Unmanned aerial vehicle monitoring, early warning and disposal method suitable for holiday tourist flow blowout of scenic spots

Abstract

The present application relates to the technical field of intelligent monitoring and emergency management, and particularly relates to a method for unmanned aerial vehicle monitoring, early warning and disposal suitable for holiday tourist flow blowout in scenic spots, comprising the steps of system initialization, unmanned aerial vehicle deployment, tourist flow data acquisition, tourist flow prediction and early warning, tourist flow source identification, cooperative scheduling planning, linkage disposal guidance, effect feedback optimization and closed-loop termination; by collecting scenic spot geographic data to divide monitoring units, a fluid mechanics model is constructed to predict tourist flow risks, the source of tourist flow is traced and unmanned aerial vehicles are dispatched, disposal is started in stages and dynamic guidance is provided, and closed-loop optimized operation is realized in combination with parameter feedback. In the present application, the tourist flow blowout is predicted in advance by relying on the continuity equation and Bernoulli equation, the early warning opportunity is moved forward to before congestion is formed, and the tourist flow is diverted from the source by combining precise tourist flow tracing, multi-vehicle cooperative scheduling, hierarchical linkage disposal and closed-loop self-optimization mechanism, thereby greatly improving the pertinence, timeliness and operation safety of holiday tourist flow control in scenic spots.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring and emergency management technology, and in particular to a method for monitoring, early warning and handling of drones in response to surges in tourist traffic during holidays in scenic areas. Background Technology

[0002] Tourist attractions are prone to explosive growth in visitor numbers during holidays, posing significant challenges to safety management and crowd control. To address this issue, existing technologies have adopted drone monitoring to assist in visitor flow management. Drones, equipped with image acquisition devices, capture images of visitor flow, analyze visitor density, identify congested areas, and provide managers with dynamic visitor flow information to assist in handling situations, thereby reducing the intensity of manual labor. However, existing drone-based passenger flow monitoring and early warning methods have a core technical problem: they can only identify and issue alarms for existing congestion, which is a passive "post-event detection and post-event handling" mode, unable to proactively predict the risk of passenger flow surges. By the time the passenger flow density reaches a preset threshold and triggers an alarm, congestion has already formed, and the delayed response can easily lead to safety hazards. Therefore, there is an urgent need for a technical solution that can predict the risk of passenger flow surges in advance and move the early warning timing forward, solving the core pain point of delayed early warning in traditional methods. Summary of the Invention

[0003] To overcome the above shortcomings, this invention provides a drone monitoring, early warning, and handling method for peak tourist flow in scenic areas during holidays, aiming to improve the problems of delayed early warning and reactive handling in existing technologies.

[0004] This invention provides the following technical solution, applicable to the monitoring, early warning, and handling of surges in tourist traffic during holidays in scenic areas, comprising the following steps: S1. System initialization: Collect scenic area geographic data to divide monitoring units, calibrate drone and visitor flow benchmark parameters, and configure ground control system communication. S2. Drone Deployment: The drone swarm establishes a self-organizing network, allocates drones according to monitoring units to plan initial routes, detects battery power and responds to return-to-base resupply requests; S3. Passenger flow data collection: Each UAV collects passenger flow images, extracts passenger flow density, flow speed, and flow direction parameters, completes fluid parameter mapping, and summarizes the parameters to the ground control system; S4. Passenger Flow Forecast and Early Warning: Based on the aggregated passenger flow parameters, a model is constructed by combining the continuity equation and Bernoulli equation to calculate the risk assessment and early warning level, and the forecast results are then issued. S5. Passenger Flow Source Identification: Based on the issued early warning instructions and passenger flow parameters, retrieve the electronic map to trace the source path, identify key nodes, and transmit the results back to the ground control system. S6. Collaborative scheduling and planning: Based on the returned traceability information and instructions, adjust the deployment and flight path of the UAV and avoid collisions, plan and guide the dedicated flight path of the UAV, and send out the scheduling parameters. The guiding UAV is equipped with a guidance projection device. S7. Coordinated Response Guidance: Based on the issued early warning level instructions, initiate corresponding response operations to guide the drone to project dynamic guidance markers at key nodes along the flight path, with response and projection executed simultaneously; S8. Effect Feedback Optimization: After the data is collected and processed, the parameters are summarized to the ground control system. The ground control system will feed back the parameters to update the model, optimize the rules, and adjust the plan. S9. Closed-loop termination: Repeat the closed-loop operation from S3 to S8. When the passenger flow density is lower than the first critical interval and the passenger flow dispersion is greater than or equal to 0 for a preset time, the command is terminated, the drone resumes normal mode, and a report is generated at the end of the holiday.

[0005] Preferably, in step S1, the system initialization specifically includes the following process: Collect data on terrain features, regional carrying capacity thresholds, and trail distribution in various areas of the scenic area, and divide the scenic area into N monitoring units according to terrain features and points prone to congestion. The system calibrates the cruising radius, endurance, autonomous return threshold battery level, and response speed of a single drone, as well as the benchmark values ​​for passenger flow density, flow speed, and flow direction. Configure communication links and data protocols between the ground control system and each processing stage to ensure real-time transmission of data and commands between stages.

[0006] Preferably, in step S2, the deployment of the drone specifically includes the following process: Once the drone swarm is started, it automatically establishes a distributed self-organizing network without a central node, sharing the location, flight path, and battery data of each drone in real time. Based on the results of the monitoring unit division, allocate 1-2 drones to each monitoring unit and plan the initial cruise routes along the boundaries of the monitoring units and points prone to passenger congestion. The drone monitors its own battery level in real time. When the battery level is lower than the autonomous return-to-home threshold, it sends a return-to-home request, matches the nearest drone mobile resupply station, and plans a return-to-home resupply route.

[0007] Preferably, in step S3, the passenger flow data collection specifically includes the following process: Each drone uses its onboard high-definition camera to collect real-time images of passenger flow from its monitoring unit, and uses image recognition algorithms to extract the number of tourists and their displacement data from a single frame image. The flow density is calculated based on the number of tourists and the actual area of ​​the scenic spot corresponding to the image. The flow velocity is calculated based on the displacement distance of tourists in two consecutive frames of images and the time interval between image acquisition. The flow direction is calculated based on the horizontal and vertical components of the tourist displacement. The calculated passenger flow density, flow speed, and flow direction parameters are aggregated in real time to the ground control system via a self-organizing network, and the ground control system distributes the parameters to subsequent processing stages.

[0008] Preferably, in step S4, the construction of the passenger flow fluid model specifically includes the following process: A three-dimensional continuity equation is established based on the law of conservation of mass, and then simplified into a one-dimensional continuity equation by combining the topographic features of the bottleneck area of ​​the scenic spot. Based on the law of conservation of energy and combined with the topographic slope of the scenic area and the regional carrying capacity threshold, the Bernoulli equation is established, and the critical value of passenger flow density in the bottleneck area is derived through the continuity equation and the Bernoulli equation. By combining real-time passenger flow parameters collected by drones, the model parameters are dynamically corrected. The passenger flow density time change rate, passenger flow dispersion, and passenger flow pressure value are calculated through the passenger flow fluid flow model. The warning level is determined based on the calculation results.

[0009] Preferably, in step S4, the early warning level determination and instruction issuance specifically include the following process: When the passenger flow density in the bottleneck area is in the first critical range and the flow speed continues to decrease, a Level 1 warning is triggered. A level-two warning is triggered when the passenger flow density is in the second critical range and the passenger flow dispersion is less than 0. A Level 3 warning is triggered when the passenger flow density is greater than or equal to the third critical interval or the passenger flow pressure value is greater than or equal to the congestion critical pressure value. The warning level, risk area, predicted time and passenger flow parameters are aggregated and sent to the ground control system, which then issues warning instructions simultaneously.

[0010] Preferably, in step S5, the passenger flow tracing and identification specifically includes the following process: Based on the issued early warning instructions, passenger flow direction and flow speed data, retrieve the scenic area's electronic map; Based on passenger flow direction and speed, the passenger flow origin path algorithm is used to trace the passenger flow origin path of risk areas and form a set of passenger flow origin paths; Identify key nodes such as entrance diversion points, pedestrian intersections, and parking lot exits in the passenger flow origin path, extract the coordinates of key nodes, and transmit the passenger flow origin path information and key node coordinates back to the ground control system.

[0011] Preferably, in step S6, the collaborative scheduling plan specifically includes the following process: Based on the returned passenger flow source path information, key node coordinates and early warning instructions; Increase the number of drones deployed at key nodes along passenger flow origin routes and increase the frequency of drone patrols in key node areas; The patrol range of drones in risk areas is adjusted according to the warning level. Level 1 warnings reduce the patrol range, while Level 2 and above warnings focus on the core risk areas. By sharing drone location and flight path data through a self-organizing network, a drone collision avoidance algorithm combining distance priority and speed adjustment is adopted. To guide drones in planning dedicated routes along key nodes of passenger flow origin paths, and to plan return and resupply routes based on the drones' battery levels.

[0012] Preferably, in step S7, the coordinated handling guidance specifically includes the following process: Based on the issued warning level instructions, the corresponding response operations will be automatically initiated; Level 1 alerts trigger voice broadcasts to provide passenger flow information; Level 2 alerts trigger voice broadcasts and warning lights to broadcast flow restriction notices and project warning lights; Level 3 alerts trigger voice broadcasts, warning lights, temporary fencing, and location markers, including the deployment of temporary fencing and location of the congestion source. The drones are guided to fly along a dedicated route, and dynamic guidance signs are projected at key nodes along the passenger flow path. The content of the guidance signs is dynamically adjusted based on the real-time passenger flow data, so that the handling operations and dynamic guidance are executed simultaneously.

[0013] Preferably, in step S8, the effect feedback optimization specifically includes the following process: The system collects and processes passenger flow density, flow velocity, and flow direction parameters, which are then aggregated to the ground control system. The processed parameters are fed back to update the total energy constant and various critical values ​​of the passenger flow fluid flow model. The processed parameters are also fed back to optimize the identification rules for key nodes in the passenger flow source path. Finally, the processed parameters are fed back to adjust the deployment and patrol plans of UAVs.

[0014] The present invention has the following beneficial effects: 1. In this invention, by introducing the continuity equation and Bernoulli equation to construct a passenger flow fluid flow model, the tourist group is regarded as a continuous fluid medium for dynamic analysis, which realizes the forward prediction of the risk of passenger flow surge. It solves the problem of the lag in the traditional method that can only detect congestion after the fact, and moves the early warning time to before the congestion is formed, thus gaining valuable time for subsequent handling.

[0015] 2. In this invention, the source path algorithm is used to trace the source path of passenger flow in risk areas based on passenger flow direction and flow speed, and to identify key nodes such as entrance diversion points and pedestrian intersections. This enables accurate source tracing of passenger flow from congestion points, allowing intervention measures to be implemented from the source rather than just at the end, significantly improving the pertinence and effectiveness of traffic management.

[0016] 3. In this invention, the location, flight path and power data of the UAV swarm are shared in real time through a self-organizing network, and a collision avoidance algorithm combining distance priority and speed adjustment is adopted. At the same time, the number of UAVs deployed and the patrol frequency of key nodes are dynamically adjusted according to the traceability information, so as to realize the resource optimization and safe operation guarantee under multi-aircraft collaborative operation.

[0017] 4. In this invention, a graded response operation is initiated according to the warning level, and the drone is simultaneously guided to project dynamic guidance signs at key nodes. This enables voice broadcasts, warning lights, and guidance light strips to work together, achieving a gradient response from mild warnings to emergency intervention. This avoids the impact of excessive intervention on the tourist experience while ensuring rapid control under serious risks.

[0018] 5. In this invention, the passenger flow parameters collected after the treatment are fed back to the fluid flow model, key node identification rules and UAV deployment scheme, so as to realize the dynamic correction of model parameters and the continuous optimization of scheduling strategy, forming a complete closed-loop learning mechanism, enabling the system to continuously improve itself according to the actual treatment effect and adapt to passenger flow changes in different time periods and scenarios. Attached Figure Description

[0019] Figure 1 This is a flowchart of the drone monitoring, early warning, and handling method for surges in tourist traffic during holidays, as proposed in this invention. Detailed Implementation

[0020] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0021] In embodiments of the present invention, the present invention provides a method for monitoring, early warning, and handling of sudden surges in tourist traffic during holidays in scenic areas. The following example uses a mountainous scenic area. Figure 1 As shown, it includes the following steps: S1. System initialization: Collect scenic area geographic data to divide monitoring units, calibrate drone and visitor flow benchmark parameters, and configure ground control system communication. Furthermore, the system initialization specifically includes the following process: Collect data on terrain features, regional carrying capacity thresholds, and trail distribution in various areas of the scenic area, and divide the scenic area into N monitoring units according to terrain features and points prone to congestion. The system calibrates the cruising radius, endurance, autonomous return threshold battery level, and response speed of a single drone, as well as the benchmark values ​​for passenger flow density, flow speed, and flow direction. Configure communication links and data protocols between the ground control system and each processing stage to ensure real-time transmission of data and commands between stages.

[0022] Specifically, system initialization is the foundation of the entire method, aiming to complete the collection of scenic area geographic information, the division of monitoring units, the calibration of key parameters, and the configuration of communication links, so as to provide necessary data support and hardware guarantee for subsequent collaborative operation of UAVs and passenger flow analysis.

[0023] Geographic data collection for the scenic area: First, using drone aerial photography, satellite remote sensing, or existing GIS data, data on the terrain features, regional carrying capacity thresholds, and trail distribution of each area within the target scenic area are collected. Terrain features include altitude, slope, and topographic relief, used to assess the accessibility of different areas; the regional carrying capacity threshold refers to the maximum number of visitors that can be accommodated per unit area, pre-set according to scenic area management regulations and safety standards; trail distribution data records the start and end points, width, length, and intersection points of all tour routes. This data is integrated into a 3D digital map of the scenic area and stored in the ground control system.

[0024] Monitoring Unit Division: Based on collected terrain features and historically frequent congestion points, the scenic area is divided into N monitoring units. The division principle is that each monitoring unit should have relatively consistent terrain attributes and independent visitor flow characteristics to facilitate targeted monitoring by drones. Therefore, a congestion risk index R is defined for each grid area i. i : ; Where: θ i W represents the average slope of region i, in degrees, reflecting the steepness of the terrain. i P represents the average width of the main walkways within area i, in meters. Smaller widths indicate greater congestion. i The probability of congestion in the area during the same period in history is represented by a value ranging from 0 to 1, which is derived from the statistical data of visitor flow in previous years. α, β, and γ are weighting coefficients that can be set according to the actual management experience of the scenic area, for example, α=0.3, β=0.4, and γ=0.3.

[0025] Calculate R for each grid cell iThen, a region growing algorithm is used to merge adjacent grids with similar risk indices to form monitoring units. Taking the target scenic area as an example, areas with higher risk indices, such as the entrance plaza, core viewing platform, and narrow walkways, are divided into independent monitoring units; while gentler recreational areas are merged into larger units. The final set of monitoring units is U = {U1, U2, ..., U3}. N Each unit has a unique identifier and spatial scope.

[0026] UAV and Passenger Flow Benchmark Parameter Calibration: To ensure the efficient and safe operation of the UAV swarm, it is necessary to calibrate the key performance parameters of individual UAVs. Cruise radius r u The maximum distance a drone can fly stably with a full charge. Battery life t u : The longest operating time after a single charge; Autonomous return threshold battery level e th : Remaining battery percentage that triggers automatic return to home; Response speed of handling operations v resp The average delay between receiving instructions and completing the corresponding action by the drone.

[0027] Passenger flow baseline parameters are used for subsequent real-time comparisons and predictions, including: Passenger flow density benchmark value ρ base The average density of monitoring units during non-congestion periods is calculated using the following formula: ; Where ρ t The data represents the passenger flow density at each sampling time during the same historical period, in person / m², where T is the total number of sampling times.

[0028] Flow velocity reference value v base Average walking speed of tourists, in m / s, usually taken as 0.8 to 1.2 m / s, which can be adjusted according to the slope of the scenic trail.

[0029] Flow reference value φ: The angle of the mainstream direction of tourists in the absence of guidance, in degrees, obtained through statistical analysis of historical trajectory data.

[0030] Taking the target scenic area as an example, by analyzing the visitor flow data of previous National Day Golden Weeks, the ρ of the entrance plaza was calculated. base =0.5 people / m², with 0.8 people / m² at the core viewing platform; average visitor speed v base =1.0m / s; the main direction of traffic is along the trail towards the exit.

[0031] Communication Link Configuration: A stable, low-latency communication link must be established between the ground control system and all UAVs and data processing components. The configuration employs a self-organizing network protocol to ensure real-time sharing of location, flight path, and battery data among UAVs. The ground control system issues commands to UAVs and receives collected passenger flow parameters via a dedicated data link. Simultaneously, data interfaces and transmission protocols between each processing component are defined to ensure smooth data flow. For example, all passenger flow parameters are specified to be encapsulated in JSON format and transmitted via the MQTT protocol to ensure real-time performance.

[0032] Through the above initialization work, the target scenic area has the basic conditions for implementing subsequent monitoring and early warning: the ground control system stores detailed geographic information and benchmark parameters, the drone swarm has completed performance calibration and established reliable communication with the ground, and the division of monitoring units provides spatial granularity for refined management and control.

[0033] S2. Drone Deployment: The drone swarm establishes a self-organizing network, allocates drones according to monitoring units to plan initial routes, detects battery power and responds to return-to-base resupply requests; Furthermore, the deployment of the drone specifically includes the following process: Once the drone swarm is started, it automatically establishes a distributed self-organizing network without a central node, sharing the location, flight path, and battery data of each drone in real time. Based on the results of the monitoring unit division, allocate 1-2 drones to each monitoring unit and plan the initial cruise routes along the boundaries of the monitoring units and points prone to passenger congestion. The drone monitors its own battery level in real time. When the battery level is lower than the autonomous return-to-home threshold, it sends a return-to-home request, matches the nearest drone mobile resupply station, and plans a return-to-home resupply route.

[0034] Specifically, after system initialization, the drone swarm enters the deployment phase. The core tasks of this step are establishing a collaborative communication network among the drones, rationally allocating drone resources based on the division of monitoring units, planning initial cruise routes, and establishing a battery monitoring and autonomous return-to-base resupply mechanism, laying the foundation for subsequent passenger flow data collection and emergency response. Taking a mountainous scenic area as an example, the area is divided into N=8 monitoring units, covering key areas such as the entrance plaza, the core viewing platform, and narrow walkways.

[0035] Self-organizing network establishment: After the drone swarm is launched, a distributed self-organizing network without a central node is automatically established through the onboard communication module. This network employs a dynamic source routing protocol or an optimized link-state routing protocol, enabling each drone to act as both a data acquisition node and a relay node, forwarding information from other drones. After the network is established, each drone broadcasts its own position coordinates (x, y, y) in real time. u ,y u ,z u), current route planning and remaining battery percentage e u This enables full network status sharing. This decentralized structure avoids single points of failure and enhances system robustness. In the target scenic area, the eight drones completed self-organizing a network within 30 seconds of takeoff, and the ground control system could obtain the real-time status of the entire fleet through any one of the drones.

[0036] UAV allocation for monitoring units: Based on the set of monitoring units U={U1,U2,…,U...} partitioned in step S1 N To ensure comprehensive coverage and redundancy, a certain number of drones need to be allocated to each unit. The allocation strategy comprehensively considers the area of ​​the monitoring unit, historical congestion risk, and the coverage capability of a single drone. The monitoring unit U is defined. i Coverage demand index D i : ; Among them: A i For monitoring unit U i The area, in m², is calculated from the geographic data collected in step S1; R i For monitoring unit U i The congestion risk index is based on the one defined in step S1. The weighting coefficients α, β, and γ have been pre-calibrated; C represents the effective coverage capability of a single drone, in m² / drone, which indicates the maximum area that a drone can clearly monitor at a typical cruising altitude. This parameter is obtained by pre-calibrating the drone's camera field of view and cruising altitude.

[0037] Then it is assigned to the monitoring unit U i Number of drones n i for: ; In the formula This indicates rounding up to ensure at least one drone is allocated to each unit. For the target scenic area, the entrance plaza area A1 = 5000m², risk index R1 = 0.8, and single-drone coverage C = 4000m², resulting in D1 = 1.0, therefore n1 = 1 drone is allocated; the core viewing platform area A2 = 3000m², risk index R2 = 1.5, resulting in D2 = 1.125, rounded up to allocate n2 = 2 drones to address high congestion risk. The total number of allocated drones must match the actual available fleet size. If ∑n i If the number of drones exceeds the total number of drones, priority will be given to protecting high-risk units.

[0038] Initial cruise route planning: Plan an initial cruise route for each UAV to ensure efficient coverage of the assigned monitoring unit. The principle of route planning is to fly along the boundary of the monitoring unit and pass through points within the unit prone to congestion, such as intersections and viewpoint entrances. For each monitoring unit U... i First, extract the set of vertex coordinates B of its boundary polygon. i ={b i1 ,b i2 ...}, and the predefined set of coordinates of easily congested points P within the unit. i ={p i1 ,p i2 The initial flight path is generated using a key-point-based round-trip path algorithm: starting from a point on the boundary of the monitoring unit, all congestion-prone points are visited sequentially, and finally the path returns to the starting point along the boundary, forming a closed loop. The sequence of flight path points is denoted as w1, w2, ..., w M Each point contains three-dimensional coordinates. The drones fly in a cyclical pattern according to this sequence to achieve continuous monitoring. For a monitoring unit assigned two drones, the flight path can be divided into two segments, with the two drones cruising in opposite directions to shorten the patrol cycle.

[0039] Battery level monitoring and return-to-home resupply response: The drone continuously monitors its remaining battery level during flight. u And with the preset autonomous return threshold battery level e th Compare. When e u <e th Immediately, the UAV sends a return-to-home request to the ground control system and nearby UAVs via a self-organizing network. The request includes the current coordinates and remaining battery power. The ground control system uses a shortest path algorithm to find the nearest resupply station based on the real-time location of each UAV and the locations of pre-deployed mobile resupply stations. Let the coordinates of the resupply station be s. k If the current position of the returning drone is u, then choose to use Euclidean distance. The smallest resupply station. The system then plans a safe return route for the drone from its current location to the resupply station and updates the shared information in the ad hoc network. Other drones adjust their routes to avoid conflict with the returning drone. After resupply, the drone automatically rejoins the fleet and is reassigned tasks based on current needs. At the target scenic area, a drone responsible for the core viewing platform requested resupply when its battery level dropped to 20%. The system matched it with a mobile resupply station only 200 meters away, planned a straight return route, and the drone completed resupply and took off again 5 minutes later. The entire process did not affect the continuity of monitoring.

[0040] S3. Passenger flow data collection: Each UAV collects passenger flow images, extracts passenger flow density, flow speed, and flow direction parameters, completes fluid parameter mapping, and summarizes the parameters to the ground control system; Furthermore, the passenger flow data collection specifically includes the following process: Each drone uses its onboard high-definition camera to collect real-time images of passenger flow from its monitoring unit, and uses image recognition algorithms to extract the number of tourists and their displacement data from a single frame image. The flow density is calculated based on the number of tourists and the actual area of ​​the scenic spot corresponding to the image. The flow velocity is calculated based on the displacement distance of tourists in two consecutive frames of images and the time interval between image acquisition. The flow direction is calculated based on the horizontal and vertical components of the tourist displacement. The calculated passenger flow density, flow speed, and flow direction parameters are aggregated in real time to the ground control system via a self-organizing network, and the ground control system distributes the parameters to subsequent processing stages.

[0041] Specifically, after the drones are deployed, each drone continuously patrols its assigned monitoring unit according to its planned initial route, collecting passenger flow data in real time. The goal of this step is to transform the raw image information into quantified passenger flow parameters, including passenger density, flow speed, and flow direction, and to complete fluid parameter mapping, providing basic data for subsequent passenger flow prediction and source identification. Taking the target scenic area as an example, two drones, numbered UAV-2-1 and UAV-2-2, responsible for the core viewing platform, fly in opposite directions in a loop, collecting high-definition images every Δt = 2 seconds.

[0042] Image Acquisition and Tourist Identification: Each drone is equipped with an onboard high-definition camera that acquires overhead images at a fixed sampling frequency of f=0.5Hz. The image resolution is 1920×1080, and the size of the ground area covered depends on the drone's current flight altitude. Assuming the drone's cruising altitude is h=50m and the camera's field of view is 60°, the actual ground area corresponding to a single frame is approximately 86.6×48.8m², with an area of ​​A. frame ≈4225m². This area can be adjusted in real time according to the drone's altitude and recorded in the image metadata.

[0043] For each frame of the image, a lightweight object detection algorithm based on deep learning is used to identify tourist heads or human bodies in the image, and the number of tourists N is obtained. people Simultaneously, the displacement of the same tourist in two adjacent image frames is tracked using feature point matching or optical flow. To reduce computational load, only some salient feature points in the image can be extracted for tracking, obtaining the pixel displacement vector (Δu, Δv) for each feature point.

[0044] Passenger flow density calculation: Passenger flow density ρ is defined as the number of tourists per unit area within the monitoring unit. For a single frame image, the instantaneous density corresponding to that frame is: ; Where N people For the number of tourists identified in this frame, Aframe The area represents the actual ground area corresponding to the image, in m². Since drones may be located at different positions within the monitoring unit, a single frame density only reflects a local area. Therefore, a spatially weighted average of the densities from multiple sampling points is needed to obtain the overall density of the monitoring unit. However, for simplicity, the single frame density is used here as the instantaneous value at that location at that moment. The subsequent prediction module will fuse multi-frame data from multiple drones.

[0045] At a certain moment, UAV-2-1, located at the core viewing platform of the target scenic area, captured an image showing 87 tourists. The drone's altitude at this time was 52 meters, corresponding to an area A. frame =4500m², then ρ frame =87 / 4500≈0.0193 people / m². This value is far below the scenic area's carrying capacity threshold, indicating that the area is currently relatively unobstructed.

[0046] Passenger flow velocity calculation: Passenger flow velocity v reflects the speed at which tourists move. By tracking the displacement of the same feature point in two consecutive image frames and combining this with the image acquisition time interval Δt, the instantaneous velocity of that point can be calculated. For frame t and frame t+1, let the pixel coordinates of a feature point be (u... t ,v t ) and (u t+1 ,v t+1 First, the pixel displacement needs to be converted into actual ground displacement. This requires coordinate transformation using the camera's intrinsic and extrinsic parameters. For simplification, we can assume the ground is flat and the camera is pointing vertically downwards. The conversion relationship between pixel displacement and actual displacement is then: ; Where h is the drone's altitude above the ground, in meters; f pix The equivalent focal length in pixels can be obtained through camera calibration. However, for a more precise description, we use the following formula to calculate the actual ground velocity vector of this feature point: ; Where M is the transformation matrix from the pixel coordinate system to the ground coordinate system, containing height and camera intrinsic parameters. For simplicity, we directly give the formula for calculating the velocity magnitude here. In practical systems, the velocity vectors of all valid tracking points in the image are usually statistically analyzed, and the average value is taken as the overall passenger flow velocity v in that area. The velocity vector includes magnitude v = |v| and direction φ. The formula for calculating the velocity magnitude v is: ; In the formula: K is the number of feature points successfully matched in two consecutive frames; Δx k Δy represents the displacement component of the k-th feature point from frame t to frame t+1 in the ground coordinate system, in meters; Δt represents the inter-frame time interval, in seconds.

[0047] In the target scenic area, 150 feature points were successfully matched in two consecutive frames of images from UAV-2-1. The average displacement Δs = 0.8m and Δt = 2s were calculated, so the passenger flow speed v = 0.8 / 2 = 0.4m / s. Since many tourists stop to take photos in the viewing platform area, the speed is relatively low, which is in line with expectations.

[0048] Passenger flow direction calculation: The passenger flow direction φ represents the angle of the mainstream direction of tourist movement, defined with true north as 0°, and clockwise or counterclockwise. Similarly, based on the displacement vectors of matched feature points, the average displacement direction of all valid points is calculated. Let φ be the angle between the displacement vector of the k-th feature point and true north. k The overall flow direction φ can then be obtained through vector composition or circular averaging. A common method is to first calculate the average components of the displacement vector: ; The flow direction angle φ is: ; According to , The sign is adjusted in the quadrant. For example, the target scenic area is calculated at a certain time. =0.3m, =0.6m, then φ=arctan(0.3 / 0.6)≈26.6°, that is, the tourists are moving in a direction of approximately 26.6° north of east.

[0049] Fluid parameter mapping and aggregation: The calculated instantaneous parameters of a single frame are encapsulated into a data packet along with the corresponding monitoring unit identifier, timestamp, and UAV location information, and transmitted in real time to the ground control system via a self-organizing network. The ground control system performs spatiotemporal alignment and fusion of the received data according to the monitoring unit, for example, by performing a moving average of multiple sampled values ​​from the same monitoring unit over a short period of time to obtain a more stable parameter estimate. The fused parameters are then distributed to subsequent processing stages: the fluid-based passenger flow prediction module uses density and speed for risk prediction, while the passenger flow tracing module traces the source of congestion based on flow direction and speed.

[0050] Taking the target scenic area as an example, UAV-2-1 and UAV-2-2 each generate a set of data every 2 seconds. After receiving the data, the ground control system averages the 30 sets of data from the core viewing platform unit within a 1-minute window to obtain the current average density ρ = 0.02 people / m², average speed v = 0.45 m / s, and mainstream direction φ = 25° for that unit. These parameters are then fed into the prediction model in step S4.

[0051] S4. Passenger Flow Forecast and Early Warning: Based on the aggregated passenger flow parameters, a model is constructed by combining the continuity equation and Bernoulli equation to calculate the risk assessment and early warning level, and the forecast results are then issued. Furthermore, the construction of the passenger flow fluid flow model specifically includes the following process: A three-dimensional continuity equation is established based on the law of conservation of mass, and then simplified into a one-dimensional continuity equation by combining the topographic features of the bottleneck area of ​​the scenic spot. Based on the law of conservation of energy and combined with the topographic slope of the scenic area and the regional carrying capacity threshold, the Bernoulli equation is established, and the critical value of passenger flow density in the bottleneck area is derived through the continuity equation and the Bernoulli equation. By combining the passenger flow parameters collected in real time by UAVs, the model parameters are dynamically corrected. The passenger flow density time change rate, passenger flow dispersion and passenger flow pressure value are calculated by the passenger flow fluid flow model. The warning level is determined by combining the calculation results. Furthermore, the warning level determination and instruction issuance specifically include the following process: When the passenger flow density in the bottleneck area is in the first critical range and the flow speed continues to decrease, a Level 1 warning is triggered. A level-two warning is triggered when the passenger flow density is in the second critical range and the passenger flow dispersion is less than 0. A Level 3 warning is triggered when the passenger flow density is greater than or equal to the third critical interval or the passenger flow pressure value is greater than or equal to the congestion critical pressure value. The warning level, risk area, predicted time and passenger flow parameters are aggregated and sent to the ground control system, which then issues warning instructions simultaneously.

[0052] Specifically, after obtaining real-time passenger flow parameters such as density ρ, velocity v, and flow direction φ, the core of this step is to construct a passenger flow model using the continuity equation and Bernoulli equation from fluid mechanics. This model dynamically analyzes the passenger flow status within the monitoring unit, predicts potential congestion risks, and triggers corresponding levels of early warning based on the degree of risk. Taking the core viewing platform of the target scenic area as an example, the terrain of this area is a platform-type viewing platform with a gradually narrowing walkway at the exit, forming a typical "bottleneck" area.

[0053] Passenger Flow Model Construction: The passenger flow is considered as a continuous fluid medium, whose motion satisfies the fundamental laws of mass and energy conservation. First, based on the law of mass conservation, the continuity equation for passenger flow motion in three-dimensional space can be expressed as: ; Where ρ is the passenger flow density, in people / m²; v is the passenger flow velocity vector, in m / s; and t is time, in seconds. • This is the divergence operator. Since scenic trails typically have a clear directionality, and the visitor flow parameters collected by drones have been aggregated by monitoring unit, the problem can be simplified to a one-dimensional form. Considering the position coordinate x of the bottleneck area along the trail direction, the one-dimensional continuity equation is: ; In the equation, v(x,t) represents the velocity component along the path. This equation shows that the rate of change of density at a certain point over time is equal to the gradient of passenger flow into that point.

[0054] Secondly, based on the law of conservation of energy, Bernoulli's equation is introduced to describe the energy relationship of passenger flow under ideal conditions. For an incompressible fluid along a streamline, Bernoulli's equation is: p+1 / 2ρv 2 +ρgh = constant; Where p represents pressure, analogous to the "crowding pressure" experienced by tourists, with units of Pa; g is the acceleration due to gravity, and h is the height. However, in a scenic environment, terrain slope and psychological factors affecting tourists are more crucial. Therefore, we construct a simplified Bernoulli equation suitable for passenger flow. Considering two cross-sections before and after the bottleneck area, neglecting the influence of height changes, and introducing "passenger flow pressure" p to represent the degree of crowding felt by tourists, the Bernoulli equation can be written as: p1 + 1 / 2ρ1v1 2 =p² + 1 / 2ρ²v² 2 In the formula, subscripts 1 and 2 represent the upstream and bottleneck points, respectively. According to the law of conservation of mass, the passenger flow through the bottleneck should satisfy: ρ1vA1 = ρ2v2A2; where A1 and A2 are the cross-sectional areas of the walkway at the upstream and bottleneck points, respectively. From the above formula, the relationship between the density ρ2 at the bottleneck and the density ρ1 at the upstream can be derived. When the density at the bottleneck reaches the critical value ρ... crit At this point, the flow velocity drops sharply, leading to congestion. Through theoretical analysis or historical data calibration, the critical value ρ for passenger flow density in the bottleneck area can be obtained. In the target scenic area, by analyzing data from previous years, the critical density ρ for the walkway at the exit of the core viewing platform was determined. crit =0.8 people / m².

[0055] Real-time risk parameter calculation: Based on passenger flow parameters collected in real time by drones, the ground control system calculates the following three key indicators for each monitoring unit to quantify congestion risk: Passenger flow density time variation rate This indicator reflects the trend of density change over time. For monitoring unit U i Approximated by the difference in density values ​​between adjacent time points: ; Where ΔT is the sampling interval. If this value is positive and large, it indicates that the density is increasing rapidly, suggesting that congestion may worsen. For example, at a certain moment, ρ(t) = 0.65 people / m² at the core viewing platform of the target scenic area, and at the previous moment, ρ(t) ΔT) = 0.60 people / m², ΔT = 5s, then =0.01 people / (m²·s), indicating that the density is slowly increasing.

[0056] Passenger flow dispersion (ρv) According to the continuity equation, divergence directly determines the density change. In the one-dimensional case, divergence can be approximated as: ; Where Δx is the spatial step length along the walkway. A negative divergence indicates that inflow exceeds outflow, and passenger flow is accumulating in the area. In practical systems, since changes along the path cannot be directly measured, data from drones before and after the bottleneck area can be used for estimation. A simpler approach is to directly calculate the net inflow rate within the monitoring unit: if the density increases and the velocity decreases within the unit, it usually means that the divergence is negative. In the target scenic area, the flux divergence value at the viewing platform exit can be estimated by coordinating multiple drones. For example, based on drone data, the flux entering the viewing platform upstream is ρ1v = 0.5 × 1.0 = 0.5 people / (m·s), and the flux exiting downstream is ρ2v2 = 0.6 × 0.4 = 0.24 people / (m·s), then the divergence is approximately negative, indicating that passenger flow is accumulating.

[0057] The passenger flow pressure value *p* can be defined by Bernoulli's equation, and it comprehensively reflects the effects of density and speed. When no direct pressure measurement method is available, the following empirical formula can be used for estimation: p=k·ρ·v 2 ; Where k is an empirical coefficient, related to the scenic area environment. This formula originates from the concept of dynamic pressure; the larger p is, the stronger the coupling between passenger flow energy and density, and the higher the risk of congestion. In the target scenic area, we take k=0.5. If ρ=0.6 people / m² and v=0.4m / s, then p=0.5×0.6×0.4 2 =0.048Pa. This is compared to the preset congestion threshold pressure value p. crit By comparison, it can be determined whether a Level 3 warning has been reached.

[0058] Warning Level Determination and Command Issuance: The ground control system dynamically determines the warning level based on the above three indicators and a preset threshold range. The warning level is divided into three levels, corresponding to different levels of risk and response requirements: Level 1 Warning, Mild Risk: When the passenger flow density ρ in the bottleneck area is within the first critical range ρ crit1 ≤ρ<ρ crit2 Furthermore, the flow velocity v continuously decreases, for example, triggered when v decreases for three consecutive sampling periods. The first critical interval is typically below the critical density, but an upward trend has already been observed. For example, the target scenic area is set with ρ... crit1 =0.4 people / m², ρ=0.6 people / m². If the density of people on the observation deck is 0.5 people / m², and the speed decreases from 0.5m / s to 0.45m / s and then to 0.4m / s, a level one warning will be triggered, indicating that attention is needed.

[0059] Level 2 warning, moderate risk: when passenger flow density ρ is in the second critical range ρ crit2 ≤ρ<ρ crit3 And passenger flow dispersion Triggered when (ρv) < 0. The second critical interval is close to the critical density, and a negative divergence indicates that accumulation is occurring. For the target scenic area, ρ = 0.6 people / m². crit3 =ρ=0.8 people / m². If the density rises to 0.7 people / m² and the estimated divergence is negative, a level-two warning will be triggered, and evacuation preparations should be made.

[0060] Level 3 warning, severe risk: When passenger flow density ρ ≥ ρ or passenger flow pressure value p ≥ p crit This is triggered when congestion has occurred or is about to occur, requiring immediate intervention. For example, if the density at the observation deck reaches 0.85 people / m², or the pressure value exceeds 0.1 Pa, a Level 3 warning is triggered, and emergency response is initiated.

[0061] In the target scenic area example, at a certain moment, the density of the core viewing platform was 0.75 people / m², still below the critical value, but the divergence was negative and the velocity continued to decrease, so the system judged it as a level two warning. The ground control system immediately encapsulated the warning level, risk area, predicted time, and current passenger flow parameters into a warning command, and issued it to the passenger flow tracing module and subsequent handling module in step S5 through the self-organizing network. At the same time, the monitoring interface of the ground control system displayed the warning information to remind management personnel to pay attention.

[0062] S5. Passenger Flow Source Identification: Based on the issued early warning instructions and passenger flow parameters, retrieve the electronic map to trace the source path, identify key nodes, and transmit the results back to the ground control system. Furthermore, the passenger flow tracing and identification specifically includes the following process: Based on the issued early warning instructions, passenger flow direction and flow speed data, retrieve the scenic area's electronic map; Based on passenger flow direction and speed, the passenger flow origin path algorithm is used to trace the passenger flow origin path of risk areas and form a set of passenger flow origin paths; Identify key nodes such as entrance diversion points, pedestrian intersections, and parking lot exits in the passenger flow origin path, extract the coordinates of key nodes, and transmit the passenger flow origin path information and key node coordinates back to the ground control system.

[0063] Specifically, upon receiving the warning instruction issued in step S4, the ground control system needs to quickly locate the source of passenger flow causing congestion in the risk area and identify key traffic management nodes along the route. The core of this step is to construct a topology map of the scenic area's road network. Based on real-time passenger flow direction and speed data, a reverse tracing algorithm is used to trace possible source paths, providing precise target locations for subsequent drone scheduling and dynamic guidance. Taking a level-two warning triggered at the core viewing platform of the target scenic area as an example, the coordinates of its risk area are r=(xr ,y r At this time, the direction of passenger flow collected by the drone is φ=25°, that is, the tourists are moving in a roughly north-northeast direction with a flow speed of v=0.4m / s.

[0064] Scenic Area Road Network Map Construction: First, the ground control system retrieves the electronic map of the scenic area stored in step S1, extracts all geographical elements such as trails, intersections, attraction entrances, and parking lot exits, and constructs a directed graph model G=(V,E). The node set V contains: Physical intersection: A point where two or more trails meet; Key facilities: scenic area entrance, parking lot exit, cable car station, main viewing platforms, etc. Monitoring unit boundary points: To facilitate traceability, the boundary of each monitoring unit can also be abstracted as a node.

[0065] The edge set E represents the path segment connecting adjacent nodes, and each edge e ij ∈E stores the following attributes: Starting point v i And the endpoint v; trail length L ij Unit: m; Average width W ij Unit: m; Traffic direction restrictions; Historical average traffic speed .

[0066] Within the target scenic area, the core viewing platform is located at node v. r Nearby, there are three connecting edges: one leading south, one east, and one north. Based on the real-time flow direction φ=25°, it can be determined that the tourists mainly come from the south or east. However, careful analysis is needed: the flow direction is the direction the tourists are moving, not the direction of origin. In the risk area, we observe tourists moving in a 25° direction, meaning they are heading in that direction, indicating they entered the area from the opposite direction. Therefore, the direction of origin should be the opposite of the flow direction. Let the direction of origin angle φ be... src =φ + 180°, i.e., φ src =205°. This direction roughly points southwest, corresponding to the connecting node v. r In the edges, those pointing towards the source direction should be those that point from other nodes to v. r And the direction is the same as φ src The matched edges. Specifically, this can be determined by the cosine of the angle between the edge's direction vector and the source direction.

[0067] Reverse tracing of passenger flow origins: Based on the above analysis, risk area node v is defined. r Candidate source edge set E in (v r )={e ir∈E}, that is, all directed edges terminating at v. For each candidate edge e ir Its geometric direction vector d ir It can start from the starting point v i Pointing to destination v. Actual passenger flow from v. i Flow to v r Its direction should satisfy the real-time source direction φ src They are basically the same. Therefore, the direction matching degree is calculated: ; Where u src =(sinφ,cosφ src ) represents the unit vector in the direction of origin, assuming true north is 0°, and increasing clockwise. m ir Range of values ​​[ [1,1], the closer to 1, the more consistent the directions. Set a threshold m=0.7, and only retain m. ir >m th The edges are used as valid source edges.

[0068] In the target scenic area, the following calculations were performed: Side e sr The direction vector is approximately (0, 1), and the source direction u src =(sin205°,cos205°)≈( 0.4226, 0.9063), the dot product is 0×( 0.4226)+( 1)×( (0.9063) = 0.9063, matching degree m sr =0.9063, greater than the threshold; edge e er The direction is ( The dot product of (1,0) is ( 1)×( 0.4226)+0×( (0.9063) = 0.4226, matching degree 0.4226, less than the threshold, excluded; edge e nr The direction is (0,1), and the dot product is negative, so it is obviously a mismatch.

[0069] Therefore, the main source direction is determined to be edge e. sr That is, visitors enter the viewing platform from the south trail.

[0070] Next, with node v s Starting anew, continue the reverse tracing. At node v s Also looking for v sThe edge is the endpoint, and combined with real-time or historical passenger flow data on that edge, the previous level of origin is determined. This process is repeated until the scenic area entrance or boundary node is reached, forming a complete passenger flow origin path P={v r ,v s ,v s2 ,…,v entry}. In the target scenic area, from v s Continuing to trace backwards, we obtain node v. s2 v s3 Finally, we arrived at the south entrance of the scenic area. entry The path length is L path =350m.

[0071] Key Node Identification and Information Feedback: On the obtained source path P, all nodes that may affect passenger flow control are considered key nodes, including: Fork in the road is the best place to divert traffic; the source node is the entrance for visitors to enter the scenic area; special facilities are places where a large number of visitors usually converge.

[0072] Extract the coordinates p of these key nodes k =(x k ,y k This data, along with the path information P, is encapsulated into a source tracing result data packet. Simultaneously, the contribution weight of each path to congestion can be estimated based on the passenger flow volume on each edge. Passenger flow volume q ij The density ρ and velocity v can be calculated from the real-time data collected by the drone: q ij =ρ·v ij ·W ij ; Where W represents the width of the walkway. If multiple paths lead to the risk area, the primary source path is determined based on the passenger flow volume of each path, and guidance resources are deployed to it first.

[0073] Within the target scenic area, reverse tracing confirmed the primary source routes as the South Gate entrance, the parking lot intersection, and the viewing platform. Key nodes along this route included the South Gate entrance (coordinates (x1, y1) and the parking lot intersection (coordinates (x2, y2)). The real-time passenger flow along this route was estimated by the drone to be q = 1.2 people / s, significantly higher than other potential routes, thus identifying it as the primary intervention target. The ground control system transmitted the coordinates of these key nodes and route information to the self-organizing scheduling module in step S6 via a self-organizing network, providing precise data for subsequent drone flight path planning.

[0074] S6. Collaborative scheduling and planning: Based on the returned traceability information and instructions, adjust the deployment and flight path of the UAV and avoid collisions, plan and guide the dedicated flight path of the UAV, and send out the scheduling parameters. The guiding UAV is equipped with a guidance projection device. Furthermore, the collaborative scheduling plan specifically includes the following process: Based on the returned passenger flow source path information, key node coordinates and early warning instructions; Increase the number of drones deployed at key nodes along passenger flow origin routes and increase the frequency of drone patrols in key node areas; The patrol range of drones in risk areas is adjusted according to the warning level. Level 1 warnings reduce the patrol range, while Level 2 and above warnings focus on the core risk areas. By sharing drone location and flight path data through a self-organizing network, a drone collision avoidance algorithm combining distance priority and speed adjustment is adopted. To guide drones in planning dedicated routes along key nodes of passenger flow origin paths, and to plan return and resupply routes based on the drones' battery levels.

[0075] Specifically, after receiving the passenger flow origin and path information, key node coordinates, and early warning instructions transmitted in step S5, the ground control system dynamically optimizes the deployment of the drone swarm, ensuring that monitoring resources are tilted towards key areas and guiding the drones to plan dedicated routes for precise intervention. Taking the triggering of a level-two early warning at the core viewing platform of the target scenic area as an example, its key nodes include the south gate entrance p1, the parking lot intersection p2, and the risk area p. r The source routes are the south gate entrance, the parking lot intersection, and the viewing platform.

[0076] Optimization of drone deployment at key nodes: Passenger flow q based on key nodes k Define deployment priority index , where λ k Node type weights are assigned, with entrances receiving 1.2, intersections 1.0, and risk areas 1.5. Additional drones are allocated from the idle drone pool based on priority index to the vicinity of key nodes, ensuring at least one drone is permanently stationed at each key node for monitoring, and the patrol frequency is increased from the default every 30 seconds to every 10 seconds. Within the target scenic area, one drone is allocated from a low-risk area to the south entrance to collaborate with the existing drone in covering the entrance and intersections.

[0077] Risk area patrol range adjustment: The patrol range of drones in risk areas is adaptively adjusted according to the warning level: for Level 1 warnings, the range is reduced to within a 50-meter radius of the risk point; for Level 2 and above warnings, it is further reduced to within a 30-meter radius, and the flight speed is reduced to increase the data collection frequency. In the target scenic area, the flight path of the two drones at the core viewing platform is changed from boundary patrol to circling around point p. rIt flies in a circle with a radius of 25 meters, and its speed decreases from 5 m / s to 3 m / s.

[0078] Drone collision avoidance planning: sharing real-time drone locations via a self-organizing network. i (t), velocity v i (t) and flight path. Each drone calculates its distance from neighboring drones. When d ij (t) <d safe Obstacle avoidance is triggered on time. A speed adjustment strategy is adopted: ; Where α=0.3 is the deceleration coefficient. In the target scenic area, when UAV-2-1 and UAV-3 are 15 meters apart, the speed of UAV-2-1 drops from 3m / s to 2.1m / s, and accelerates after the distance is restored.

[0079] Guided drone route planning: Based on passenger flow origin path P={p r The system plans a dedicated flight path for the guidance drone equipped with a guidance projection device, passing over each key node sequentially and hovering to project dynamic guidance markers. The flight path is generated using a shortest path algorithm, and the node order is consistent with the source path. This is combined with the real-time battery level e of the guidance drone. u Planning the return resupply route: When e u Below the autonomous return threshold e th When resupplying, insert the nearest mobile supply station as a transit point in the flight path, and automatically return to the original flight path after resupply is completed. In the target scenic area, guide the UAV-G1 drone to fly along the route of "viewing platform → parking lot intersection → south gate entrance", projecting a light strip pointing towards the risk area at each node; when the battery drops to 25%, go to the south gate supply station to recharge, and continue to perform guidance after recharging.

[0080] S7. Coordinated Response Guidance: Based on the issued early warning level instructions, initiate corresponding response operations to guide the drone to project dynamic guidance markers at key nodes along the flight path, with response and projection executed simultaneously; Furthermore, the coordinated response guidance specifically includes the following process: Based on the issued warning level instructions, the corresponding response operations will be automatically initiated; Level 1 alerts trigger voice broadcasts to provide passenger flow information; Level 2 alerts trigger voice broadcasts and warning lights to broadcast flow restriction notices and project warning lights; Level 3 alerts trigger voice broadcasts, warning lights, temporary fencing, and location markers, including the deployment of temporary fencing and location of the congestion source. The drones are guided to fly along a dedicated route, and dynamic guidance signs are projected at key nodes along the passenger flow path. The content of the guidance signs is dynamically adjusted based on the real-time passenger flow data, so that the handling operations and dynamic guidance are executed simultaneously.

[0081] Specifically, after receiving the warning level instruction issued in step S4 and the dedicated flight path for guiding the UAV planned in step S6, the ground control system simultaneously initiates the corresponding level of handling operations and directs the UAV to project dynamic guidance markers at key nodes along the flight path, thereby achieving coordinated operation between aerial guidance and ground handling.

[0082] Tiered response procedures are initiated: The ground control system automatically triggers corresponding response procedures based on the warning level: For a Level 1 warning, the voice broadcast module is activated to play crowd flow alerts, such as "Crowds are dense ahead, please maintain order"; for a Level 2 warning, in addition to upgrading the voice broadcast to a flow restriction notice, the warning light module is activated to project flashing red light spots to mark the boundaries of the risk area; for a Level 3 warning, based on the Level 2 warning, a temporary fence deployment and positioning marker module is added, deploying flexible isolation strips to form a physical fence and projecting fluorescent markers to accurately locate the source of congestion. Taking a Level 2 warning triggered at the core viewing platform of the target scenic area as an example, the drone over the viewing platform immediately broadcasts "Crowds are dense at the viewing platform, it is recommended to postpone your visit" and projects red light spots to cover the entrance area.

[0083] Dynamic guidance drone projection: Guide drones equipped with projection equipment fly sequentially over key nodes along the passenger flow origin path, such as parking lot intersections and the south entrance, hovering at each node to project dynamic guidance signs. The projected content is a warm-toned light band flowing in the desired diversion direction, with a light band flow speed v. light Based on the real-time passenger flow density ρ at the node node Dynamic adjustment: ; Where v base ρ is the reference flow velocity, β is the adjustment coefficient, and ρ is the flow velocity. crit This represents the critical density. As the node density increases, the light band's flow velocity accelerates to enhance the sense of urgency in the splitting process, while the brightness increases proportionally. For example, at a parking lot intersection, if ρ = 0.3 people / m² is detected, v is calculated. light The light band moves at approximately 1.19 m / s, guiding the flow towards secondary attractions.

[0084] Coordinated Execution of Handling and Guidance: GPS timing enables microsecond-level synchronization of all devices, ensuring simultaneous activation of voice broadcasts, warning lights, and guidance projections. Voice broadcasts and guidance signage are aligned to avoid information conflicts. If the guidance drone detects poor passenger flow response at a certain node, it immediately relays this information to the ground control system to adjust guidance strategies for subsequent nodes. At the target scenic area, after the light strip was projected at the parking lot intersection, passenger flow in that direction increased from 0.9 people / s to 1.3 people / s, curbing the upward trend in density at the core viewing platform, and the Level 2 warning remained unchanged.

[0085] S8. Effect Feedback Optimization: After the data is collected and processed, the parameters are summarized to the ground control system. The ground control system will feed back the parameters to update the model, optimize the rules, and adjust the plan. Furthermore, the effect feedback optimization specifically includes the following process: The system collects and processes passenger flow density, flow velocity, and flow direction parameters, which are then aggregated to the ground control system. The processed parameters are fed back to update the total energy constant and various critical values ​​of the passenger flow fluid flow model. The processed parameters are also fed back to optimize the identification rules for key nodes in the passenger flow source path. Finally, the processed parameters are fed back to adjust the deployment and patrol plans of UAVs.

[0086] Specifically, after the handling guidance is implemented, the ground control system continuously collects passenger flow parameters and feeds them back to each processing stage. This enables adaptive updates of model parameters, dynamic optimization of recognition rules, and continuous improvement of UAV deployment schemes, forming a closed-loop learning mechanism. Taking the Level II early warning handling of the core viewing platform of the target scenic area as an example, passenger flow data collected within 10 minutes after the handling is used for subsequent optimization.

[0087] Post-processing passenger flow parameter collection: Each monitoring unit's drones continue to collect passenger flow images according to step S3, and calculate the post-processing passenger flow density ρ. post Flow velocity v post and flow direction φ post The data are then aggregated to the ground control system after being processed by a moving average. For example, after the treatment at the core observation deck, the average density decreased from 0.75 people / m² to 0.68 people / m², and the velocity increased from 0.4 m / s to 0.55 m / s.

[0088] Passenger flow fluid flow model parameter update: The processed parameters are fed back to the fluid-based passenger flow prediction module to update the total energy constant E0 and density critical value ρ in the model. crit Based on Bernoulli's equation, the energy constant E is calculated using cross-sectional data before and after treatment. k And updated using a weighted average: ; Where γ = 0.2 is the learning rate. Simultaneously, ρ is fine-tuned based on congestion triggering conditions during processing. crit For example, the critical value for the core observation deck was revised from 0.80 people / m² to 0.78 people / m².

[0089] Optimization of key node identification rules for passenger flow origin paths: Feedback the processed parameters to the passenger flow tracing module to optimize the node type weight λ. k Define the bootstrap responsiveness. ,in , These represent the passenger flow at each node before and after the intervention. The node weight update formula is: ; Where δ=0.1 is the adjustment step size. For example, the response η2 of a parking lot intersection is 0.44, and its weight is increased from 1.0 to 1.04.

[0090] Adjustments to the drone deployment and patrol plan: Feedback of post-processing parameters to the self-organizing scheduling module to recalculate the coverage demand index D for each monitoring unit. i The allocation of drones is dynamically adjusted. At the same time, the patrol frequency is optimized based on the monitoring blind spots discovered during the operation. For example, the default patrol frequency at the intersection of the parking lot is adjusted from 30 seconds / time to 20 seconds / time, and one drone from the core observation deck is moved to a new risk area to achieve dynamic resource balance.

[0091] S9. Closed-loop termination: Repeat the closed-loop operation from S3 to S8. When the passenger flow density is lower than the first critical interval and the passenger flow dispersion is greater than or equal to 0 for a preset time, the command is terminated, the drone resumes normal mode, and a report is generated at the end of the holiday.

[0092] Specifically, the system automatically repeats steps S3 to S8, forming a continuous monitoring-prediction-response-optimization closed loop. Within each cycle, the drone swarm collects passenger flow parameters in real time, and the ground control system dynamically performs prediction, source tracing, scheduling, response, and feedback optimization. When the passenger flow density ρ in the risk area is lower than the lower limit ρ of the first critical interval... crit1 And passenger flow dispersion ·(ρv)≥0 and continues for a preset time T hold Upon determining that passenger flow has returned to normal and there is no risk of overcrowding, the ground control system immediately issues instructions to all drones to terminate early warning and handling operations: stop voice broadcasts, warning lights, and dynamic signage projection; guide drones to return to base or switch to regular monitoring tasks; and have each monitoring unit's drones resume the initial cruise routes planned in step S2. After the entire holiday period ends, the system automatically summarizes all operational data, generating a scenic area passenger flow monitoring and handling report that includes passenger flow time-series data, early warning event records, guidance effect evaluation, model parameter update history, and drone operation logs, providing data support for subsequent management.

[0093] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring, early warning, and handling drone-based surges in tourist traffic during holidays in scenic areas, characterized in that: Includes the following steps: S1. System initialization: Collect scenic area geographic data to divide monitoring units, calibrate drone and visitor flow benchmark parameters, and configure ground control system communication. S2. Drone Deployment: The drone swarm establishes a self-organizing network, allocates drones according to monitoring units to plan initial routes, detects battery power and responds to return-to-base resupply requests; S3. Passenger flow data collection: Each UAV collects passenger flow images, extracts passenger flow density, flow speed, and flow direction parameters, completes fluid parameter mapping, and summarizes the parameters to the ground control system; S4. Passenger Flow Forecast and Early Warning: Based on the aggregated passenger flow parameters, a model is constructed by combining the continuity equation and Bernoulli equation to calculate the risk assessment and early warning level, and the forecast results are then issued. S5. Passenger Flow Source Identification: Based on the issued early warning instructions and passenger flow parameters, retrieve the electronic map to trace the source path, identify key nodes, and transmit the results back to the ground control system. S6. Collaborative scheduling and planning: Based on the returned traceability information and instructions, adjust the deployment and flight path of the UAV and avoid collisions, plan and guide the dedicated flight path of the UAV, and send out the scheduling parameters. The guiding UAV is equipped with a guidance projection device. S7. Coordinated Response Guidance: Based on the issued early warning level instructions, initiate corresponding response operations to guide the drone to project dynamic guidance markers at key nodes along the flight path, with response and projection executed simultaneously; S8. Effect Feedback Optimization: After the data is collected and processed, the parameters are summarized to the ground control system. The ground control system will feed back the parameters to update the model, optimize the rules, and adjust the plan. S9. Closed-loop termination: Repeat the closed-loop operation from S3 to S8. When the passenger flow density is lower than the first critical interval and the passenger flow dispersion is greater than or equal to 0 for a preset time, the command is terminated, the drone resumes normal mode, and a report is generated at the end of the holiday.

2. The method for monitoring, early warning, and handling of sudden surges in tourist traffic during holidays in scenic areas, as described in claim 1, is characterized in that... In step S1, the system initialization specifically includes the following process: Collect data on terrain features, regional carrying capacity thresholds, and trail distribution in various areas of the scenic area, and divide the scenic area into N monitoring units according to terrain features and points prone to congestion. The system calibrates the cruising radius, endurance, autonomous return threshold battery level, and response speed of a single drone, as well as the benchmark values ​​for passenger flow density, flow speed, and flow direction. Configure communication links and data protocols between the ground control system and each processing stage to ensure real-time transmission of data and commands between stages.

3. The method for monitoring, early warning, and handling drone-based surges in tourist traffic during holidays, as described in claim 1, is characterized in that... In step S2, the deployment of the drone specifically includes the following process: Once the drone swarm is started, it automatically establishes a distributed self-organizing network without a central node, sharing the location, flight path, and battery data of each drone in real time. Based on the results of the monitoring unit division, allocate 1-2 drones to each monitoring unit and plan the initial cruise routes along the boundaries of the monitoring units and points prone to passenger congestion. The drone monitors its own battery level in real time. When the battery level is lower than the autonomous return-to-home threshold, it sends a return-to-home request, matches the nearest drone mobile resupply station, and plans a return-to-home resupply route.

4. The method for monitoring, early warning, and handling drone surges in tourist traffic during holidays, as described in claim 1, is characterized in that... In step S3, the passenger flow data collection specifically includes the following process: Each drone uses its onboard high-definition camera to collect real-time images of passenger flow from its monitoring unit, and uses image recognition algorithms to extract the number of tourists and their displacement data from a single frame image. The flow density is calculated based on the number of tourists and the actual area of ​​the scenic spot corresponding to the image. The flow velocity is calculated based on the displacement distance of tourists in two consecutive frames of images and the time interval between image acquisition. The flow direction is calculated based on the horizontal and vertical components of the tourist displacement. The calculated passenger flow density, flow speed, and flow direction parameters are aggregated in real time to the ground control system via a self-organizing network, and the ground control system distributes the parameters to subsequent processing stages.

5. The method for monitoring, early warning, and handling drone surges in tourist traffic during holidays, as described in claim 1, is characterized in that... In step S4, the construction of the passenger flow fluid flow model specifically includes the following process: A three-dimensional continuity equation is established based on the law of conservation of mass, and then simplified into a one-dimensional continuity equation by combining the topographic features of the bottleneck area of ​​the scenic spot. Based on the law of conservation of energy and combined with the topographic slope of the scenic area and the regional carrying capacity threshold, the Bernoulli equation is established, and the critical value of passenger flow density in the bottleneck area is derived through the continuity equation and the Bernoulli equation. By combining the passenger flow parameters collected in real time by UAVs to dynamically correct the model parameters, the passenger flow density time change rate, passenger flow dispersion and passenger flow pressure value are calculated through the passenger flow fluid flow model, and the warning level is determined based on the calculation results.

6. The method for monitoring, early warning, and handling drone surges in tourist traffic during holidays, as described in claim 1, is characterized in that... In step S4, the early warning level determination and instruction issuance specifically include the following process: When the passenger flow density in the bottleneck area is in the first critical range and the flow speed continues to decrease, a Level 1 warning is triggered. A level-two warning is triggered when the passenger flow density is in the second critical range and the passenger flow dispersion is less than 0. A Level 3 warning is triggered when the passenger flow density is greater than or equal to the third critical interval or the passenger flow pressure value is greater than or equal to the congestion critical pressure value. The warning level, risk area, predicted time and passenger flow parameters are aggregated and sent to the ground control system, which then issues warning instructions simultaneously.

7. The method for monitoring, early warning, and handling drone surges in tourist traffic during holidays, as described in claim 1, is characterized in that... In step S5, the passenger flow tracing and identification specifically includes the following process: Based on the issued early warning instructions, passenger flow direction and flow speed data, retrieve the scenic area's electronic map; Based on passenger flow direction and speed, the passenger flow origin path algorithm is used to trace the passenger flow origin path of risk areas and form a set of passenger flow origin paths; Identify key nodes such as entrance diversion points, pedestrian intersections, and parking lot exits in the passenger flow origin path, extract the coordinates of key nodes, and transmit the passenger flow origin path information and key node coordinates back to the ground control system.

8. The method for monitoring, early warning, and handling surges in tourist traffic during holidays in scenic areas, as described in claim 1, is characterized in that... In step S6, the collaborative scheduling plan specifically includes the following process: Based on the returned passenger flow source path information, key node coordinates and early warning instructions; Increase the number of drones deployed at key nodes along passenger flow origin routes and increase the frequency of drone patrols in key node areas; The patrol range of drones in risk areas is adjusted according to the warning level. Level 1 warnings reduce the patrol range, while Level 2 and above warnings focus on the core risk areas. By sharing drone location and flight path data through a self-organizing network, a drone collision avoidance algorithm combining distance priority and speed adjustment is adopted. To guide drones in planning dedicated routes along key nodes of passenger flow origin paths, and to plan return and resupply routes based on the drones' battery levels.

9. The method for monitoring, early warning, and handling of sudden surges in tourist traffic during holidays in scenic areas, as described in claim 1, is characterized in that... In step S7, the coordinated handling guidance specifically includes the following process: Based on the issued warning level instructions, the corresponding response operations will be automatically initiated; Level 1 alerts trigger voice broadcasts to provide passenger flow information; Level 2 alerts trigger voice broadcasts and warning lights to broadcast flow restriction notices and project warning lights; Level 3 alerts trigger voice broadcasts, warning lights, temporary fencing, and location markers, including the deployment of temporary fencing and location of the congestion source. The drones are guided to fly along a dedicated route, and dynamic guidance signs are projected at key nodes along the passenger flow path. The content of the guidance signs is dynamically adjusted based on the real-time passenger flow data, so that the handling operations and dynamic guidance are executed simultaneously.

10. The method for monitoring, early warning, and handling drone surges in tourist traffic during holidays, as described in claim 1, is characterized in that... In step S8, the effect feedback optimization specifically includes the following process: Collect and process passenger flow density, flow velocity, and flow direction parameters, and summarize them to the ground control system; feed back the processed parameters to update the total energy constant and various critical values ​​of the passenger flow fluid flow model; The processed parameters will be fed back to optimize the rules for identifying key nodes in the passenger flow source path; the processed parameters will also be fed back to adjust the drone deployment and patrol plan.

Images