention, within the framework of step S100, when a tourist reaches the entrance of the cave, the "real-time environmental physical data" collected by the terminal will instantly undergo a step change: the ambient light intensity drops sharply, and the air pressure fluctuates at a specific frequency due to the cave effect. The system immediately compares this real-time physical data with the entrance node in the "environmental feature map". The system compares the data with preset "physical baseline characteristics". Even if the signal-to-noise ratio of the global navigation satellite system begins to decrease at this time, the system can accurately determine from a physical perspective that "the tourist is crossing from the outdoor environment to an extreme indoor environment" based on the difference in multi-data characteristics obtained by S100.

[0036] Furthermore, the specific implementation process of S200 is as follows: In order to overcome the technical bias that traditional multi-source fusion positioning is prone to divergence due to the complex terrain in tourist attractions (such as canyons, caves, dense forests and indoor exhibition halls), this embodiment elaborates in depth on the underlying computational logic of step S200 "adaptive calculation of positioning and reliability based on environmental physical features".

[0037] Traditional fusion positioning mostly employs Extended Kalman Filter (EKF), whose observation noise covariance matrix is ​​typically set to a fixed empirical value or linearly adjusted solely based on the signal-to-noise ratio of the receiver. When visitors suddenly move from an open outdoor area into an underground exhibition hall, traditional models often exhibit significant lag due to a lack of perception of drastic changes in the physical environment. This embodiment breaks through this framework by constructing an adaptive factor graph optimization model directly driven by physical sensors.

[0038] First, the system calculates the environmental mutation index based on physical data.

[0039] After the system obtains the real-time environmental physical data and the physical reference features of the corresponding topology node output in step S100, it does not directly input them into the positioning solver, but extracts the transient difference between the two in the time dimension.

[0040] Specifically, the system calculates the current time. The environmental abrupt change index between real-time environmental physical data and the physical baseline characteristics at the corresponding topology node The index is calculated by incorporating the gradient of ambient light intensity decrease and the frequency of air pressure fluctuations, reflecting the degree of abrupt changes in the spatial morphology of the terminal.

[0041] Secondly, the observation weights of multi-source localization are reshaped using the nonlinearity of the environmental mutation index.

[0042] In obtaining the environmental mutation index Subsequently, the system dynamically updates the observation error covariance matrix of the multi-source positioning data in the factor graph optimization model using a set decay function. The update process follows the nonlinear evolution formula below:

[0043] in, Indicates the first Location data source The observation error covariance matrix updated at each time step. This represents the initial reference observation error covariance matrix. Indicates the first The environmental sensitivity coefficient of a location data source. In factor graph optimization theory, the information matrix (i.e., observation weights) of the location data source during the fusion process is the inverse of its covariance matrix. .

[0044] The advantage of the above design is that: when the environmental mutation index... When the gradient representing ambient illuminance drops beyond a set threshold and air pressure fluctuations meet indoor physical characteristics, the system automatically increases the sensitivity coefficient corresponding to the Global Navigation Satellite System (GNSS). This causes the GNSS observation error covariance matrix to increase exponentially, and its corresponding information matrix (weights) instantaneously approaches zero. Simultaneously, the system maintains or reduces the sensitivity coefficients of Bluetooth Low Energy (BLE) and the Inertial Measurement Unit (IMU). Through this strong coupling mechanism between physics and mathematics, the system achieves a "soft handover" of the weights of different positioning data sources within an extremely short transition window.

[0045] Subsequently, the global optimal state and real-time location confidence are solved by factor graph optimization.

[0046] After reconstructing the observation weights from each data source, the system constructs a factor graph within a sliding window. This factor graph contains state nodes (representing the terminal's position, velocity, and attitude) and factor nodes connecting the state nodes. The system constructs a nonlinear least-squares cost function by minimizing the sum of Mahalanobis distances across all factor nodes, and iteratively solves this function using the Gauss-Newton method or the Levenberg-Marquardt algorithm to extract the optimal state at the current moment as the real-time position coordinates.

[0047] More importantly, in order to provide a reliable trigger for subsequent dynamic route planning, the system goes beyond simply outputting location coordinates and further extracts the marginal covariance matrix during the solution process. And based on this matrix, the real-time location confidence level is calculated. :

[0048] in, The set scale adjustment constant, This is the trace of the edge covariance matrix. Geometrically, this trace represents the volume of the positioning error ellipsoid. A smaller error ellipsoid volume indicates lower uncertainty in the current position calculation, resulting in higher positioning reliability.

[0049] Experimental scenario and comparative analysis: To conclusively demonstrate the significant advancement of this embodiment compared to the prior art, this embodiment provides an extreme transition scenario experiment of "entering a dimly lit cave from an outdoor plaza with bright light".

[0050] Comparative Analysis (Traditional EKF Algorithm): In the first 10 seconds after a tourist enters the cave, the GNSS signal is severely interfered with by multipath reflections from the rock walls, but is not completely lost. The traditional EKF algorithm only detects a decrease in the signal-to-noise ratio, and its covariance matrix adjustment has a significant lag. Experimental data shows that during the 10-second transition period, the positioning trajectory output by the comparative analysis experienced random jumps of up to 45 meters, causing frequent errors in the background route planning engine.

[0051] In this embodiment of the invention: the instant a visitor crosses the cave entrance, a light sensor detects a sharp drop in lux, and a barometer senses the unique pressure change characteristic of caves. The system instantly calculates the extremely large environmental change index. Substituting into the above exponential update formula, the covariance matrix of GNSS is exponentially amplified within 0.5 seconds, its observation weights are forcibly stripped away, and the system relies entirely on the preset high-weight Bluetooth fingerprint and inertial navigation calculations for takeover positioning.

[0052] Experimental data shows that within the same 10-second transition period, the smoothness of the positioning trajectory in this embodiment is improved by 92%, with the error controlled within 3 meters. More importantly, the system synchronously outputs real-time positioning reliability. After a controllable slight drop, the system quickly rebounded, accurately reflecting the brief period of uncertainty experienced by the system.

[0053] Furthermore, the specific implementation process of S300 is as follows: Phase 1: Quiet state based on dead reckoning (maintained by extremely low power consumption).

[0054] When the system determines that the terminal is not currently in a route deviation state and the real-time positioning reliability is output in step S200 When the threshold value is consistently greater than or equal to the preset safety threshold, the system forcibly controls the terminal's wireless radio frequency communication module (such as a 4G / 5G / Wi-Fi antenna) to enter a deep sleep state.

[0055] During this period of silence, the terminal completely cuts off the uplink data link with the edge server and cloud server. To ensure the continuity of the navigation location on the tourist interface, the system locally calls the acceleration and angular velocity output by the inertial measurement unit (IMU) and combines them with the locally cached local road network topology to perform short-range dead reckoning.

[0056] Meanwhile, the system monitors tourists' "route deviation status" in real time on a local coprocessor with extremely low computing power. The system extracts the terminal's current velocity vector. Tangential vector of the nearest road segment to the current location on the preset planned route Calculate the heading angle between the two. cosine value:

[0057] By introducing the geometric properties of the aforementioned vector dot product, the system can determine the relative movement trend of tourists with extremely high accuracy and low cost, without the need for complex map matching algorithms.

[0058] Phase Two: Abnormal Wake-up and Simplified Transmission under Strict Physical Constraints.

[0059] A hardware interrupt signal is generated to wake up the dormant communication module only when the terminal detects a breach of a specific physical or mathematical boundary. This embodiment sets two independent and extremely stringent event-driven conditions: Firstly (preventing false triggering due to signal-to-noise fluctuations): requiring real-time location reliability. The time span during which the values ​​remain below the preset safety threshold exceeds the set time tolerance window. This time window design cleverly filters out transient signal fluctuations caused by tourists walking past a single large tree or being briefly blocked by birds, thus avoiding ineffective wake-ups.

[0060] Secondly (to prevent accidental touches due to local deviation): the vertical projection distance from the tourist's actual coordinates to the preset route must be greater than the set spatial deviation tolerance. Furthermore, the cosine value of the heading angle calculated by the above formula It is consistently less than zero. In mathematical geometry, Meaning the included angle The angle exceeds 90 degrees, which definitively indicates that the tourists have not only gone astray, but are "walking in the opposite direction from their goal."

[0061] When any of the above conditions are met, the communication module is instantly woken up. At this time, the system executes a lightweight transmission strategy with "negative constraints": the real-time location coordinates, the type of event that triggered the wake-up (e.g., 0x01 represents confidence loss, 0x02 represents reverse deviation), and the current motion velocity vector are serialized and encapsulated to generate a minimally simplistic state feature packet. The data payload of this state feature packet is forcibly stripped (excluded) from the massive original environmental physical sampling sequence and multi-source positioning observation messages, and transmitted only through a single uplink handshake to the physically nearest edge server.

[0062] Long battery life and high concurrency resistance comparative experiment: To intuitively verify the significant technical effects brought about by this event-driven mechanism, this embodiment sets up a real-world comparison experiment of "4-hour hike in a mountain scenic area".

[0063] Experimental conditions: Two groups of tourists walked for 4 hours on the same mountain scenic route during a holiday when the network was congested, using a terminal equipped with a traditional high-frequency positioning and navigation scheme (comparative example) and a terminal equipped with the scheme of this embodiment (experimental example).

[0064] Comparative analysis: The traditional solution uses a fixed frequency of 1Hz to send full GPS messages to the server to request route verification. Over 4 hours, the terminal's radio frequency module worked at full load, consuming 42% of the terminal's battery power. At the same time, due to the excessive concurrency of base stations in scenic areas during holidays, the traditional terminal encountered an average packet loss delay of 1200 milliseconds when requesting route updates, resulting in frequent stuttering of the navigation screen.

[0065] The embodiments of this invention demonstrate that, during 92% of the normal walking time, the communication module of the experimental terminal remained in a zero-power sleep / silent state, relying entirely on formulaic vector comparison for local security verification. "Reverse walking" only occurred when a tourist mistakenly entered one of two forks in the road. Only then does the system precisely wake up the communication module and send a minimal status feature packet of less than 1KB.

[0066] Ultimately, the 4-hour hike consumed only 11% of the terminal's battery (nearly 4 times the battery life). Furthermore, due to the extremely small data packets and on-demand triggering, the actual uplink handshake latency was controlled to within 80 milliseconds even in extremely congested base station environments.

[0067] This embodiment addresses the technical pain points of traditional tourist navigation software, which requires high-frequency and continuous network connectivity, leading to rapid battery depletion of tourist terminals and network congestion at scenic area base stations. It overturns the traditional polling paradigm of "terminal periodically reporting coordinates" and introduces a "state machine event-driven mechanism" based on underlying hardware interrupts. The terminal's communication state is strictly divided into two mutually exclusive lifecycles: "silent calculation" and "abnormal wake-up".

[0068] Furthermore, the specific implementation process of S400 is as follows: This step establishes a hierarchical processing and time-sequence blocking mechanism based on "event severity level" between the edge server and the cloud server, which is divided into two independent processing branches: "emergency anti-lost" and "routine anti-congestion".

[0069] The first branch: Localized emergency planning for preventing people from getting lost based on physical barriers (for extreme getting lost scenarios).

[0070] When the edge server receives the status feature packet reported by the terminal, it first parses the event type code. If the code indicates that "real-time location reliability is lower than the preset safety threshold" (i.e., the tourist is in a deep signal blind zone or is lost), the edge server will trigger the highest priority emergency response.

[0071] To completely avoid uncontrollable network latency caused by cross-WAN transmission, the edge server immediately disconnects the route synchronization link between the current terminal and the cloud server at the hardware logic level. Subsequently, the edge server directly calls the "micro-security topology map" stored in its local memory.

[0072] The edge server uses the real-time location coordinates and current velocity vector carried in the state feature packet as the initial state (nodes and directions) for the algorithm search. Employing an improved heuristic search algorithm, it prioritizes searching the shortest connecting path in the micro-safety topology map to the nearest open physical safety area (such as a helicopter rescue point or plaza) or the nearest manual assistance base station. This path is generated as a "local anti-wandering emergency route" and distributed to the terminal with the highest communication privileges.

[0073] The second branch: Time difference collaboration and spatiotemporal graph convolutional global planning (for local congestion scenarios). If the status feature packet indicates that the "route deviation state meets the triggering condition" (such as the aforementioned reverse heading angle) and the real-time positioning reliability is at a normal level, the system determines that the tourist is actively avoiding congested crowds or obstacles ahead. At this time, the system activates the cloud-edge time difference coordination mechanism.

[0074] First, the edge server determines at the micro level whether the local micro-level pedestrian density within a preset physical radius around the current deviation location exceeds a congestion threshold. If the determination is true, the edge server directly generates an immediate turning instruction (e.g., "Congestion ahead, please turn right immediately onto the auxiliary road") based on the local micro-road network and issues it out rapidly.

[0075] After completing the millisecond-level "micro-level congestion avoidance" described above, the edge server then transmits its real-time location coordinates to the cloud server, which then performs macro-level replanning.

[0076] Derivation of a physically constrained spatiotemporal graph convolutional neural network on a cloud server: After receiving the transmitted coordinates, the cloud server invokes a pre-trained spatiotemporal graph convolutional neural network model to output the optimal global alternative route. To completely address the shortcomings of traditional algorithms that are detached from physical reality, this embodiment forcibly introduces "spatial-temporal joint physical constraint evaluation" in the graph space aggregation operation of the graph neural network.

[0077] Specifically, the cloud server performs the following physical quantization calculations for each candidate Point of Interest (POI) across the entire domain: Firstly, time-based accessibility filtering. The estimated physical walking time to a candidate point of interest is calculated based on real-time location coordinates. Extract the current system time. Cloud server comparison Remaining open time window for the candidate point of interest If the opening hours have expired, the candidate node will be pruned directly in the directed graph to avoid guiding visitors to attractions that are already closed.

[0078] Secondly, 3D terrain and physical attenuation penalty. The cloud server extracts real-time location coordinates to the digital elevation model (DEM) of the connecting road segment where the candidate point of interest is located, and calculates the cumulative elevation difference of the terrain undulation of the road segment. (Focus on accumulating all uphill elevation increments). Combined with the total distance walked by the tourist that day as recorded by the system, a tourist's physical exertion index is generated. Based on this, the physical exertion penalty weight is calculated. :

[0079] in, This is the terrain transformation coefficient. The formula indicates that the more fatigued the tourist is and the greater the elevation difference of the mountain road ahead, the more exponentially the penalty weight increases.

[0080] Third, dynamic aggregation and optimization of edge weights. The cloud server will use the calculated physical decay penalty weights... Real-time pedestrian flow thermal characteristics of this road section This serves as a dynamic adjustment parameter for the connection edges in the spatiotemporal graph convolutional neural network model. The model aggregates features within a multi-scale neighborhood through graph convolutional layers, calculating the total cost of each feasible path. Ultimately, the system selects the connected sequence with the lowest total value, generates the "optimal global alternative route" that perfectly matches the tourist's current remaining energy, avoids congestion, and ensures the attraction remains open, and sends it to the terminal.

[0081] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic planning method of a travel route based on multi-source positioning data fusion, characterized in that, Includes the following steps: S100: Acquire real-time environmental physical data and multi-source positioning data collected by the terminal's built-in sensors, and a pre-constructed environmental feature map, wherein the environmental feature map contains topological nodes associated with physical reference features; S200: Compare the real-time environmental physical data with the physical reference features at the corresponding topological node, dynamically adjust the observation weights of different positioning data sources in the factor graph optimization model according to the degree of difference in the comparison, and output the real-time location coordinates and real-time positioning reliability of the terminal through the factor graph optimization model. S300: Monitor the real-time location reliability and the route deviation status of the terminal. When the real-time location reliability is lower than a preset safety threshold or the route deviation status meets the triggering condition, wake up the terminal's sleep communication module and send a status feature packet containing the real-time location coordinates to the edge server. S400: The edge server responds to the state feature packet, blocks the conventional global planning based on the state of the real-time location confidence and forcibly generates a local anti-lost emergency route, or collaborates with the cloud server to generate a global alternative route based on the current scenic area congestion, and sends the generated dynamic route to the terminal.

2. The dynamic planning method for tourism routes based on multi-source positioning data fusion according to claim 1, characterized in that, The specific methods for acquiring the real-time environmental physical data and the multi-source positioning data include: The system utilizes the built-in ambient light sensor, barometer, and magnetometer of the terminal to collect real-time ambient light intensity, air pressure fluctuation frequency, and magnetic field strength variance at the current location, which serve as the real-time environmental physical data. The terminal's built-in wireless communication module and inertial measurement unit are invoked to collect the global navigation satellite system signal-to-noise ratio, Bluetooth low-power received signal strength indication, terminal acceleration, and angular velocity in real time as the multi-source positioning data.

3. The dynamic planning method for tourism routes based on multi-source positioning data fusion according to claim 2, characterized in that, The environmental feature map is a directed graph data structure containing multiple topological nodes and connecting edges, wherein: The topology node includes three-dimensional spatial coordinates and the physical reference features, which are composed of pre-collected historical ambient light illuminance reference values, historical magnetic field variance reference values, and low-power Bluetooth fingerprint reference vectors. The connecting edge includes the road segment connectivity attributes, physical walking distance, and historical average network signal coverage strength between two adjacent topological nodes.

4. The dynamic planning method for tourism routes based on multi-source positioning data fusion according to claim 3, characterized in that, The dynamic adjustment of observation weights from different location data sources in the factor graph optimization model based on the degree of difference in the comparison includes: Calculate the environmental abrupt change index between the real-time environmental physical data and the physical baseline characteristics at the corresponding topology node; Based on the environmental mutation index, the observation error covariance matrix of the multi-source positioning data in the factor graph optimization model is dynamically updated using a set decay function. The update formula is as follows: in, Indicates the first Location data source The observation error covariance matrix updated at each time step. This shows the initial observation error covariance matrix. Indicates the first The environmental sensitivity coefficient of the location data source This represents the environmental mutation index; When the environmental abrupt change index, which represents a sudden drop in ambient light intensity, exceeds a set threshold and the frequency of air pressure fluctuations meets indoor characteristics, the system automatically increases the environmental sensitivity coefficient corresponding to the global navigation satellite system, causing the corresponding observation error covariance matrix to be exponentially amplified, thereby reducing its observation weight in the factor graph optimization model.

5. The dynamic planning method for tourism routes based on multi-source positioning data fusion according to claim 4, characterized in that, The factor graph optimization model includes state nodes and factor nodes connected to the state nodes, wherein: The state node contains the physical location, velocity, and attitude of the terminal at various historical moments and the current moment; The factor nodes include kinematic prior factor nodes and multi-source observation factor nodes; The kinematic prior factor node is constructed based on the terminal acceleration and the angular velocity, and the multi-source observation factor node is constructed based on the signal-to-noise ratio of the global navigation satellite system and the low-power Bluetooth received signal strength indication; the information matrix of the multi-source observation factor node is the inverse matrix of the observation error covariance matrix corresponding to the positioning data source.

6. The dynamic planning method for tourism routes based on multi-source positioning data fusion according to claim 5, characterized in that, The step of solving and outputting the real-time location coordinates and real-time positioning reliability of the terminal through the factor graph optimization model includes: The optimal state node sequence is solved by minimizing the sum of Mahalanobis distances of all the factor nodes, and the optimal state at the current moment is extracted as the real-time position coordinates. Extract the marginal covariance matrix corresponding to the real-time location coordinates when solving the factor graph optimization model, and calculate the real-time location confidence using the following confidence evaluation formula: in, express The real-time location reliability at that moment. The set scale adjustment constant, Represents the edge covariance matrix The smaller the value of the trace, the higher the reliability of the real-time positioning.

7. The dynamic planning method for tourism routes based on multi-source positioning data fusion according to claim 6, characterized in that, Monitoring the real-time location reliability and the route deviation status of the terminal includes: If the triggering condition is not met and the real-time positioning confidence is greater than or equal to the preset safety threshold, the communication module of the terminal is controlled to remain in a sleep state, and dead reckoning is performed using the local road network topology cached locally by the terminal and the pose increment output by the inertial measurement unit, and the real-time position coordinates are updated. The vertical projection distance from the real-time position coordinates to the preset planned route is calculated in real time, and the heading angle between the current motion speed vector of the terminal and the tangential vector of the preset planned route is calculated in advance. These are combined as the route deviation state.

8. The dynamic planning method for tourism routes based on multi-source positioning data fusion according to claim 7, characterized in that, When the real-time positioning reliability is lower than a preset safety threshold or the route deviation state meets the triggering conditions, the terminal's sleep communication module is woken up, including: When any of the following event-driven conditions are detected, the system generates a hardware interrupt signal to wake up the communication module: Condition 1: The time span during which the real-time location confidence is continuously lower than the preset safety threshold exceeds the set time tolerance window. ; Condition 2: The vertical projection distance is greater than the set spatial deviation tolerance. Furthermore, the cosine value of the heading angle is consistently less than zero; Sending a status feature packet containing the real-time location coordinates to the edge server includes: The real-time position coordinates, the event type encoding that triggered the wake-up, and the current motion velocity vector are serialized and encapsulated to generate a lightweight state feature package; The data payload of the state feature packet excludes the original sampling sequence of the real-time environmental physical data and the original observation message of the multi-source positioning data, and transmits the state feature packet uplink once only through the wireless radio frequency channel to the edge service that is closest to the terminal in physical distance.

9. A dynamic planning method for tourism routes based on multi-source positioning data fusion according to claim 8, characterized in that, Based on the state-based real-time location confidence, conventional global planning is disrupted and local anti-wandering emergency routes are forcibly generated, including: When the event type encoding in the status feature packet indicates that the real-time location confidence is lower than the preset security threshold, the edge server immediately disconnects the route synchronization link between the current terminal and the cloud server. In the micro-security topology map stored locally on the edge server, using the real-time location coordinates and the current movement speed vector in the state feature packet as the initial state, the shortest connecting path to the nearest open physical safety area or the nearest manual assistance base station is searched and prioritized as the local anti-lost emergency route to block cross-regional network delays.

10. A dynamic planning method for tourism routes based on multi-source positioning data fusion according to claim 8, characterized in that, The collaborative cloud server generates global alternative routes based on the current congestion level of the scenic area, including: When the state feature package indicates that the route deviation state meets the triggering condition and the real-time positioning confidence is at a normal level, the edge server determines whether the local micro-human flow density within a preset physical radius around the current deviation position is greater than the congestion threshold. If the congestion threshold is exceeded, the edge server directly generates an instant turning instruction to avoid the current congested intersection based on the local micro-road network and sends it to the terminal. After issuing the instant turn command, the edge server transmits the real-time location coordinates to the cloud server; the cloud server calls the spatiotemporal graph convolutional neural network model, inputs the real-time human flow heat map features, opening time windows and tourist physical exertion index of each candidate point of interest in the whole domain, and outputs the optimal global alternative route to avoid high-risk congestion areas. When the cloud server calls the spatiotemporal graph convolutional neural network model to output the optimal global alternative route, it performs the following spatial-temporal joint physical constraint evaluation: Calculate the estimated physical walking time from the real-time location coordinates to the candidate point of interest, compare the estimated physical walking time with the remaining open time window of the candidate point of interest, and filter out nodes that are unreachable by time. Extract the cumulative elevation difference of the terrain undulations from the real-time location coordinates to the connecting road segment where the candidate point of interest is located, and calculate the physical exertion penalty weight by combining it with the tourist physical exertion index. The physical exertion penalty weight and the real-time human flow thermal features are used as dynamic adjustment parameters for the edge weights in the spatiotemporal graph convolutional neural network model. The connected sequence with the minimum total algebraic value is selected through graph space aggregation operation to generate the optimal global alternative route.

Images