Airport scene aircraft collision early warning system and method based on real-time positioning

Abstract

The invention discloses a scene aircraft collision early warning system and method based on real-time positioning, and belongs to the technical field of airport scene monitoring and safety protection, and the method comprises the steps: obtaining the three-dimensional space coordinates of an aircraft based on the real-time positioning technology, building a kinematic model through the fusion of a velocity vector and a course angle parameter, and carrying out the real-time positioning. Millimeter-level positioning of key parts of the aircraft is realized through three-dimensional point cloud data processing; a plurality of three-dimensional collision situation analysis models are established, and multi-dimensional risk assessment is realized in combination with an improved safety anti-collision time (TTC) algorithm. According to the method, a distributed computing platform is adopted, and through cooperative operation of a point cloud feature matching algorithm and a track prediction model, the problem of monitoring blind areas of a traditional method in complex scenes such as taxiway intersections is effectively solved, and the accuracy of aircraft collision early warning on the airport scene is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of airport scene monitoring and safety protection, and in particular to a scene aircraft collision warning system and method based on real-time positioning. Background Art

[0002] With the rapid development of the air transport industry, airport traffic density has increased significantly, and the incidence of aircraft collisions has climbed annually. According to statistics, approximately 58% of aircraft airworthiness issues caused by collisions at airports over the past three years can be attributed to the inadequate early warning capabilities of traditional surveillance technologies. Traditional surface surveillance technologies (such as radar and ADS-B) simplify aircraft tracking as point signals and are unable to accurately determine the three-dimensional spatial location of vulnerable areas such as wingtips and tails, resulting in low collision warning accuracy and high false alarm rates. Existing point signal-based positioning solutions can monitor the global position of aircraft, but lack the ability to accurately locate local areas in real time. They only support early warning for a single collision type (such as head-on collisions) and fail to cover complex collision scenarios such as rear-end and side collisions. A collision warning method based on multi-sensor fusion has been proposed, integrating radar and visual data to improve positioning accuracy. However, this method relies on a fixed threshold to determine collision risk and lacks a dynamic time-to-collision (TTC) model, making it difficult to adapt to real-time traffic flow changes at airports. A collision warning algorithm was designed for drone-vehicle collaborative scenarios, but its model did not integrate three-dimensional point cloud positioning technology and did not consider the impact of key parameters such as aircraft wingspan on lateral distance calculations, resulting in a warning response delay of up to 3.5 seconds, which failed to meet civil aviation safety standards.

[0003] To sum up, the existing technologies have the following core defects: insufficient positioning accuracy. Traditional technologies are unable to obtain millimeter-level coordinates of aircraft areas prone to scratches (such as wingtips), resulting in large errors in collision judgment; the collision type coverage is single. Existing solutions are mostly targeted at single scenarios (such as straight-line head-on collisions) and lack the ability to dynamically model multiple types of collisions; dynamic adaptability is poor. The static threshold warning model does not combine dynamic factors such as driver reaction time and braking parameters, resulting in high false alarm and missed alarm rates; the verification system is imperfect and lacks high-precision simulation platform verification, making it difficult to ensure system stability in complex scene environments.

[0004] Therefore, there is an urgent need for a technical solution that integrates high-precision positioning, multi-type collision dynamic modeling and real-time TTC calculation to comprehensively improve the accuracy and reliability of airport scene collision warning and reduce the risk of aircraft operations. Summary of the Invention

[0005] The present invention aims to provide a scene aircraft collision warning system and method based on real-time positioning. By integrating high-precision three-dimensional point cloud positioning technology with multi-type collision dynamic modeling algorithms, it solves the problems of insufficient positioning accuracy and single collision warning type of traditional surveillance technology, provides intelligent and full-coverage technical support for the safe operation of airport scenes, and meets the strict requirements of the civil aviation field for real-time and multi-scenario adaptability.

[0006] To achieve the above object, the technical solution provided by the present invention is:

[0007] A scene aircraft collision warning system based on real-time positioning includes the following modules:

[0008] The positioning data acquisition module, based on real-time positioning technology, uses a 3D point cloud and surveillance image matching point detection algorithm and integrates the SQPnP algorithm to obtain the millimeter-level real-time 3D coordinates of key aircraft parts in the camera coordinate system. These key parts include the wing tip, the fuselage center of gravity axis point, and the tail vertex.

[0009] A motion parameter calculation module, based on the time series data of the millimeter-level real-time three-dimensional coordinates, obtains the instantaneous velocity of the key parts of the aircraft through differential calculation, and fuses the velocity vectors of multiple key points to calculate the average driving speed and relative heading angle;

[0010] A collision type identification module pre-determines the collision type based on the relative heading angle and velocity relationship, including head-on collision, rear-end collision, and side collision;

[0011] The early warning model processing module establishes corresponding mathematical models to calculate the collision time for different collision types and compares it with the safe collision avoidance time, namely TTC. Based on the real-time positioning results of the aircraft, combined with the relative heading angle, average driving speed and collision time, different trigger conditions are set for three typical collision types to trigger early warnings.

[0012] Furthermore, the positioning data acquisition module includes:

[0013] The matching point detection unit collects 750 sets of surveillance images using a 1:200 scene simulation platform, annotates the aircraft categories and matching point coordinates, and divides the images into 690 training sets and 60 validation sets. This training and validation sets are used to train a multi-category matching point detection model. The model uses a multi-category top-down matching point detection algorithm to extract the 2D coordinates of the aircraft target area and matching points from the surveillance images. These are then feature-matched with the coordinates of preset key parts in the 3D point cloud model to generate 2D and 3D coordinate correspondences as input parameters for the pose solution unit.

[0014] The pose solving unit inputs the correspondence between two-dimensional and three-dimensional coordinates and integrates them with the SQPnP algorithm to solve the pose matrix of the three-dimensional point cloud coordinate system and the camera coordinate system, and projects the high-resolution point cloud to the camera coordinate system to obtain the millimeter-level real-time three-dimensional coordinates of the key parts of the aircraft in the camera coordinate system, thereby realizing millimeter-level real-time positioning of the key parts of the aircraft.

[0015] Furthermore, the matching point detection unit collects 750 sets of surveillance images through a 1:200 scene simulation platform, annotates aircraft categories and matching point coordinates, and divides them into 690 training sets and 60 validation sets. The multi-category matching point detection model is trained using the training set and validation set. The model uses a multi-category top-down matching point detection algorithm to extract the two-dimensional coordinates of the aircraft target area and matching points from the surveillance images, and performs feature matching with the coordinates of preset key parts in the three-dimensional point cloud model to generate a two-dimensional and three-dimensional coordinate correspondence as the input parameters of the pose solution unit. The two-dimensional coordinates of the aircraft target area and matching points are extracted from the surveillance images. The specific content is:

[0016] The surveillance image is input into the target detection module, which identifies the aircraft type and generates a target frame. Based on the target frame, the Lite-HRNet network outputs the two-dimensional coordinates of the aircraft target area and matching points. The detection error is controlled within 10 pixels. The Lite-HRNet network is also denoted as a lightweight high-resolution network.

[0017] Furthermore, the motion parameter calculation module includes:

[0018] a speed calculation unit for calculating an average speed of the aircraft based on the three-dimensional coordinate differences of adjacent frames obtained from the millimeter-level real-time three-dimensional coordinates and a camera frame rate;

[0019] The heading angle calculation unit uses the law of cosines to solve the relative heading angle through the coordinates of the aircraft's key points.

[0020] Furthermore, the specific calculation formulas for the average driving speed and the relative heading angle are:

[0021]

[0022] Among them S A is the distance from aircraft A to the collision point, S B is the distance from aircraft B to the collision point, n' and m' are time series indexes n'>m', n is the number of key points, v i is the instantaneous velocity of each point, v1,…,v n are the speeds of the 1st, ..., nth key points respectively, v is the average speed of the aircraft, K2, K1 and K'2, K1' are the key points of the two aircraft respectively, θ1, θ2 are the relative heading angles.

[0023] Furthermore, the judgment logic of the collision type identification module is:

[0024] Head-on collision: triggered when the relative heading angle difference is ≤5° and the lateral distance is less than the wingspan;

[0025] Rear-end collision, when the vehicles are traveling in the same direction and the average speed difference exceeds the threshold, that is, v A ≠v B When triggered, the average speed of aircraft A is v A cm / s, the average speed of aircraft B is v B cm / s;

[0026] A side collision is triggered when the relative heading angle is greater than 5° and less than 175°.

[0027] Furthermore, the early warning model processing module includes:

[0028] The safety collision avoidance time calculation unit dynamically calculates TTC by combining the driver's reaction time, brake response time, and braking duration;

[0029] The collision time comparison unit compares the predicted collision time with the TTC and outputs warning information through a visual interface based on the comparison results. The warning information includes a collision type identification code, predicted collision time, TTC threshold and recommended avoidance strategy, where the identification code is mapped to head-on, rear-end and side collision types.

[0030] Furthermore, the TTC calculation formula is:

[0031]

[0032] Among them, the driver's reaction time t1 = 2s, the system processing time t2, the brake response time t3 = 1.2s, the braking duration t4, the aircraft's speed v, and the gravity acceleration g = 9.8m / s 2 And the road adhesion coefficient μ, dry road μ = 0.8, wet road μ = 0.4.

[0033] Furthermore, the system also includes a multi-scenario verification module, which verifies the warning accuracy under different perspectives and matching point distributions based on a 1:200 scene simulation platform.

[0034] In addition, the present invention also proposes a scene aircraft collision warning method based on real-time positioning. The scene aircraft collision warning method based on real-time positioning is applicable to the scene aircraft collision warning system based on real-time positioning. The specific steps of the method are as follows:

[0035] Step S1: Processing the surveillance image using the matching point detection unit in the positioning data acquisition module, using a multi-category top-down matching point detection algorithm, using the target detection module to identify the aircraft type and generate a target frame. Based on the target frame, the Lite-HRNet network is used to extract the aircraft target area and the two-dimensional coordinates of the matching points. The extracted two-dimensional coordinates of the matching points are feature matched with the coordinates of the preset key parts in the three-dimensional point cloud model to generate a correspondence between the two-dimensional and three-dimensional coordinates.

[0036] Step S2: The pose calculation unit in the positioning data acquisition module receives the two-dimensional and three-dimensional coordinate correspondences generated in step S1 and inputs them into the SQPnP algorithm to solve the pose matrix between the three-dimensional point cloud coordinate system and the camera coordinate system. The pose matrix includes the rotation matrix R and the translation vector T. The three-dimensional point cloud is projected onto the camera coordinate system through the pose matrix to obtain the millimeter-level real-time three-dimensional coordinates of the key parts of the aircraft in the camera coordinate system.

[0037] Step S3: Based on the millimeter-level real-time 3D coordinates, the 3D coordinate differences of adjacent frames are obtained. In combination with the camera frame rate, the instantaneous velocity of each key point is calculated, and then the average instantaneous velocity of all key points is calculated to obtain the average speed of the aircraft. The heading angle calculation unit uses the law of cosines to solve the relative heading angle of the two aircraft key point coordinates.

[0038] Step S4: The collision type identification module receives the relative heading angle and average speed calculated in step S3 and performs a preliminary judgment of the collision type in combination with the aircraft wingspan parameters:

[0039] Identify collision types based on speed and heading angle, including head-on collision, rear-end collision, and side collision;

[0040] When the relative heading angle difference between two aircraft is less than or equal to 5° and the lateral distance between them is less than the wingspan of the aircraft, it is determined to be a head-on collision risk;

[0041] When two aircraft are traveling in the same direction and their speed difference exceeds the set threshold, that is, v A ≠v B When the vehicle is in the rear-end collision risk range, it is determined as a rear-end collision risk;

[0042] When the relative heading angle between the two aircraft is greater than 5° and less than 175°, it is determined to be a side collision risk;

[0043] And output the corresponding collision type identification;

[0044] In step S5, the safe collision avoidance time calculation unit in the warning model processing module calculates the TTC based on the collision type identification in step S4, combined with the driver's reaction time, brake response time and braking duration; the collision time comparison unit compares the predicted collision time with the TTC, and outputs warning information through a visual interface based on the comparison result.

[0045] Compared with the prior art, the present invention adopts the above technical solution and has the following beneficial effects:

[0046] (1) The present invention proposes a scene aircraft collision warning system and method based on real-time positioning. Through the three-dimensional point cloud and surveillance image matching point detection technology, combined with high-precision three-dimensional scanners and surveillance cameras, it can accurately obtain the millimeter-level coordinates of the aircraft's easily scratched areas (such as wingtips), effectively reducing the error in collision judgment and providing more accurate collision warnings.

[0047] (2) The present invention proposes a scene aircraft collision warning system and method based on real-time positioning, which uses a combination of a high-precision three-dimensional scanner and a surveillance camera to obtain the real-time position information of the aircraft from multiple dimensions, further improving the accuracy and reliability of positioning.

[0048] (3) The present invention proposes a scene aircraft collision warning system and method based on real-time positioning. Based on a multi-category top-down matching point detection algorithm, the system extracts the two-dimensional coordinates of the aircraft target area and matching points from the surveillance image, and combines them with three-dimensional point cloud data for positioning. This system can effectively deal with various collision scenarios in complex scene environments.

[0049] (4) The present invention proposes a scene aircraft collision warning system and method based on real-time positioning. Through dynamic modeling of the relationship between relative heading angle and velocity, it can accurately predict the type of collision and establish corresponding mathematical models for different collision types to calculate the collision time, thereby providing more comprehensive warning capabilities.

[0050] (5) The present invention proposes a scene aircraft collision warning system and method based on real-time positioning. By verifying the warning effect in different scenarios through a simulation platform, the system performance can be comprehensively evaluated, and the algorithm and model can be further optimized to make it more adaptable and stable in complex scene environments, thereby improving the verification system. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flowchart for aircraft collision warning processing;

[0052] Figure 2 This is a schematic diagram of the relative relationship between the two aircraft;

[0053] Figure 3 This is a schematic diagram of the head-to-head collision algorithm model;

[0054] Figure 4 This is a schematic diagram of the rear-end collision algorithm model;

[0055] Figure 5 This is a schematic diagram of the side collision algorithm model;

[0056] Figure 6 Screenshot of the early warning model verification scenario. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0058] Example 1

[0059] A scene aircraft collision warning system based on real-time positioning includes the following modules:

[0060] The positioning data acquisition module, based on real-time positioning technology, uses a 3D point cloud and surveillance image matching point detection algorithm and integrates the SQPnP algorithm to obtain the millimeter-level real-time 3D coordinates of key aircraft parts in the camera coordinate system. These key parts include the wing tip, the fuselage center of gravity axis point, and the tail vertex.

[0061] A motion parameter calculation module, based on the millimeter-level real-time three-dimensional coordinate time series data, obtains the instantaneous velocity of the key parts of the aircraft through differential calculation, and fuses the velocity vectors of multiple key points to calculate the average driving speed and relative heading angle;

[0062] The collision type recognition module pre-determines the collision type based on the relative heading angle and speed relationship. The collision types include head-on collision, rear-end collision and side collision. The three collision types are shown in the figure below. Figure 3 、 4 , as shown in 5;

[0063] The early warning model processing module establishes corresponding mathematical models to calculate the collision time for different collision types and compares it with the safe collision avoidance time, namely TTC. Based on the real-time positioning results of the aircraft, combined with the relative heading angle, average driving speed and collision time, different trigger conditions are set for three typical collision types to trigger early warnings.

[0064] Furthermore, the positioning data acquisition module includes:

[0065] The matching point detection unit collects 750 sets of surveillance images using a 1:200 scene simulation platform, annotates the aircraft categories and matching point coordinates, and divides the images into 690 training sets and 60 validation sets. This training and validation sets are used to train a multi-category matching point detection model. The model uses a multi-category top-down matching point detection algorithm to extract the 2D coordinates of the aircraft target area and matching points from the surveillance images. These are then feature-matched with the coordinates of preset key parts in the 3D point cloud model to generate 2D and 3D coordinate correspondences as input parameters for the pose solution unit.

[0066] The pose solving unit inputs the correspondence between two-dimensional and three-dimensional coordinates and integrates them with the SQPnP algorithm to solve the pose matrix of the three-dimensional point cloud coordinate system and the camera coordinate system, and projects the high-resolution point cloud to the camera coordinate system to obtain the millimeter-level real-time three-dimensional coordinates of the key parts of the aircraft in the camera coordinate system, thereby realizing millimeter-level real-time positioning of the key parts of the aircraft.

[0067] Furthermore, the matching point detection unit collects 750 sets of surveillance images through a 1:200 scene simulation platform, annotates aircraft categories and matching point coordinates, and divides them into 690 training sets and 60 validation sets. The multi-category matching point detection model is trained using the training set and validation set. The model uses a multi-category top-down matching point detection algorithm to extract the two-dimensional coordinates of the aircraft target area and matching points from the surveillance images, and performs feature matching with the coordinates of preset key parts in the three-dimensional point cloud model to generate a two-dimensional and three-dimensional coordinate correspondence as the input parameters of the pose solution unit. The two-dimensional coordinates of the aircraft target area and matching points are extracted from the surveillance images. The specific content is:

[0068] The surveillance image is input into the target detection module, which identifies the aircraft type and generates a target frame. Based on the target frame, the Lite-HRNet network outputs the two-dimensional coordinates of the aircraft target area and matching points. The detection error is controlled within 10 pixels. The Lite-HRNet network is also denoted as a lightweight high-resolution network.

[0069] Furthermore, the motion parameter calculation module includes:

[0070] a speed calculation unit for calculating an average speed of the aircraft based on the three-dimensional coordinate differences of adjacent frames obtained from the millimeter-level real-time three-dimensional coordinates and a camera frame rate;

[0071] The heading angle calculation unit uses the law of cosines to solve the relative heading angle through the coordinates of the aircraft's key points.

[0072] Furthermore, the specific calculation formulas for the average driving speed and the relative heading angle are:

[0073]

[0074] Among them S A is the distance from aircraft A to the collision point, S B is the distance from aircraft B to the collision point, n' and m' are time series indexes n'>m', n is the number of key points, v i is the instantaneous velocity of each point, v1,…,v n are the speeds of the 1st, ..., nth key points respectively, v is the average speed of the aircraft, K2, K1 and K'2, K1' are the key points of the two aircraft respectively, θ1, θ2 are the relative heading angles.

[0075] Furthermore, the judgment logic of the collision type identification module is:

[0076] Head-on collision: triggered when the relative heading angle difference is ≤5° and the lateral distance is less than the wingspan;

[0077] Rear-end collision, when the vehicles are traveling in the same direction and the average speed difference exceeds the threshold, that is, v A ≠v B When triggered, the average speed of aircraft A is v A cm / s, the average speed of aircraft B is v B cm / s;

[0078] A side collision is triggered when the relative heading angle is greater than 5° and less than 175°.

[0079] Furthermore, the early warning model processing module includes:

[0080] The safety collision avoidance time calculation unit dynamically calculates TTC by combining the driver's reaction time, brake response time, and braking duration;

[0081] The collision time comparison unit compares the predicted collision time with the TTC and outputs warning information through a visual interface based on the comparison results. The warning information includes a collision type identification code, predicted collision time, TTC threshold and recommended avoidance strategy, where the identification code is mapped to head-on, rear-end and side collision types.

[0082] Furthermore, the TTC calculation formula is:

[0083]

[0084] Among them, the driver's reaction time t1 = 2s, the system processing time t2, the brake response time t3 = 1.2s, the braking duration t4, the aircraft's speed v, and the gravity acceleration g = 9.8m / s 2 And the road adhesion coefficient μ, dry road μ = 0.8, wet road μ = 0.4.

[0085] Furthermore, the system also includes a multi-scenario verification module, such as Figure 6 As shown in the figure, based on a 1:200 scene simulation platform, the warning accuracy under different viewing angles and matching point distributions is verified.

[0086] In addition, the present invention also proposes a scene aircraft collision warning method based on real-time positioning, and the scene aircraft collision warning method based on real-time positioning is applicable to the scene aircraft collision warning system based on real-time positioning, such as Figure 1 As shown, the specific steps of the method are as follows:

[0087] Step S1: Processing the surveillance image using the matching point detection unit in the positioning data acquisition module, using a multi-category top-down matching point detection algorithm, using the target detection module to identify the aircraft type and generate a target frame. Based on the target frame, the Lite-HRNet network is used to extract the aircraft target area and the two-dimensional coordinates of the matching points. The extracted two-dimensional coordinates of the matching points are feature matched with the coordinates of the preset key parts in the three-dimensional point cloud model to generate a correspondence between the two-dimensional and three-dimensional coordinates.

[0088] Step S2: The pose calculation unit in the positioning data acquisition module receives the two-dimensional and three-dimensional coordinate correspondences generated in step S1 and inputs them into the SQPnP algorithm to solve the pose matrix between the three-dimensional point cloud coordinate system and the camera coordinate system. The pose matrix includes the rotation matrix R and the translation vector T. The three-dimensional point cloud is projected onto the camera coordinate system through the pose matrix to obtain the millimeter-level real-time three-dimensional coordinates of the key parts of the aircraft in the camera coordinate system.

[0089] Step S3: Based on the millimeter-level real-time 3D coordinates, the 3D coordinate differences of adjacent frames are obtained. In combination with the camera frame rate, the instantaneous velocity of each key point is calculated, and then the average instantaneous velocity of all key points is calculated to obtain the average speed of the aircraft. The heading angle calculation unit uses the law of cosines to solve the relative heading angle of the two aircraft key point coordinates.

[0090] Step S4: The collision type identification module receives the relative heading angle and average speed calculated in step S3 and performs a preliminary judgment of the collision type in combination with the aircraft wingspan parameters:

[0091] Identify collision types based on speed and heading angle, including head-on collision, rear-end collision, and side collision;

[0092] When the relative heading angle difference between two aircraft is less than or equal to 5° and the lateral distance between them is less than the wingspan of the aircraft, it is determined to be a head-on collision risk;

[0093] When two aircraft are traveling in the same direction and their speed difference exceeds the set threshold, that is, v A ≠v BWhen the vehicle is in the rear-end collision risk range, it is determined as a rear-end collision risk;

[0094] When the relative heading angle between the two aircraft is greater than 5° and less than 175°, it is determined to be a side collision risk;

[0095] And output the corresponding collision type identification;

[0096] In step S5, the safe collision avoidance time calculation unit in the warning model processing module calculates the TTC based on the collision type identification in step S4, combined with the driver's reaction time, brake response time and braking duration; the collision time comparison unit compares the predicted collision time with the TTC, and outputs warning information through a visual interface based on the comparison result.

[0097] Example 2

[0098] System construction and data preparation

[0099] 1. Hardware Deployment: Deploy surveillance cameras (such as the Intel RealSense L515) at the airport to capture images from a bird's-eye view. Configure a high-precision 3D scanner to obtain point cloud models of aircraft (such as the A321 and B737) and annotate the 3D coordinates of eight key matching points (such as the nose, wingtip, and tail).

[0100] 2. Dataset Construction: 750 sets of surveillance images were collected using a 1:200 scene simulation platform. The images were labeled with aircraft categories and matching point coordinates. The images were divided into 690 training sets and 60 validation sets for training the multi-category matching point detection model (SSDLite + Lite-HRNet).

[0101] Real-time positioning process

[0102] 1. Matching point detection:

[0103] The monitoring image is input into the target detection module (SSDLite), which identifies the aircraft type and generates a target box;

[0104] Based on the target frame, the Lite-HRNet network outputs the two-dimensional coordinates of 8 matching points, and the detection error is controlled within 10 pixels.

[0105] 2. Pose calculation:

[0106] The two-dimensional matching points are paired with the three-dimensional coordinates in the point cloud model, the pose matrix (R, T) is solved by the SQPnP algorithm, the point cloud is projected to the camera coordinate system, and the three-dimensional coordinates of the local area of the aircraft (such as the wingtip coordinates) are obtained.

[0107] Collision warning execution steps

[0108] 1. Motion parameter calculation:

[0109] Speed calculation: Based on the coordinate differences between two adjacent frames (with an interval of 40 ms), calculate the average speed of 8 key points. The formula is:

[0110]

[0111] - Course angle calculation: Solve the relative course angles θ1 and θ2 of the key points (such as K2, K2') of two aircraft through the cosine theorem. The formula is:

[0112]

[0113] 2. Collision type identification:

[0114] Head-on collision: If |θ1 - θ2| ≤ 5° and the lateral distance k ≥ |d · sinθ1| (k is the average wingspan), it is determined as a head-on collision risk;

[0115] Rear-end collision: When traveling in the same direction, if v1 > v2 and |θ1| < |θ2| (active rear-end collision) or v1 < v2 and |θ1| > |θ2| (passive rear-end collision), it is determined as a rear-end collision risk;

[0116] Side collision: If 5° < |θ1 + θ2| < 175°, it is determined as a side collision risk.

[0117] 3. Early warning model processing: <了

[0118] Calculation of the safe anti-collision time (TTC):

[0119]

[0120] Among them, the driver's reaction time t1 = 2 s, and the system processing time t 2, The brake response time t3 = 1.2 s, and the braking duration is calculated based on the speed v, the gravitational acceleration g (9.8 m / s 2 ), and the road surface adhesion coefficient μ (0.8 for dry road surface, 0.4 for wet road surface).

[0121] Collision time comparison: According to the geometric models of different collision types, calculate the predicted collision time. If the collision time is less than TTC, send out an audible and visual early warning through the system interface.

[0122] Verification and optimization:

[0123] Verify the early warning accuracy under different scenarios (such as side view angle, concentrated distribution of matching points) through the simulation platform, adjust the number of matching points (recommended 8 points evenly distributed) and algorithm parameters to ensure that the absolute error ≤ 0.64 cm and the relative error ≤ 10%, meeting the airport surface safety requirements.

[0124] It should be noted that there seems to be a misspelling in "了0000280" in the original text. I translated it as "了0000280" as it is, but it might need to be corrected in the original content.This technical solution achieves real-time monitoring and dynamic risk prediction of areas where aircraft are prone to scratches through the integration of high-precision positioning and multi-model collision warning, significantly improving the intelligent level of airport safety protection.

[0125] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A scene aircraft collision warning system based on real-time positioning, characterized in that: Includes the following modules: The positioning data acquisition module, based on real-time positioning technology, uses a 3D point cloud and surveillance image matching point detection algorithm and integrates the SQPnP algorithm to obtain the millimeter-level real-time 3D coordinates of key aircraft parts in the camera coordinate system. These key parts include the wing tip, the fuselage center of gravity axis point, and the tail vertex. A motion parameter calculation module, based on the time series data of the millimeter-level real-time three-dimensional coordinates, obtains the instantaneous velocity of the key parts of the aircraft through differential calculation, and fuses the velocity vectors of multiple key points to calculate the average driving speed and relative heading angle; A collision type identification module pre-determines the collision type based on the relative heading angle and velocity relationship, including head-on collision, rear-end collision, and side collision; The early warning model processing module establishes corresponding mathematical models to calculate the collision time for different collision types and compares it with the safe collision avoidance time, namely TTC. Based on the real-time positioning results of the aircraft, combined with the relative heading angle, average driving speed and collision time, different trigger conditions are set for three typical collision types to trigger early warnings.

2. The scene aircraft collision warning system based on real-time positioning according to claim 1, characterized in that: The positioning data acquisition module includes: The matching point detection unit collects 750 sets of surveillance images using a 1:200 scene simulation platform, annotates the aircraft categories and matching point coordinates, and divides the images into 690 training sets and 60 validation sets. This training and validation sets are used to train a multi-category matching point detection model. The model uses a multi-category top-down matching point detection algorithm to extract the 2D coordinates of the aircraft target area and matching points from the surveillance images. These are then feature-matched with the coordinates of preset key parts in the 3D point cloud model to generate 2D and 3D coordinate correspondences as input parameters for the pose solution unit. The pose solving unit inputs the correspondence between two-dimensional and three-dimensional coordinates and integrates them with the SQPnP algorithm to solve the pose matrix of the three-dimensional point cloud coordinate system and the camera coordinate system, and projects the high-resolution point cloud to the camera coordinate system to obtain the millimeter-level real-time three-dimensional coordinates of the key parts of the aircraft in the camera coordinate system, thereby realizing millimeter-level real-time positioning of the key parts.

3. The scene aircraft collision warning system based on real-time positioning according to claim 2, characterized in that: The matching point detection unit collects 750 sets of surveillance images through a 1:200 scene simulation platform, annotates aircraft categories and matching point coordinates, and divides them into 690 training sets and 60 validation sets. The training and validation sets are used to train a multi-category matching point detection model. The model uses a multi-category top-down matching point detection algorithm to extract the two-dimensional coordinates of the aircraft target area and matching points from the surveillance images, and performs feature matching with the coordinates of preset key parts in the three-dimensional point cloud model to generate a two-dimensional and three-dimensional coordinate correspondence as input parameters of the pose solution unit. The two-dimensional coordinates of the aircraft target area and matching points are extracted from the surveillance images. The specific content is: The surveillance image is input into the target detection module, which identifies the aircraft type and generates a target frame. Based on the target frame, the Lite-HRNet network outputs the two-dimensional coordinates of the aircraft target area and matching points. The detection error is controlled within 10 pixels. The Lite-HRNet network is also denoted as a lightweight high-resolution network.

4. The scene aircraft collision warning system based on real-time positioning according to claim 1, characterized in that: The motion parameter calculation module includes: a speed calculation unit for calculating an average speed of the aircraft based on the three-dimensional coordinate differences of adjacent frames obtained from the millimeter-level real-time three-dimensional coordinates and a camera frame rate; The heading angle calculation unit uses the law of cosines to solve the relative heading angle through the coordinates of the aircraft's key points.

5. The scene aircraft collision warning system based on real-time positioning according to claim 4, characterized in that: The specific calculation formulas for the average driving speed and relative heading angle are: Among them S A is the distance from aircraft A to the collision point, S B is the distance from aircraft B to the collision point, n' and m' are time series indexes n'>m', n is the number of key points, v i is the instantaneous velocity of each point, v1,…,v n are the speeds of the 1st, ..., nth key points respectively, v is the average speed of the aircraft, K2, K1 and K'2, K1' are the key points of the two aircraft respectively, θ1, θ2 are the relative heading angles.

6. The scene aircraft collision warning system based on real-time positioning according to claim 1, characterized in that: The judgment logic of the collision type identification module is: Head-on collision: triggered when the relative heading angle difference is ≤5° and the lateral distance is less than the wingspan; Rear-end collision, when the vehicles are traveling in the same direction and the average speed difference exceeds the threshold, that is, v A ≠v B When triggered, the average speed of aircraft A is v A cm / s, the average speed of aircraft B is v B cm / s; A side collision is triggered when the relative heading angle is greater than 5° and less than 175°.

7. The scene aircraft collision warning system based on real-time positioning according to claim 1, characterized in that: The early warning model processing module includes: The safety collision avoidance time calculation unit dynamically calculates TTC by combining the driver's reaction time, brake response time, and braking duration; The collision time comparison unit compares the predicted collision time with the TTC and outputs warning information through a visual interface based on the comparison results. The warning information includes a collision type identification code, predicted collision time, TTC threshold and recommended avoidance strategy, where the identification code is mapped to head-on, rear-end and side collision types.

8. The scene aircraft collision warning system based on real-time positioning according to claim 1, characterized in that: The TTC calculation formula is: Among them, the driver's reaction time t1 = 2s, the system processing time t2, the brake response time t3 = 1.2s, the braking duration t4, the aircraft's speed v, and the gravity acceleration g = 9.8m / s 2 And the road adhesion coefficient μ, dry road μ = 0.8, wet road μ = 0.

4.

9. A scene aircraft collision warning system based on real-time positioning according to any one of claims 1 to 8, characterized in that: It also includes a multi-scenario verification module, which verifies the accuracy of warnings under different perspectives and matching point distributions based on a 1:200 scene simulation platform.

10. A scene aircraft collision warning method based on real-time positioning, characterized in that: The method for early warning of a scene aircraft collision based on real-time positioning is applicable to the method for early warning of a scene aircraft collision based on real-time positioning according to any one of claims 1 to 9, and the specific steps of the method are as follows: Step S1: Processing the surveillance image using the matching point detection unit in the positioning data acquisition module, using a multi-category top-down matching point detection algorithm, using the target detection module to identify the aircraft type and generate a target frame. Based on the target frame, the Lite-HRNet network is used to extract the aircraft target area and the two-dimensional coordinates of the matching points. The extracted two-dimensional coordinates of the matching points are feature matched with the coordinates of the preset key parts in the three-dimensional point cloud model to generate a correspondence between the two-dimensional and three-dimensional coordinates. Step S2: The pose calculation unit in the positioning data acquisition module receives the two-dimensional and three-dimensional coordinate correspondences generated in step S1 and inputs them into the SQPnP algorithm to solve the pose matrix between the three-dimensional point cloud coordinate system and the camera coordinate system. The pose matrix includes the rotation matrix R and the translation vector T. The three-dimensional point cloud is projected onto the camera coordinate system through the pose matrix to obtain the millimeter-level real-time three-dimensional coordinates of the key parts of the aircraft in the camera coordinate system. Step S3: Based on the millimeter-level real-time 3D coordinates, the 3D coordinate differences of adjacent frames are obtained. In combination with the camera frame rate, the instantaneous velocity of each key point is calculated, and then the average instantaneous velocity of all key points is calculated to obtain the average speed of the aircraft. The heading angle calculation unit uses the law of cosines to solve the relative heading angle of the two aircraft key point coordinates. Step S4: The collision type identification module receives the relative heading angle and average speed calculated in step S3 and performs a preliminary judgment of the collision type in combination with the aircraft wingspan parameters: Identify collision types based on speed and heading angle, including head-on collision, rear-end collision, and side collision; When the relative heading angle difference between two aircraft is less than or equal to 5° and the lateral distance between them is less than the wingspan of the aircraft, it is determined to be a head-on collision risk; When two aircraft are traveling in the same direction and their speed difference exceeds the set threshold, that is, v A ≠v B When the vehicle is in the rear-end collision risk range, it is determined as a rear-end collision risk; When the relative heading angle between the two aircraft is greater than 5° and less than 175°, it is determined to be a side collision risk; And output the corresponding collision type identification; In step S5, the safe collision avoidance time calculation unit in the warning model processing module calculates the TTC based on the collision type identification in step S4, combined with the driver's reaction time, brake response time and braking duration; the collision time comparison unit compares the predicted collision time with the TTC, and outputs warning information through a visual interface based on the comparison result.

Images