Airport clearance area unmanned aerial vehicle intrusion monitoring and positioning method based on improved YOLOv8n

Abstract

The invention relates to the technical field of computer vision and aviation safety, and discloses an airport clearance area unmanned aerial vehicle intrusion monitoring and positioning method based on improved YOLOv8n, and the method comprises the steps: constructing an improved YOLOv8n detection model, enhancing multi-scale feature fusion through integrating BiFPN at a neck network, and introducing an attention mechanism at a prediction head to improve feature selectivity; deploying the trained and optimized model at an intelligent sensing node, and detecting and acquiring pixel coordinates of the unmanned aerial vehicle in real time; the geographic coordinates and the attitude parameters of the nodes are fused, and the three-dimensional geographic coordinates of the unmanned aerial vehicle are calculated through space coordinate transformation in combination with a camera model and target prior information; and finally, intelligent alarm and visual display are realized. According to the method, the detection capability of the tiny unmanned aerial vehicle target under the complex background is effectively improved, an integrated solution from image recognition to precise geographic positioning is realized, and the method has the advantages of accurate detection, precise positioning and reliable system.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and aviation safety technology, and more specifically to a method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n. Background Technology

[0002] Airport airspace is a critical area for ensuring the safety of aircraft takeoffs and landings, and the core of its safety management lies in the real-time and accurate detection and handling of intruders. In recent years, the risk of intrusion by "low, slow, and small" targets brought by consumer drones has become increasingly prominent, posing a new challenge to airport operational safety.

[0003] In existing technologies, monitoring of such targets primarily relies on radar systems, radio spectrum monitoring, and optical sensors. Radar has limited detection capabilities against targets with low radar cross-sections and is susceptible to ground clutter interference, resulting in a high false alarm rate. Radio monitoring cannot cope with drones operating in pre-programmed silent flight modes, creating blind spots. In contrast, computer vision methods based on optical imagery, due to their low cost and rich information dimensions, have become the most promising complementary solution. Among them, lightweight deep learning models such as YOLOv8n are being explored for application in this field due to their good balance between speed and accuracy.

[0004] However, directly applying existing target detection models to UAV monitoring in airport airspace presents a series of unresolved technical bottlenecks, preventing them from meeting the reliability requirements of actual deployment:

[0005] First, the detection model lacks generalization and robustness. Airport environments are complex, with clouds, birds, and building edges in the background easily causing interference, while distant drone targets have a very small pixel count and weak features. The standard YOLOv8n model is inefficient in processing features at different scales during the feature fusion stage and lacks a focusing mechanism for key features, resulting in a high rate of missed and false detections of small-sized drones in complex backgrounds.

[0006] Second, the model training process is mismatched with the target characteristics. Drone targets are typically small in images, and the boundaries of their bounding boxes are inherently somewhat ambiguous. Standard loss functions exhibit instability during training when regressing these poor-quality bounding boxes, affecting the model's convergence efficiency and localization accuracy.

[0007] Third, there is a lack of effective and accurate geolocation capabilities. Existing vision-based solutions are mostly limited to outputting the pixel coordinates of the target image, and cannot directly obtain its three-dimensional location information (longitude, latitude, and altitude) in the geodetic coordinate system. The latter is crucial for air traffic control departments to conduct situational assessments and command decisions. Although there is a theory of ranging and positioning using monocular cameras, there is a lack of a high-precision, systematic solution that is tightly coupled with a high-performance detection model and is feasible in engineering.

[0008] Fourth, there is still room for optimization in system-level performance. The performance of a single model is limited by its lightweight design. How to further improve model performance without significantly increasing computational overhead, and how to integrate information from multiple sources to improve positioning accuracy and system reliability when there are multiple monitoring nodes, are all system-level problems that need to be solved.

[0009] Based on the above, this invention proposes a method for monitoring and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n to solve the aforementioned problems. Summary of the Invention

[0010] The technical problem to be solved by the present invention

[0011] To address the shortcomings of the aforementioned background technology, this invention aims to provide a method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n. It primarily solves the following technical problems:

[0012] How to improve the detection accuracy and robustness of models for small drone targets in complex sky backgrounds, and significantly reduce the false negative and false positive rates.

[0013] How to effectively integrate visual inspection results with sensor data to achieve accurate and stable conversion from image pixel coordinates to three-dimensional geographic coordinates in the geodetic coordinate system.

[0014] How can we further explore the performance potential of the model without sacrificing real-time performance, and improve the overall reliability of the solution through system-level design?

[0015] How can we implement tiered intelligent alarms based on accurate location information to provide intuitive decision support for security personnel?

[0016] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: a method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n, comprising the following steps:

[0017] S1. Constructing an UAV detection model: Based on the YOLOv8n network structure, a bidirectional feature pyramid network (BiFPN) is integrated into its neck network to perform multi-scale feature fusion, and an attention mechanism module is introduced into its prediction head to construct an improved YOLOv8n detection model.

[0018] S2. Model Training: Obtain a training dataset containing images of drones in the airport airspace and train the improved YOLOv8n detection model to obtain a trained drone detection model.

[0019] S3. Visual Perception and Preliminary Detection: Deploy at least one intelligent perception node, which is equipped with the trained UAV detection model and connected to the monitoring video stream; perform inference on the input video frame image and output the bounding box information of the UAV target and its pixel coordinates in the image;

[0020] S4. Geospatial Positioning Calculation: Obtain the geographic coordinates and attitude parameters of the intelligent sensing node; calculate the relative distance and angle of the UAV relative to the intelligent sensing node based on the pixel coordinates, camera imaging model, and prior information on the physical size of the UAV; fuse the geographic coordinates and attitude parameters of the intelligent sensing node, and calculate the three-dimensional geographic coordinates of the UAV in the geodetic coordinate system through spatial coordinate transformation.

[0021] S5. Alarm and Visualization: Based on the three-dimensional geographic coordinates and detection confidence level, generate intrusion alarm information and overlay the drone location information onto the digital map.

[0022] In step S1, the attention mechanism module introduced is a Coordinate-Attention module, which is embedded after the convolutional layer of the prediction head.

[0023] In step S1, during model training, the Wise-IoU loss function is used as the loss function for bounding box regression.

[0024] In step S2, the improved YOLOv8n detection model is trained using knowledge distillation technology. Specifically, a larger-scale pre-trained model is used as the teacher model, and the improved YOLOv8n detection model is used as the student model. The model is trained by minimizing the distillation loss between the outputs of the teacher model and the student model.

[0025] In step S4, the specific method for calculating the relative distance based on the prior information of the UAV's physical size is as follows: infer the pixel size of the UAV in the image based on the bounding box information, combine the known actual physical size of the UAV and the camera focal length, and calculate the relative distance between the UAV and the camera based on the pinhole imaging model.

[0026] As a further aspect of the present invention: the specific process of calculating the three-dimensional geographic coordinates through spatial coordinate transformation in step S4 includes:

[0027] S4.1. Transform the pixel coordinates to normalized coordinates in the camera coordinate system according to the camera intrinsic parameter matrix;

[0028] S4.2. Combine the calculated relative distance to obtain the three-dimensional coordinates of the UAV in the camera coordinate system;

[0029] S4.3. Based on the rotation matrix and translation vector formed by the attitude parameters of the intelligent sensing node, transform the three-dimensional coordinates in the camera coordinate system to the local horizontal coordinate system.

[0030] S4.4 Based on the precise geographic coordinates of the intelligent sensing node, the coordinates in the local horizontal coordinate system are transformed to the geodetic coordinate system to obtain the longitude, latitude, and altitude of the UAV.

[0031] As a further aspect of the present invention: the intelligent sensing nodes are multiple, and the method further includes:

[0032] S4.5 Multi-node data fusion: When the same UAV target is detected simultaneously by multiple intelligent sensing nodes, the preliminary three-dimensional geographic coordinates obtained by each node are calculated separately; Kalman filtering or least squares algorithm is used to fuse the multiple preliminary positioning results to obtain an optimal estimated three-dimensional geographic coordinates.

[0033] As a further aspect of the present invention: in step S5, the alarm is a multi-level dynamic alarm, and the alarm level is dynamically determined based on one or more factors among the following: the confidence level of the UAV target, the calculated altitude, the relative distance to the runway, and the movement trend.

[0034] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the program, implements the steps of the method as described in any one of the preceding claims.

[0035] A computer-readable storage medium having a computer program stored thereon, characterized in that, when executed by a processor, the program implements the steps of the method as described in any one of the preceding claims.

[0036] The technical effects and advantages of this invention are as follows:

[0037] (1) Significantly improved overall detection performance: This invention significantly enhances the model's ability to fuse multi-scale features and focus on key features by integrating a bidirectional feature pyramid network (BiFPN) into its neck network and introducing an attention mechanism into the prediction head. This improvement enables the model to more effectively capture the weak feature information of distant UAV targets with extremely small pixel ratios. At the same time, it can effectively suppress false detections when facing complex and ever-changing sky background interference, thus achieving a balance between high detection rate and low false alarm rate overall.

[0038] (2) Achieved accurate geospatial positioning: This invention breaks through the limitation of traditional visual detection that can only provide image coordinates. It deeply integrates the visual detection results with the high-precision pose sensor data of the intelligent sensing node. Through a complete set of spatial coordinate transformation chains, the pixel coordinates in the image are accurately converted into three-dimensional coordinates of the geodetic coordinate system with clear physical meaning. This provides air traffic control departments with absolute position information that can be directly used for decision-making and response, greatly enhancing the practical value of the technology.

[0039] (3) Enhanced system robustness and reliability: By adopting the knowledge distillation training strategy, the generalization performance and detection accuracy of the lightweight student model are further improved without increasing the computational load of edge devices. In addition, the multi-node data fusion mechanism supported in the scheme can optimize and fuse positioning information from different perspectives, effectively overcoming the problem of blind spots or occasional detection errors in single-node vision, making the output results of the entire monitoring system more stable and reliable.

[0040] (4) Highly intelligent and practical: This invention is not a single algorithm improvement, but a complete system solution. Its built-in multi-factor dynamic risk assessment model can make intelligent judgments based on target confidence, spatial location and movement behavior, and realize graded alarms, making the response of security personnel more targeted. Combined with the visualization of digital maps, it forms a closed loop of perception, positioning, assessment, alarm and display. The human-computer interaction interface is intuitive, the decision support capability is strong, and it is easy to deploy and apply in actual airport security systems.

[0041] (5) Maintains good real-time performance: The core detection model of this invention is based on the lightweight YOLOv8n and is optimized by knowledge distillation. This ensures that the algorithm can still maintain high-speed inference capability on edge devices with limited computing resources. The entire data processing process is efficient, and the latency from image acquisition to final coordinate output and alarm is extremely low, which fully meets the stringent real-time requirements of airport airspace security monitoring. Attached Figure Description

[0042] The invention will now be further described with reference to the accompanying drawings.

[0043] Figure 1 This is a flowchart of a method for detecting and locating unmanned aerial vehicle intrusions in airport airspace based on an improved YOLOv8n according to the present invention.

[0044] Figure 2 This is a schematic diagram of the improved MSA-YOLOv8n model structure in this invention;

[0045] Figure 3 This is a schematic diagram of the UAV positioning principle based on monocular vision and spatial coordinate transformation in this invention. Detailed Implementation

[0046] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. These embodiments are only used to explain the present invention and are not intended to limit the scope of protection of the present invention.

[0047] Example 1

[0048] Please see Figure 1 This embodiment provides a specific implementation process for a method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n:

[0049] System hardware composition

[0050] This system consists of multiple intelligent sensing nodes deployed around the airport perimeter and a central processing unit. Each intelligent sensing node includes: a high-definition network camera with a resolution of 2 megapixels or higher (supporting RTSP video stream output), a Jetson-Nano or equivalent edge computing unit, a high-precision GNSS positioning module (supporting GPS / BeiDou, with positioning accuracy better than 1 meter), and an IMU (Inertial Measurement Unit). All intelligent sensing nodes are connected to the central processing unit via gigabit Ethernet. The central processing unit is a server equipped with a high-performance GPU (such as NVIDIA-RTX-4090) and is responsible for running data fusion, precise positioning, and situation display software.

[0051] Step S1: Building and training an improved YOLOv8n drone detection model

[0052] First, an improved YOLOv8n model (hereinafter referred to as MSA-YOLOv8n) is constructed. Specifically, using the PyTorch framework, the pre-trained YOLOv8n weights released by Ultralytics are loaded as the baseline model, and two key modifications are made to its network structure:

[0053] Please see Figure 2The original PANet structure in the Neck section is replaced with BiFPN. Specifically, in the parse_model function in the models / yolo.py file, the corresponding module name is changed from PANet to BiFPN. BiFPN achieves efficient fusion of deep and shallow features through simple weighted bidirectional cross-scale connections.

[0054] After the first convolutional layer of each prediction head, a CoordAttention (CA) module is inserted. The implementation of this module can be found in the open-source code library.

[0055] For example, https: / / github.com / houqb / CoordAttention can be integrated into the YOLO architecture as a new PyTorch module;

[0056] Next, model training was performed using a self-built airport drone dataset containing over 100,000 images, covering various weather conditions, lighting, backgrounds, and different drone models. During training, a knowledge distillation strategy was employed, using the officially pre-trained YOLOv8x model as the teacher model with its weights frozen, and the previously constructed MSA-YOLOv8n model as the student model. The loss function consisted of three parts: the student model's original detection loss, the L2 loss (distillation loss) between the output feature maps of the teacher and student models, and the bounding box regression loss using Wise-IoUv1 (WIoUv1). Training was conducted on 8 GPUs for 300 epochs with a batch size of 64. After training, the model was converted to TensorRT or ONNX format to optimize inference speed for edge deployments.

[0057] Step S2: Deployment and Real-time Detection

[0058] The trained MSA-YOLOv8n model is deployed on Jetson Nano on each intelligent sensing node. Each node acquires video streams in real time through its camera (1920x1080 resolution, 25fps frame rate) and uses a TensorRT-accelerated model for real-time inference. When a UAV target is detected, the node outputs its bounding box information and pixel coordinates (u,v) in the image, and records the timestamp of the current frame. This detection result is packaged together with the node's own GNSS coordinates (longitude, latitude, altitude) and gimbal attitude angles (pitch angle, yaw angle, roll angle) measured by the IMU, and sent to the central processing unit through the network.

[0059] Step S3: Monocular visual localization solution

[0060] Please see Figure 3After the central processing unit receives data from a node, it initiates the location calculation process:

[0061] Camera calibration and coordinate normalization: The camera's intrinsic parameter matrix K and distortion coefficients D have been obtained beforehand using Zhang Zhengyou's calibration method. K and D are used to perform distortion correction on the pixel coordinates (u,v) to obtain the normalized camera coordinates (x,v). c y c );

[0062] Distance estimation: Assuming the detected drone is a common quadcopter model with typical physical dimensions (such as wheelbase) of L = 500mm, the distance D is calculated using the pinhole imaging model as follows: Where f is the camera focal length (in pixels, which can be obtained from the intrinsic parameter matrix K), and w is the width of the detection box (in pixels).

[0063] Coordinate system transformation: After calculating the distance D, combine it with the normalized coordinates (x... c y c This allows us to obtain the three-dimensional coordinates P of the UAV in the camera coordinate system. c =[X c Y c Z c ] T Next, based on the attitude angles (Euler angles) provided by the intelligent sensing nodes, the rotation matrix R from the camera frame to the carrier frame is calculated. bc Translation vector T bc (This parameter is determined by the installation relationship between the camera and the IMU and has been obtained in advance through joint calibration.) Then, using the pose provided by GNSS and IMU data, the carrier's system coordinates are transformed to the local Northeastern Sky (ENU) coordinate system. Finally, the ENU coordinates are transformed to the WGS-84 geodetic coordinate system through Gauss-Kruger projection or UTM projection, and the accurate latitude, longitude and altitude of the UAV are finally obtained.

[0064] Step S4: Multi-node data fusion and dynamic alarm

[0065] If multiple intelligent sensing nodes in the system detect the same UAV target at the same time, the central processing unit will initiate a data fusion process and use the extended Kalman filter (EKF) algorithm to fuse the positioning results provided by each node (considered as an observation value) with the target motion model to obtain an optimal estimated three-dimensional geographic coordinate, which significantly improves positioning accuracy and stability.

[0066] The system has a three-level dynamic alarm mechanism. The central processing unit compares the calculated UAV coordinates with the airport's electronic fence (geographic boundary in GIS) in real time. At the same time, it comprehensively considers the detection confidence, altitude, distance from the runway, and motion vector (speed and direction). For example, if a UAV with a confidence > 0.9, an altitude below 200 meters, and is flying towards the runway at a speed greater than 10 m / s, it will trigger the highest level "emergency" alarm, which will be displayed as a flashing red icon on the monitoring screen and will also alert the control tower through audio-visual equipment.

[0067] Step S5: Dynamic Risk Assessment, Alerts, and Visualization

[0068] The central processing unit performs dynamic risk assessment on the calculated coordinates of each drone and triggers corresponding alarms and visualizations. The specific implementation process is as follows:

[0069] Dynamic Risk Assessment: The system has a built-in risk assessment model that calculates the comprehensive risk value R for each drone target in real time. score R score It is a function composed of multiple weighting factors:

[0070] Confidence factor (C): The detection confidence (0-1) from the output of the MSA-YOLOv8n model. The higher the confidence, the greater the probability that the target is a real drone, and the higher the risk base value.

[0071] Spatial location factor (P): Based on the UAV's geodetic coordinates, calculate its Euclidean distance to key areas such as the runway centerline, takeoff / landing routes, and airport boundaries. The closer the distance, the higher the weight.

[0072] Altitude Factor (A): The current altitude of the drone. The lower the altitude, especially when it is below the altitude of the airport tower or approach path, the greater the risk weight.

[0073] Motion characteristic factor (M): The instantaneous speed and direction of motion of the UAV are calculated by continuous frame positioning data. The faster the speed and the direction is directly towards the core of the runway or no-fly zone, the greater the risk weight.

[0074] The overall risk value can be simplified to: R score =·C+β·P+γ·A+δ·M (where α, β, γ, and δ are weighting coefficients adjusted according to the actual airport security strategy), and the system presets three risk level thresholds: low, medium, and high (e.g., R). score <0.3 indicates low risk, 0.3≤R score <0.7 indicates medium risk, R score ≥0.7 indicates high risk;

[0075] Multi-level alarm triggering: based on the calculated Rscore Depending on the level, the system will automatically trigger alarms of different levels.

[0076] Low-level alarm: When a drone intrudes into the airspace but the risk is low (such as flying at a long distance, high altitude, stationary or away from the runway), the system records it in the background log and displays it as a green entry in the monitoring software list;

[0077] Intermediate alarm: When the risk value is intermediate, the system will trigger an audio-visual alarm. The drone icon in the monitoring software interface will turn yellow and flash, and a pop-up reminder will be sent to the tower controller's workstation.

[0078] Advanced Alarm: When the risk value reaches the advanced level (such as low altitude, high speed, straight runway), the system triggers the highest level alarm. The drone icon turns red and flashes violently in the monitoring interface. A high-decibel alarm sounds in the airport control room. At the same time, the system automatically sends an emergency alarm message to the safety manager through preset communication channels (such as SMS, internal communication system). The message includes the drone's location, speed, estimated trajectory and risk level.

[0079] Geographic information visualization: All data is integrated and displayed on the visualization platform of the central processing unit, which is deeply integrated with the airport digital map (GIS) system;

[0080] The map clearly marks the runway, control tower, buildings, and electronic fence boundaries;

[0081] Each intelligent sensing node is displayed on the map with its actual geographic coordinates;

[0082] Detected drones will be displayed in real time as icons on their latitude and longitude coordinates, and the icon color will change dynamically (green / yellow / red) according to their current risk level;

[0083] The system will draw the historical movement trajectory lines of each drone, and the trajectory color will also reflect the changes in historical risk level, which will facilitate the tracing and analysis of intrusion paths;

[0084] Clicking on any drone icon will bring up a details window, displaying all information such as a real-time screenshot, precise coordinates, altitude, speed, confidence level, and risk value.

[0085] Example 2

[0086] In another implementation, to cope with nighttime or low-light environments, the ordinary optical camera in the intelligent sensing node is replaced with an infrared thermal imaging camera. The main difference between this embodiment and Embodiment 1 is:

[0087] The MSA-YOLOv8n model needs to be fine-tuned using a specially collected infrared UAV dataset to adapt it to the characteristics of thermal imaging targets.

[0088] The hardware configuration and subsequent positioning calculation, data fusion, and alarm visualization processes are exactly the same as in Example 1.

[0089] Example 3

[0090] In another implementation, to reduce system costs, a binocular vision solution can be used instead of a monocular solution that relies on prior knowledge of physical dimensions. The main difference between this embodiment and Embodiment 1 is:

[0091] Each intelligent sensing node needs to be equipped with two cameras that have been strictly synchronized and calibrated to form a binocular vision system;

[0092] During positioning, the drone targets detected by the left and right cameras are first paired, and then the target distance is directly calculated by parallax based on the principle of stereo vision, without needing to know the physical size of the drone.

[0093] The subsequent coordinate transformation, data fusion, and alarm visualization processes are the same as in Example 1.

[0094] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n, characterized in that, Includes the following steps: S1. Constructing an UAV detection model: Based on the YOLOv8n network structure, a bidirectional feature pyramid network (BiFPN) is integrated into its neck network to perform multi-scale feature fusion, and an attention mechanism module is introduced into its prediction head to construct an improved YOLOv8n detection model. S2. Model Training: Obtain a training dataset containing images of drones in the airport airspace and train the improved YOLOv8n detection model to obtain a trained drone detection model. S3. Visual Perception and Preliminary Detection: Deploy at least one intelligent perception node, which is equipped with the trained UAV detection model and connected to the monitoring video stream; perform inference on the input video frame image and output the bounding box information of the UAV target and its pixel coordinates in the image; S4. Geospatial Positioning Calculation: Obtain the geographic coordinates and attitude parameters of the intelligent sensing node; calculate the relative distance and angle of the UAV relative to the intelligent sensing node based on the pixel coordinates, camera imaging model, and prior information on the physical size of the UAV; fuse the geographic coordinates and attitude parameters of the intelligent sensing node, and calculate the three-dimensional geographic coordinates of the UAV in the geodetic coordinate system through spatial coordinate transformation. S5. Alarm and Visualization: Based on the three-dimensional geographic coordinates and detection confidence level, generate intrusion alarm information and overlay the drone location information onto the digital map.

2. The method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n, as described in claim 1, is characterized in that: In step S1, the attention mechanism module introduced is a Coordinate-Attention module, which is embedded after the convolutional layer of the prediction head.

3. A method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n, as described in claim 1 or 2, characterized in that: In step S1, during model training, the Wise-IoU loss function is used as the loss function for bounding box regression.

4. The method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n, as described in claim 1, is characterized in that: In step S2, the improved YOLOv8n detection model is trained using knowledge distillation technology. Specifically, a larger-scale pre-trained model is used as the teacher model, and the improved YOLOv8n detection model is used as the student model. The model is trained by minimizing the distillation loss between the outputs of the teacher model and the student model.

5. The method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n, as described in claim 1, is characterized in that: In step S4, the specific method for calculating the relative distance based on the prior information of the UAV's physical size is as follows: infer the pixel size of the UAV in the image based on the bounding box information, combine the known actual physical size of the UAV and the camera focal length, and calculate the relative distance between the UAV and the camera based on the pinhole imaging model.

6. The method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n, as described in claim 1, is characterized in that: In step S4, the specific process of calculating the three-dimensional geographic coordinates through spatial coordinate transformation includes: S4.

1. Transform the pixel coordinates to normalized coordinates in the camera coordinate system according to the camera intrinsic parameter matrix; S4.

2. Combine the calculated relative distance to obtain the three-dimensional coordinates of the UAV in the camera coordinate system; S4.

3. Based on the rotation matrix and translation vector formed by the attitude parameters of the intelligent sensing node, transform the three-dimensional coordinates in the camera coordinate system to the local horizontal coordinate system. S4.4 Based on the precise geographic coordinates of the intelligent sensing node, the coordinates in the local horizontal coordinate system are transformed to the geodetic coordinate system to obtain the longitude, latitude, and altitude of the UAV.

7. The method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n, as described in claim 6, is characterized in that: The intelligent sensing nodes are multiple, and the method further includes: S4.5 Multi-node data fusion: When the same UAV target is detected simultaneously by multiple intelligent sensing nodes, the preliminary three-dimensional geographic coordinates obtained by each node are calculated separately; Kalman filtering or least squares algorithm is used to fuse the multiple preliminary positioning results to obtain an optimal estimated three-dimensional geographic coordinates.

8. The method for detecting and locating unmanned aerial vehicle (UAV) intrusions into airport airspace based on an improved YOLOv8n, as described in claim 1, is characterized in that: In step S5, the alarm is a multi-level dynamic alarm, and the alarm level is dynamically determined based on one or more factors among the following: the confidence level of the UAV target, the calculated altitude, the relative distance to the runway, and the movement trend.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images