A method for preventing and warning collision on apron

Abstract

The application discloses a kind of apron anti-collision early warning methods, the positioning data of vehicle and aircraft is first converted into H3 geographic grid index, and candidate conflict target pair set is screened out based on this by grid neighborhood query algorithm.Subsequently, the real-time geometric distance between each pair of target in the candidate conflict target pair set is calculated to generate real-time collision warning.Then, the target in the candidate conflict target pair set is future trajectory prediction, respectively get the first prediction trajectory sequence based on motion state estimation and the second prediction trajectory sequence based on vehicle operation planning route deduction.Afterwards, the first future distance sequence and the second future distance sequence corresponding to two prediction trajectory sequences are calculated respectively.Finally, according to the two future distance sequences, the collision risk level is determined by risk assessment model, and the corresponding graded warning information is output.The application improves the real-time, accuracy and intelligent level of apron anti-collision early warning.

Description

Technical Field

[0001] This invention relates to the field of aviation ground safety technology, and in particular to a collision avoidance and early warning method for aprons. Background Technology

[0002] Airport aprons are high-dynamic, high-risk areas where aircraft and ground support vehicles (such as guidance vehicles, refueling vehicles, and baggage towing vehicles) operate together. Traditional safety management relies heavily on the visual observation of drivers and verbal instructions from air traffic controllers, which has inherent defects such as human fatigue, blind spots, and communication delays, making it difficult to cope with the real-time collision avoidance requirements in complex operating environments.

[0003] To improve operational safety, automated collision avoidance schemes based on positioning data have emerged in existing technologies. These schemes typically receive real-time positioning data from vehicles (e.g., GNSS-based) and aircraft (e.g., ADS-B-based or surface surveillance radar), calculate the real-time Euclidean distance between them, and make a binary judgment (safe / dangerous) based on a single safety threshold, thereby triggering an alarm when the distance is too close.

[0004] However, such existing solutions have significant technical limitations:

[0005] The computational efficiency is low, making it difficult to meet real-time requirements: When the number of targets on the tarmac increases, a "brute-force" approach of comparing all targets pairwise is typically used, resulting in a quadratic increase in computational complexity (O(N²)). This leads to a significant increase in processing latency in high-density target scenarios, making it impossible to guarantee millisecond-level concurrent detection of hundreds of dynamic targets.

[0006] The early warning system is passive and delayed, lacking foresight: it relies solely on the static position at the current moment for collision assessment, making it a "post-event" or "instantaneous" warning. It cannot predict potential collisions between vehicles and aircraft due to continuous movement in the coming seconds to tens of seconds, leaving operators with severely insufficient reaction time.

[0007] The prediction model is simplistic and fails to fully utilize prior operational knowledge: While some improvement schemes introduce motion prediction based on simple linear extrapolation, they fail to effectively leverage the key operational characteristic of ground vehicles having pre-defined work routes in tasks such as guidance and inspection. The prediction model is disconnected from the actual operational plan, resulting in insufficient accuracy and practicality in long-term trajectory prediction.

[0008] The risk indications are vague and lack tiered decision support: the alarm output is only a simple "yes / no" binary judgment, which cannot quantify the urgency of the risk. This makes it impossible for drivers to take differentiated measures according to the level of risk, which can easily lead to complacency with frequent low-risk alarms or insufficient response to high-risk situations, thus limiting the guidance value of the alarms.

[0009] Therefore, there is an urgent need for a comprehensive technical solution that can simultaneously address the issues of computational efficiency, proactive early warning, accurate prediction, and risk quantification, in order to build a real-time, proactive, and intelligent apron collision avoidance early warning system. Summary of the Invention

[0010] To address the technical problems existing in the background art, this invention proposes a collision avoidance early warning method for aprons.

[0011] The present invention proposes a method for apron collision avoidance early warning, comprising the following steps:

[0012] S1. Within the airport apron area, based on the H3 geographic grid index of vehicles and aircraft, a bidirectional neighborhood search is performed using a grid neighborhood query algorithm to filter out vehicle and aircraft pairs that fall within each other's neighborhood range, thus obtaining a set of candidate conflict target pairs.

[0013] S2. Calculate the real-time geometric distance between each vehicle and aircraft pair in the candidate conflict target pair set. If the real-time geometric distance is less than the safety threshold, generate a real-time collision alarm.

[0014] S3. For each vehicle and aircraft pair in the candidate conflict target pair set, perform future trajectory prediction to obtain the first predicted trajectory sequence and the second predicted trajectory sequence corresponding to the vehicle and aircraft pair; wherein, the first predicted trajectory sequence is obtained based on motion state estimation, the second predicted trajectory sequence is obtained based on the vehicle's predefined operation planning route and its current state deduction, and the future time period covered by the second predicted trajectory sequence is longer than the future time period covered by the first predicted trajectory sequence.

[0015] S4. Calculate the predicted distance between the predicted position of the vehicle at each future time and the predicted position of the aircraft at the corresponding time in the first predicted trajectory sequence and the second predicted trajectory sequence respectively, to obtain the corresponding first future distance sequence and second future distance sequence.

[0016] S5. Based on the first future distance sequence and the second future distance sequence, determine the collision risk level through the risk assessment model, and output graded early warning information according to the risk level.

[0017] Preferably, the H3 geographic grid index of the vehicle and aircraft is obtained through the following steps: receiving vehicle positioning data and aircraft positioning data; converting the geographic coordinates in the vehicle positioning data and aircraft positioning data into the corresponding H3 geographic grid index; the grid neighborhood query algorithm is the kRing neighborhood query algorithm of the H3 library, and the neighborhood range is determined by converting the preset physical security monitoring radius into the H3 grid topology distance k.

[0018] Preferably, in step S3, the process of obtaining the first predicted trajectory sequence based on motion state estimation specifically includes:

[0019] A motion state model is established for each vehicle and aircraft pair in the candidate conflict target pair set. The state vector of the motion state model includes at least the position coordinates and velocity components of the target on the two-dimensional plane.

[0020] Based on the continuous vehicle positioning data and continuous aircraft positioning data corresponding to each vehicle and aircraft pair, the state vector is recursively estimated and updated using the Kalman filter algorithm to obtain the optimal state estimate of the vehicle and aircraft pair at the current moment.

[0021] Based on the optimal state estimation and the preset state transition model, the state of the vehicle and aircraft at each discrete time point in the first future time period is recursively predicted, thereby generating the first predicted trajectory sequence.

[0022] Preferably, the state transition model is a uniform motion model; the future time period covered by the first predicted trajectory sequence is 5 to 15 seconds.

[0023] Preferably, the duration of the second future time period covered by the second predicted trajectory sequence is 20 to 60 seconds; in step S3, the process of obtaining the second predicted trajectory sequence based on the vehicle's predefined operation planning route and its current state specifically includes:

[0024] For each vehicle and aircraft pair in the candidate conflict target pair set, obtain the predefined operation planning route for the vehicle in that vehicle and aircraft pair, the operation planning route being defined by a series of ordered geographical locations;

[0025] Based on the vehicle's current geographical location, determine its projected position on the planned operation route and the length of the path already traveled;

[0026] Based on the vehicle's current speed and the preset prediction step size, calculate its cumulative predicted travel distance at each prediction time in the future second time period.

[0027] Based on the cumulative predicted travel distance, the corresponding future geographical locations are determined by interpolation calculation on the planned operation route, and then connected in chronological order to form the second predicted trajectory sequence.

[0028] Preferably, the predicted distance is specifically Euclidean distance; the generation process of the first future distance sequence specifically includes:

[0029] Determine the first prediction period and the first prediction time step covered by the first predicted trajectory sequence;

[0030] According to the first prediction time step, iterate through each first future moment within the first prediction period;

[0031] For each first future moment, the predicted position coordinates of the vehicle at that first future moment are extracted from the first predicted trajectory sequence, and the predicted position coordinates of the aircraft at the corresponding first future moment are obtained based on the motion state estimation.

[0032] Calculate the Euclidean distance between the vehicle's predicted position coordinates and the aircraft's predicted position coordinates.

[0033] The Euclidean distances calculated for all first future moments are collected in chronological order to form the first future distance sequence.

[0034] Preferably, the predicted distance is specifically Euclidean distance; the generation process of the second future distance sequence specifically includes:

[0035] Determine the second prediction period and the second prediction time step covered by the second prediction trajectory sequence;

[0036] According to the second prediction time step, iterate through each second future moment within the second prediction period;

[0037] For each second future time, the predicted position coordinates of the vehicle at that second future time are extracted from the second predicted trajectory sequence, and the predicted position coordinates of the aircraft at the corresponding second future time are obtained based on the motion state estimation.

[0038] Calculate the Euclidean distance between the vehicle's predicted position coordinates and the aircraft's predicted position coordinates.

[0039] The Euclidean distances calculated at all second future moments are collected in chronological order to form the second future distance sequence.

[0040] Preferably, the step of determining the collision risk level and outputting graded early warning information through a risk assessment model specifically includes:

[0041] For the first future distance sequence, calculate its corresponding first predicted collision time, the closest approach distance between the first trajectories, the first relative velocity, and the first approach angle;

[0042] For the second future distance sequence, calculate its corresponding second first predicted collision time, the closest approach distance between the second trajectories, the second relative velocity, and the second approach angle;

[0043] Based on the first predicted collision time, the closest approach distance between the first trajectories, the first relative velocity, and the first approach angle, the first risk level is obtained through a risk assessment model.

[0044] Based on the second initial collision time, the closest approach distance between the second trajectories, the second relative velocity, and the second approach angle, the second risk level is obtained through a risk assessment model;

[0045] The first and second risk levels are combined to determine a final comprehensive risk level.

[0046] Based on the comprehensive risk level, the system matches the corresponding warning content and format from a predefined warning strategy library to generate and output tiered warning information.

[0047] Preferably, the step of comprehensively processing the first risk level and the second risk level to determine a final comprehensive risk level specifically includes:

[0048] Based on the preset risk level ranking, the lower-ranked level in the first risk level and the second risk level is determined as the comprehensive risk level.

[0049] The types of the first risk level, the second risk level, and the comprehensive risk level all include safety level, alert level, warning level, danger level, and emergency level; the risk levels are ordered as follows: safety level, alert level, warning level, danger level, and emergency level, with the emergency level being the last and the safety level being the first.

[0050] Preferably, in step S2, the calculation of the real-time geometric distance includes any of the following methods:

[0051] Calculate the planar Euclidean distance between the center points of the vehicle and the aircraft; or,

[0052] Construct outer envelope geometric models for vehicles and aircraft respectively, and calculate the shortest spatial distance between the two outer envelope geometric models.

[0053] The proposed apron collision avoidance early warning method significantly improves the computational efficiency of multi-target concurrent detection by introducing a hierarchical spatial grid index, breaking through the limitations of traditional real-time collision detection. By integrating short-term prediction based on motion state estimation and long-term prediction based on operation planning routes, it achieves a fundamental shift from passive alarm to proactive early warning. Furthermore, by constructing a quantitative risk assessment model that integrates multiple factors and a hierarchical early warning strategy, it effectively solves the problems of alarm ambiguity and insufficient decision support caused by traditional binary judgment. Thus, it constructs a real-time, proactive, accurate, and intelligent apron collision avoidance early warning technology system, comprehensively improving the safety level and decision-making intelligence of apron operations. Attached Figure Description

[0054] Figure 1 This is a schematic diagram illustrating the workflow of an apron collision avoidance early warning method proposed in this invention. Detailed Implementation

[0055] Reference Figure 1 The present invention proposes a method for apron collision avoidance early warning, comprising the following steps:

[0056] S1. Within the airport apron area, based on the H3 geographic grid index of vehicles and aircraft, a bidirectional neighborhood search is performed using a grid neighborhood query algorithm to filter out vehicle and aircraft pairs that fall within each other's neighborhood range, thus obtaining a set of candidate conflict target pairs.

[0057] In this embodiment, the H3 geographic grid index of vehicles and aircraft is obtained through the following steps:

[0058] Receive vehicle positioning data and aircraft positioning data;

[0059] The geographic coordinates in vehicle and aircraft location data are converted into corresponding H3 geographic grid indexes. The grid neighborhood query algorithm is the kRing neighborhood query algorithm from the H3 library, and the neighborhood range is determined by converting the preset physical security monitoring radius into an H3 grid topological distance k.

[0060] It should be noted that the system receives vehicle positioning data in real time from the vehicle-mounted GNSS (Global Navigation Satellite System) terminal, as well as aircraft positioning data from ADS-B (Automatic Dependent Surveillance-Broadcast) signals or surface surveillance radar, via a dedicated data interface. Simultaneously, it receives vehicle operation planning route data from the airport A-CDM (Airport Collaborative Decision Making) system or the ground vehicle dispatching system. This vehicle operation planning route data is typically defined as a series of ordered latitude and longitude coordinates.

[0061] Specifically, the received vehicle and aircraft positioning data are first cleaned, including removing positioning jumps caused by signal interference and correcting timestamp anomalies. Then, the UberH3 open-source geoindexing library is called, and the geoToH3 function is used to convert the current latitude and longitude coordinates (lat, lon) of each vehicle and aircraft into an H3 geogrid index h of the corresponding resolution.

[0062] The conversion function is: h = GeoToH3(lat,lon,r);

[0063] Here, 'r' represents the resolution, which can be set according to the monitoring granularity requirements of different areas of the airport. The higher the resolution, the smaller and finer the grid. For example, a higher resolution (e.g., r=12, grid side length approximately 9 meters) is used in critical areas such as runways and taxiways, while an even higher resolution (e.g., r=13, grid side length approximately 3.5 meters) is used in ordinary aprons. After the above processing is completed, the system maintains a dynamic "H3 grid index - target list" hash mapping table in memory.

[0064] When collision detection is required for a specific area or all dynamic targets, the system enters a rapid filtering phase. For each vehicle or aircraft currently online (referred to as the "center target"), the system uses its current H3 index. Centered on the target area, the distance to the H3 grid topology is calculated based on a preset physical security monitoring radius (e.g., 150 meters). Then, the kRing neighborhood query algorithm from the H3 library is invoked. This algorithm quickly returns a hexagonal grid index set S, which contains all grid points related to the center grid. A grid with hexagonal distances not greater than k.

[0065] It should be noted that the kRing neighborhood query algorithm is defined as follows: ;

[0066] in, The value represents the index of the central hexagon H3; k represents the mesh topological distance; the returned result S represents the set of the center point and the surrounding hexagonal mesh that meet the condition (distance ≤ k).

[0067] By querying the aforementioned "H3 grid index-target list" mapping table, the system can instantly retrieve all other dynamic targets (vehicles or aircraft) falling within the geographical area represented by set S. Through bidirectional neighborhood retrieval (i.e., simultaneously retrieving surrounding aircraft centered on vehicles and retrieving surrounding vehicles centered on aircraft), and taking the intersection, the system can efficiently and accurately filter out all vehicle-aircraft pairs falling within each other's neighborhoods, forming a candidate conflict target pair set. This strategy reduces the computational complexity from a global O(N²) to approximately O(N), significantly improving processing efficiency.

[0068] S2. Calculate the real-time geometric distance between each vehicle and aircraft pair in the candidate conflict target pair set. If the real-time geometric distance is less than the safety threshold, generate a real-time collision alarm.

[0069] In this embodiment, step S2, the calculation of real-time geometric distance includes any of the following methods:

[0070] Calculate the planar Euclidean distance between the center points of the vehicle and the aircraft; or,

[0071] Construct outer envelope geometric models for vehicles and aircraft respectively, and calculate the shortest spatial distance between the two outer envelope geometric models.

[0072] Specifically, for each vehicle and aircraft pair in the candidate conflict target pair set, real-time accurate detection is performed. First, simplified envelope geometry models are constructed for the vehicles and aircraft. For example, an aircraft is simplified to a convex polygon (or a conservative ellipse) bounded by its outline, and a vehicle is simplified to a rectangle.

[0073] Subsequently, the real-time minimum spatial distance between the two geometric models is calculated. The calculation method can be to calculate the shortest side-to-side distance between two geometric figures, or, to simplify the calculation, to calculate the planar Euclidean distance between their center points. The system has a preset safety threshold. (e.g., 10 meters). If the calculated value is... If an imminent collision risk is detected, the system immediately generates and sends a real-time collision warning with the highest priority. This warning is sent directly to the driver involved through an independent channel (such as a strong voice warning from the vehicle terminal), achieving a millisecond-level emergency response.

[0074] S3. For each vehicle and aircraft pair in the candidate conflict target pair set, perform future trajectory prediction to obtain the first predicted trajectory sequence and the second predicted trajectory sequence corresponding to the vehicle and aircraft pair; wherein, the first predicted trajectory sequence is obtained based on motion state estimation, the second predicted trajectory sequence is obtained based on the vehicle's predefined operation planning route and its current state deduction, and the future time period covered by the second predicted trajectory sequence is longer than the future time period covered by the first predicted trajectory sequence.

[0075] In this embodiment, step S3, the process of obtaining the first predicted trajectory sequence based on motion state estimation, specifically includes: establishing a motion state model for each vehicle and aircraft pair in the candidate conflict target pair set, wherein the state vector of the motion state model includes at least the position coordinates and velocity components of the target on a two-dimensional plane; based on the continuous positioning data of the vehicle and the continuous positioning data of the aircraft corresponding to each vehicle and aircraft pair, recursively estimating and updating the state vector through the Kalman filter algorithm to obtain the optimal state estimate of the vehicle and aircraft pair at the current moment; based on the optimal state estimate and the preset state transition model, recursively predicting the state of the vehicle and aircraft pair at each discrete time point in the future first time period, thereby generating the first predicted trajectory sequence.

[0076] In this embodiment, the state transition model is a uniform motion model; the future time period covered by the first predicted trajectory sequence is 5 to 15 seconds.

[0077] In this embodiment, the duration of the second predicted trajectory sequence covering the second future time period is 20 to 60 seconds. In step S3, the process of obtaining the second predicted trajectory sequence based on the vehicle's predefined operation planning route and its current state specifically includes: for each vehicle and aircraft pair in the candidate conflict target pair set, obtaining the predefined operation planning route of the vehicle in the vehicle and aircraft pair, the operation planning route being defined by a series of ordered geographical locations; determining the vehicle's projection position and the length of the path already traveled on the operation planning route based on the vehicle's current geographical location; calculating the cumulative predicted travel distance of the vehicle at each predicted moment in the second future time period based on the vehicle's current speed and a preset prediction step size; determining the corresponding future geographical locations on the operation planning route through interpolation calculation based on the cumulative predicted travel distance, and connecting them in chronological order to form the second predicted trajectory sequence.

[0078] In this embodiment, future trajectories of targets in the candidate conflict target set are predicted, and two modes are used in parallel:

[0079] Mode 1: Short-term motion state prediction based on Kalman filtering, generating the first predicted trajectory sequence; the specific processing includes:

[0080] Modeling: A uniform velocity (CV) motion model is established for each moving vehicle and aircraft. Its state vector X(k) is defined as follows: ,in For planar coordinates, For the velocity component.

[0081] Filtering estimation: Preprocessed vehicle and aircraft positioning data are used as observations and input into a Kalman filter. The Kalman filter continuously optimizes the estimation of the target's true state vector through a cyclical process of prediction and update, ultimately obtaining the optimal state estimate X(z|z) for the current time z.

[0082] State prediction: Based on the current optimal estimate X(z|z) and the preset state transition model (the state transition matrix F corresponding to the uniform velocity model), the state X(z+h|z)=F of the target is recursively predicted in the first time period in the future (e.g., the next 10 seconds) at every prediction time step Δt (e.g., 1 second). h • X(z|z). Extract the position components [x(z+h), y(z+h)] and concatenate them in chronological order to form the first predicted trajectory sequence of the target. This mode is suitable for high-precision short-term (e.g., 5-15 seconds) predictions.

[0083] Mode 2: Based on long-term behavior prediction of the task planning route, a second predicted trajectory sequence is generated; the specific processing includes:

[0084] Route Acquisition and Mapping: For vehicles performing tasks (such as guide vehicles), the system acquires their predefined operational planning routes. Based on the vehicle's current geographical location, it calculates its vertical projection point on the planned route and determines the length of the path it has already traveled. .

[0085] Path extrapolation: Based on the vehicle's current speed v, calculate the cumulative predicted travel distance at each predicted time t within the next second time period (e.g., the next 30 seconds). .

[0086] Spatial interpolation: based on On the job planning route defined by discrete points, the corresponding future geographical coordinates are calculated using linear interpolation or spline interpolation algorithms. Connecting these points in chronological order forms the vehicle's second predicted trajectory sequence. This model fully utilizes prior task knowledge and is suitable for trajectory prediction over longer periods (e.g., 20-60 seconds) that conform to planned behavior.

[0087] The dual-mode prediction mechanism of this application has clear scenario-specificity and functional complementarity. The first predicted trajectory sequence (5-15 seconds) is based on Kalman filtering and a uniform velocity model, focusing on high-precision, low-latency short-term motion situational awareness, suitable for capturing sudden maneuvers. The second predicted trajectory sequence (20-60 seconds) relies heavily on structured operation planning routes obtained from airport A-CDM (Airport Collaborative Decision Making) systems or vehicle dispatching systems. This route is a digital representation of the vehicle's mission intent. By mapping the vehicle's current position onto the planned route and performing speed integration along the route, the resulting predicted trajectory is highly consistent with the driver's mission expectations, greatly improving the reliability of long-term behavior prediction. The organic combination of short-term motion extrapolation and long-term task extrapolation is the core innovation of this application.

[0088] S4. Calculate the predicted distance between the predicted position of the vehicle at each future time and the predicted position of the aircraft at the corresponding time in the first predicted trajectory sequence and the second predicted trajectory sequence respectively, to obtain the corresponding first future distance sequence and second future distance sequence.

[0089] In this embodiment, the predicted distance is specifically Euclidean distance; the generation process of the first future distance sequence specifically includes: determining the first predicted time period and the first predicted time step covered by the first predicted trajectory sequence; sequentially traversing each first future moment within the first predicted time period according to the first predicted time step; for each first future moment, extracting the predicted position coordinates of the vehicle at that first future moment from the first predicted trajectory sequence, and obtaining the predicted position coordinates of the aircraft at the corresponding first future moment based on motion state estimation; and calculating the Euclidean distance between the vehicle's predicted position coordinates and the aircraft's predicted position coordinates based on the predicted position coordinates of the vehicle and the aircraft.

[0090] The Euclidean distances calculated for all first future moments are collected in chronological order to form the first future distance sequence.

[0091] In this embodiment, the predicted distance is specifically Euclidean distance; the generation process of the second future distance sequence specifically includes: determining the second prediction period and the second prediction time step covered by the second prediction trajectory sequence; sequentially traversing each second future moment within the second prediction period according to the second prediction time step; for each second future moment, extracting the predicted position coordinates of the vehicle at that second future moment from the second prediction trajectory sequence, and obtaining the predicted position coordinates of the aircraft at the corresponding second future moment based on motion state estimation;

[0092] Based on the predicted position coordinates of the vehicle and the predicted position coordinates of the aircraft, the Euclidean distance between them is calculated; all the Euclidean distances calculated at the second future time are collected in chronological order to form the second future distance sequence.

[0093] Specifically, generating the future distance sequence includes the following steps:

[0094] For the first predicted trajectory sequence, determine the first predicted time period and the first predicted time step it covers. Iterate through each future time step, extracting the predicted vehicle position from the first predicted trajectory sequence, and extracting the predicted aircraft position at the same time from the aircraft predicted trajectory generated based on Kalman filtering. Calculate the Euclidean distance between two points at each time step. Arrange all distance values ​​in chronological order to obtain the first future distance sequence. .

[0095] Similarly, for the second predicted trajectory sequence, its second predicted time period and second predicted time step are determined, and the Euclidean distance between the predicted position of the vehicle along the planned route and the predicted position of the aircraft is calculated traversally to form the second future distance sequence. .

[0096] For the first future distance sequence respectively Second future distance sequence Perform the following calculations:

[0097] a) First Predicted Collision Time (TTC): Set a collision safety distance threshold. (e.g., 30 meters). In the distance sequence, starting from the current moment, search backwards until the first condition is met. or The time t is called TTC.

[0098] b) Closest Approach Distance (CPA) between Trajectories: Find the first future distance sequence Second future distance sequence The global minimum value over the entire forecast period is called CPA.

[0099] c) Relative velocity and approach angle: based on the first future distance sequence Second future distance sequence The positional changes reflected in the data are used to calculate the relative velocity vectors of the vehicle and the aircraft near the nearest approach point and the angle difference between the lines connecting their centers.

[0100] S5. Based on the first future distance sequence and the second future distance sequence, determine the collision risk level through the risk assessment model, and output graded early warning information according to the risk level.

[0101] In this embodiment, a risk assessment model is used to determine the collision risk level and output graded warning information. Specifically, this includes: for a first future distance sequence, calculating the corresponding first predicted collision time, the closest approach distance between the first trajectories, the first relative velocity, and the first approach angle; for a second future distance sequence, calculating the corresponding second predicted collision time, the closest approach distance between the second trajectories, the second relative velocity, and the second approach angle; and based on the first predicted collision time, the closest approach distance between the first trajectories, the first relative velocity, and the first approach angle, obtaining the first risk level through the risk assessment model.

[0102] Based on the second initial collision prediction time, the closest approach distance between the second trajectories, the second relative velocity, and the second approach angle, a second risk level is obtained through a risk assessment model. The first and second risk levels are then combined to determine a final comprehensive risk level. Based on the comprehensive risk level, the corresponding warning content and format are matched from a predefined warning strategy library to generate and output graded warning information.

[0103] In this embodiment, the first risk level and the second risk level are comprehensively processed to determine a final comprehensive risk level, specifically including:

[0104] Based on the preset risk level ranking, the lower-ranked level in the first risk level and the second risk level is determined as the comprehensive risk level.

[0105] The types of the first risk level, the second risk level, and the comprehensive risk level all include safety level, alert level, warning level, danger level, and emergency level; the risk level is ordered as follows: safety level, alert level, warning level, danger level, and emergency level, with the emergency level being the last and the safety level being the first.

[0106] Specifically, two sets of parameters calculated based on the first and second future distance sequences are input into a pre-defined risk assessment model. This risk assessment model is a rule-based decision engine or a trained machine learning model. Based on a comprehensive evaluation of parameters such as TTC, CPA, relative speed, and approach angle, the risk assessment model outputs a preliminary risk level for each set of parameters, denoted as the first risk level (from short-term prediction) and the second risk level (from long-term prediction). The risk levels include at least: safe, alert, warning, dangerous, and emergency. The two risk levels are then combined according to a pre-defined comprehensive decision. A preferred decision is to rank the risk levels in the order of "safe < alert < warning < dangerous < emergency," selecting the level that ranks lower between the first and second risk levels as the final comprehensive risk level. This reflects the safety principle of "choosing the highest warning level over the lowest."

[0107] The levels are categorized as follows: Safety Level, Warning Level, Alert Level, Danger Level, and Emergency Level. The determination logic for each level is as follows:

[0108] Safety Level: Indicates that there is sufficient distance between the vehicle and the aircraft, and no risk of collision. A vehicle is classified as having a safety level when one of the following conditions is met:

[0109] After searching the H3 neighborhood, no potential conflicting objects were found;

[0110] The current real-time distance between the vehicle and the aircraft is greater than the preset safe distance threshold, and the calculated first predicted collision time (TTC) is greater than 30 seconds;

[0111] The nearest approach distance (CPA) between predicted trajectories is greater than the preset collision distance threshold.

[0112] Warning Level: Indicates a potential long-term approach risk, requiring the driver's attention but the situation is not urgent. It is triggered when the predicted trajectory shows that the vehicle and aircraft may approach each other within 30 seconds, and the calculated closest approach distance (CPA) is near the safety boundary.

[0113] Warning level: Indicates a clear proximity risk, requiring the driver to take proactive precautions and prepare to take action. Its triggering conditions must be met simultaneously:

[0114] The current real-time distance is approaching the safe distance threshold;

[0115] The calculated first predicted collision time (TTC) is less than 30 seconds, and the closest approach distance (CPA) is less than the collision distance threshold.

[0116] Hazard Level: Indicates a high level of danger, where the vehicle is too close to the aircraft and immediate action is required. A vehicle is classified as hazardous when one of the following conditions is met:

[0117] The current real-time distance is less than a more stringent proximity threshold;

[0118] The calculated first predicted collision time (TTC) is less than 10 seconds.

[0119] Emergency Level: Indicates an extremely urgent situation where a collision is imminent, requiring immediate emergency braking or sharp steering to avoid it. Its triggering conditions are:

[0120] The current real-time distance is less than an extreme near-distance threshold;

[0121] The calculated first predicted collision time (TTC) is less than 5 seconds.

[0122] Based on the determined comprehensive risk level, a predefined early warning strategy library is queried. This library stores specific response strategies corresponding to different risk levels. These early warning strategies include at least: terminal alarm format: icon color (e.g., green, blue, yellow, orange, red), flashing frequency, text prompt content, and broadcast voice content, tone, and repetition count displayed on vehicle-mounted terminals or monitoring screens. Finally, based on the strategies, a final tiered early warning message is generated and simultaneously output through multiple methods, including vehicle-mounted communication terminals (voice + text / image), airport operation monitoring screens (visual plotting), and control tower / dispatch positions (pop-up notifications).

[0123] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for early warning of collisions on an apron, characterized in that, Includes the following steps: S1. Within the airport apron area, based on the H3 geographic grid index of vehicles and aircraft, a bidirectional neighborhood search is performed using a grid neighborhood query algorithm to filter out vehicle and aircraft pairs that fall within each other's neighborhood range, thus obtaining a set of candidate conflict target pairs. S2. Calculate the real-time geometric distance between each vehicle and aircraft pair in the candidate conflict target pair set. If the real-time geometric distance is less than the safety threshold, generate a real-time collision alarm. S3. For each vehicle and aircraft pair in the candidate conflict target pair set, perform future trajectory prediction to obtain the first predicted trajectory sequence and the second predicted trajectory sequence corresponding to the vehicle and aircraft pair; wherein, the first predicted trajectory sequence is obtained based on motion state estimation, the second predicted trajectory sequence is obtained based on the vehicle's predefined operation planning route and its current state deduction, and the future time period covered by the second predicted trajectory sequence is longer than the future time period covered by the first predicted trajectory sequence. S4. Calculate the predicted distance between the predicted position of the vehicle at each future time and the predicted position of the aircraft at the corresponding time in the first predicted trajectory sequence and the second predicted trajectory sequence respectively, to obtain the corresponding first future distance sequence and second future distance sequence. S5. Based on the first future distance sequence and the second future distance sequence, determine the collision risk level through the risk assessment model, and output graded early warning information according to the risk level.

2. The apron collision avoidance early warning method according to claim 1, characterized in that, The H3 geographic grid index for the vehicle and aircraft is obtained through the following steps: receiving vehicle positioning data and aircraft positioning data; converting the geographic coordinates in the vehicle positioning data and aircraft positioning data into the corresponding H3 geographic grid index; the grid neighborhood query algorithm is the kRing neighborhood query algorithm of the H3 library, and the neighborhood range is determined by converting the preset physical security monitoring radius into the H3 grid topology distance k.

3. The apron collision avoidance early warning method according to claim 2, characterized in that, In step S3, the process of obtaining the first predicted trajectory sequence based on motion state estimation specifically includes: A motion state model is established for each vehicle and aircraft pair in the candidate conflict target pair set. The state vector of the motion state model includes at least the position coordinates and velocity components of the target on the two-dimensional plane. Based on the continuous vehicle positioning data and continuous aircraft positioning data corresponding to each vehicle and aircraft pair, the state vector is recursively estimated and updated using the Kalman filter algorithm to obtain the optimal state estimate of the vehicle and aircraft pair at the current moment. Based on the optimal state estimation and the preset state transition model, the state of the vehicle and aircraft at each discrete time point in the first future time period is recursively predicted, thereby generating the first predicted trajectory sequence.

4. The apron collision avoidance early warning method according to claim 3, characterized in that, The state transition model is a uniform motion model; the future time period covered by the first predicted trajectory sequence is 5 to 15 seconds.

5. The apron collision avoidance early warning method according to claim 1, characterized in that, The second predicted trajectory sequence covers a future second time period of 20 to 60 seconds; In step S3, the process of obtaining the second predicted trajectory sequence based on the vehicle's predefined operation planning route and its current state specifically includes: For each vehicle and aircraft pair in the candidate conflict target pair set, obtain the predefined operation planning route for the vehicle in that vehicle and aircraft pair, the operation planning route being defined by a series of ordered geographical locations; Based on the vehicle's current geographical location, determine its projected position on the planned operation route and the length of the path already traveled; Based on the vehicle's current speed and the preset prediction step size, calculate its cumulative predicted travel distance at each prediction time in the future second time period. Based on the cumulative predicted travel distance, the corresponding future geographical locations are determined by interpolation calculation on the planned operation route, and then connected in chronological order to form the second predicted trajectory sequence.

6. The apron collision avoidance early warning method according to claim 1, characterized in that, The predicted distance is specifically Euclidean distance; the generation process of the first future distance sequence specifically includes: Determine the first prediction period and the first prediction time step covered by the first predicted trajectory sequence; According to the first prediction time step, iterate through each first future moment within the first prediction period; For each first future moment, the predicted position coordinates of the vehicle at that first future moment are extracted from the first predicted trajectory sequence, and the predicted position coordinates of the aircraft at the corresponding first future moment are obtained based on the motion state estimation. Calculate the Euclidean distance between the vehicle's predicted position coordinates and the aircraft's predicted position coordinates. The Euclidean distances calculated for all first future moments are collected in chronological order to form the first future distance sequence.

7. The apron collision avoidance early warning method according to claim 1, characterized in that, The predicted distance is specifically Euclidean distance; the generation process of the second future distance sequence specifically includes: Determine the second prediction period and the second prediction time step covered by the second prediction trajectory sequence; According to the second prediction time step, iterate through each second future moment within the second prediction period; For each second future time, the predicted position coordinates of the vehicle at that second future time are extracted from the second predicted trajectory sequence, and the predicted position coordinates of the aircraft at the corresponding second future time are obtained based on the motion state estimation. Calculate the Euclidean distance between the vehicle's predicted position coordinates and the aircraft's predicted position coordinates. The Euclidean distances calculated at all second future moments are collected in chronological order to form the second future distance sequence.

8. The apron collision avoidance early warning method according to claim 1, characterized in that, The process of determining the collision risk level and outputting graded early warning information through a risk assessment model specifically includes: For the first future distance sequence, calculate its corresponding first predicted collision time, the closest approach distance between the first trajectories, the first relative velocity, and the first approach angle; For the second future distance sequence, calculate its corresponding second first predicted collision time, the closest approach distance between the second trajectories, the second relative velocity, and the second approach angle; Based on the first predicted collision time, the closest approach distance between the first trajectories, the first relative velocity, and the first approach angle, the first risk level is obtained through a risk assessment model. Based on the second initial collision time, the closest approach distance between the second trajectories, the second relative velocity, and the second approach angle, the second risk level is obtained through a risk assessment model; The first and second risk levels are combined to determine a final comprehensive risk level. Based on the comprehensive risk level, the system matches the corresponding warning content and format from a predefined warning strategy library to generate and output tiered warning information.

9. The apron collision avoidance early warning method according to claim 8, characterized in that, The process of comprehensively processing the first risk level and the second risk level to determine a final comprehensive risk level specifically includes: Based on the preset risk level ranking, the lower-ranked level in the first risk level and the second risk level is determined as the comprehensive risk level. The types of the first risk level, the second risk level, and the comprehensive risk level all include safety level, alert level, warning level, danger level, and emergency level; the risk levels are ordered as follows: safety level, alert level, warning level, danger level, and emergency level, with the emergency level being the last and the safety level being the first.

10. The apron collision avoidance early warning method according to claim 1, characterized in that, In step S2, the calculation of the real-time geometric distance includes any of the following methods: Calculate the planar Euclidean distance between the center points of the vehicle and the aircraft; or, Construct outer envelope geometric models for vehicles and aircraft respectively, and calculate the shortest spatial distance between the two outer envelope geometric models.

Images