Method for identifying and analyzing bird flock geolocation using high point cameras

Abstract

The present application relates to the technical field of image data processing and object positioning and identification, in particular to a method for identifying and analyzing the geographic position of bird flocks by high-definition cameras, comprising: deploying multiple high-definition cameras, collecting video streams in real time and synchronously recording the spatial position parameters and attitude parameters of each camera itself; using an improved YOLOv8 model to detect bird targets in the video; the improvement includes introducing P2 layer feature maps from Backbone and adding corresponding P2 detection heads in the detection head, while removing the P5 detection head. By improving the YOLOv8 model, the present application introduces the P2 detection head to enhance the small target recognition capability, and fuses the observation angles of multiple cameras, uses spatial geometric solution and dynamic clustering technology, and finally outputs the accurate three-dimensional coordinates of the bird flock in the real world. While significantly improving the positioning accuracy and recognition rate, the present application has the advantages of strong system robustness, low deployment cost and easy expansion.

Description

Technical Field

[0001] This invention relates to the fields of image data processing technology and object positioning and recognition technology, specifically a method for identifying and analyzing the geographical location of bird flocks using a high-point camera. Background Technology

[0002] With the rapid development of human economy and society and the continuous improvement of the ecological environment, new airports are constantly being built, and the number and frequency of various aircraft are constantly increasing. At the same time, the number of birds is also constantly increasing. As a result, there is a fierce space encroachment situation between aircraft and birds, and the flight risks are constantly rising.

[0003] For example, patent CN116630897A discloses an intelligent auxiliary control system for airport bird control based on image recognition, which includes: classifying birds during flight hours into suspected birds and normal birds; obtaining the fanning frequency and warning frequency of suspected birds; obtaining the credibility of suspected birds based on the difference between the fanning frequency of suspected birds and the fanning frequency of normal birds, as well as the difference between the warning frequency of suspected birds and the warning frequency of normal birds; identifying untrusted birds among the suspected birds based on the credibility of the suspected birds; obtaining the driving priority of each type of bird based on the number of times untrusted birds appear in each type of bird and the total number of each type of bird; and performing bird driving control based on the driving priority of each type of bird.

[0004] In practical use, the aforementioned technology only identifies and classifies the danger level of birds and only performs driving-away operations on birds with higher danger levels. However, its key drawback is that it fails to output the specific spatial location of birds (such as latitude, longitude, and altitude) in real time through image recognition technology. This results in a lack of accurate target positioning basis for subsequent bird driving operations, making it impossible to determine the real-time activity area of ​​high-risk birds or plan efficient driving-away paths. Ultimately, this significantly restricts the implementation of actual prevention and control effects and weakens the practical value of the technology in airport bird strike prevention scenarios. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying and analyzing the geographical location of bird flocks using high-point cameras, in order to solve the problem that the inability to identify the specific spatial location of birds reduces the practical value of the technology in airport bird strike prevention scenarios.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying and analyzing the geographical location of bird flocks using a high-point camera, comprising the following steps:

[0007] S1. Data Acquisition and Transmission: Deploy several high-definition cameras around the monitoring area to collect video streams in real time and simultaneously record the spatial position parameters and attitude parameters of each camera, and send the data to the processing terminal.

[0008] S2. Feature Extraction and Target Detection: An improved YOLOv8 model is used to detect birds in the video, identify the birds in the image and obtain their pixel coordinates; the improvement includes introducing the P2 layer feature map from the Backbone into the Neck part of the model, adding a corresponding P2 detection head to the detection head, and removing the P5 detection head;

[0009] S3. Bird relative angle calculation: Based on the spatial position parameters and attitude parameters in step S1 and the bird pixel coordinates in step S2, calculate the azimuth and pitch angles of the bird target relative to the camera.

[0010] S4. Constructing the spatial observation line equation: Establish a spatial rectangular coordinate system for the camera and bird positions. Combine the angle of each camera to the bird target calculated in step S3 to establish a spatial line equation from each camera to the corresponding bird target.

[0011] S5. Approximation and Filtering of Intersection Points of Skew Lines: Calculate the minimum skew distance D between every two lines in all spatial line equations in step S4, and simultaneously calculate the coordinates P of the center point of the common perpendicular line where the minimum distance is located; set a distance threshold and discard data where the minimum skew distance D is greater than the distance threshold.

[0012] S6. Dynamic clustering and location determination: The dynamic clustering algorithm is used to cluster the set A composed of all the centroids obtained in step S5 to form multiple subsets, and the subsets with fewer than the preset threshold number of elements are discarded.

[0013] S7. Coordinate Output: Convert the position coordinates in the subset obtained in step S6 into geodetic coordinates and output them.

[0014] Preferably, in step S1, the spatial position parameters of the high-definition camera include longitude, latitude, and altitude, and the attitude parameters include azimuth and pitch angles.

[0015] Preferably, in step S3, the azimuth angle of the bird target relative to the camera... and pitch angle The calculation formula is:

[0016] ;

[0017] ;

[0018] ;

[0019] ;

[0020] in, The pitch angle of the bird relative to the center line of the camera. The azimuth angle of the bird relative to the center line of the camera. This refers to the camera's own orientation and angle. The camera's own tilt angle. This refers to the absolute azimuth angle of the bird. The absolute pitch angle for birds; The width of the video image in pixels. X represents the pixel height of the video image, X and Y represent the pixel coordinates of the bird's center point in the video, HOV represents the horizontal field of view of the camera, and VOV represents the vertical field of view.

[0021] Preferably, the specific calculation formula for the spatial straight line equation L of each camera pointing to the corresponding bird target in step S4 is as follows:

[0022] ;

[0023] ;

[0024] ;

[0025] ;

[0026] The spatial coordinates of the airport reference point are ( The spatial coordinates of the camera are ( ); ); ( () represents the Cartesian coordinates of the camera; The coordinates of the bird are Cartesian coordinates in a coordinate system with the airport reference point as the origin. This term is unknown and has the aforementioned constraints.

[0027] Preferably, in step S5, the minimum distance between any two straight lines is... The calculation formula is:

[0028] ;

[0029] in and It is the vector corresponding to two skew lines. It is a vector formed by the line connecting the two cameras.

[0030] Preferably, the formula for calculating the coordinates P of the center point of the common perpendicular line where the minimum distance is located in step S5 is as follows:

[0031] ;

[0032] Where A0 ( ) and B0 ( ) are the coordinates of the perpendicular feet on the two skew lines, respectively.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] This invention achieves lightweight optimization of the YOLOv8 model by removing the P5 large target detection head and adding a P2 small target detection head. While avoiding computational redundancy and ensuring training stability and real-time monitoring latency requirements, it significantly improves the accuracy of small target bird flock recognition. Relying on the collaborative deployment of multiple cameras, a unified spatial rectangular coordinate system is established. By combining the calculation of skew line distances, the solution of the center point of the common perpendicular, and dynamic clustering analysis, the absolute latitude, longitude, and altitude information of birds can be accurately calculated, filling the gap in existing technologies that only identify but do not locate. At the same time, it reuses the existing high-point camera network of communication base stations, selects conventional high-definition cameras, and can flexibly adjust the number of cameras, greatly reducing deployment costs and adapting to different airport security needs. It comprehensively solves the pain points of existing technologies in terms of accuracy, real-time performance, positioning capability, and economy. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the overall process of the method for identifying and analyzing the geographical location of bird flocks using a high-point camera according to the present invention;

[0036] Figure 2 This is a schematic diagram of the dynamic clustering algorithm flow for the method of identifying and analyzing the geographical location of bird flocks using a high-point camera according to the present invention;

[0037] Figure 3 This is a schematic diagram of the improved YOLOv8 recognition algorithm for the method of identifying and analyzing the geographical location of bird flocks using a high-point camera, as described in this invention.

[0038] Figure 4 The method for identifying and analyzing the geographical location of bird flocks using high-point cameras in this invention is illustrated in the diagram of camera distribution and viewing angle planning.

[0039] Figure 5 This is a schematic diagram showing the comparison of the number of birds identified hourly from the 28th to the 30th of this invention.

[0040] Figure 6 This is a schematic diagram of the flight trajectory of the bird target identified on the 28th in an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Faced with the severe risk of bird strikes, relying on single-camera image recognition methods for bird target identification is not very accurate. Furthermore, limitations in recognition algorithm performance prevent bird location, primarily due to the large number of birds and their flight postures. To address the issue of low recognition efficiency, in addition to the image recognition-based intelligent auxiliary control system for airport bird control disclosed in patent CN116630897A mentioned above, existing technical solutions include:

[0043] The YOLO-birds algorithm, based on YOLOv8, replaces the PANet fusion network in the Neck layer with BiFPN. It introduces an EIoU bounding box regression loss function, calculates the difference between width and height to replace the aspect ratio based on CIoU, and introduces FocalLoss.

[0044] The airport bird target detection algorithm based on the improved YOLOv8s addresses the issue that since the camera used for detection is in a fixed position while birds fly at different altitudes, the target not only varies greatly in scale but also suffers from motion blur when there are many dense targets. Therefore, the improved algorithm adds ODConv full-dimensional dynamic convolution to the YOLOv8 algorithm.

[0045] Deep learning-based image recognition systems: These systems typically use convolutional neural networks (CNNs) to process and analyze image data to improve the accuracy of bird flock detection. For example, some systems use ResNet or Inception architectures instead of YOLOv8 to achieve higher detection accuracy.

[0046] However, the above technology still has the following drawbacks when used:

[0047] Increased complexity: While replacing the fusion network improves the feature fusion effect, it also increases computational complexity, which may lead to a decrease in real-time performance; introducing a new regression loss function improves detection accuracy, but may also lead to instability in the training process.

[0048] Bird absolute coordinate positioning: While receiving and parsing image information through a single data source (camera) enables bird identification and determines the pixel position of the bird in the image, it cannot resolve the bird's position in the WGS-84 space. Therefore, it limits the further use of bird identification information, such as bird avoidance and bird scare, and thus still poses a certain risk to aircraft operation.

[0049] Please see Figure 1-6 This invention provides a technical solution: a method for identifying and analyzing the geographical location of bird flocks using a high-point camera, comprising the following steps:

[0050] S1. Data Acquisition and Transmission: Based on the characteristics of bird activity around the airport, monitoring areas are divided, and multiple high-definition cameras are deployed around the monitoring areas. Each camera covers the bird activity hotspots at a preset azimuth angle, acquiring video streams in real time and simultaneously recording the spatial position and attitude parameters of each camera. The spatial position parameters include longitude, latitude, and altitude in the GNSS coordinate system, and the video stream includes a timestamp T (accurate to milliseconds). The attitude parameters include (α, β) (horizontal azimuth α∈-22.5°, 22.5°, pitch β∈-17.5°, 17.5°). The video data from multiple cameras and the camera's own image acquisition parameter data are sent to the edge fusion device. A data buffer is set up at the edge fusion device to temporarily store the received video stream data, ensuring the continuity and stability of data transmission.

[0051] S2. Feature Extraction and Target Detection: An improved YOLOv8 model is used to detect birds in the video, identify the birds in the image and obtain their pixel coordinates. The improvement includes introducing the P2 layer feature map from the Backbone into the Neck part of the model and adding a corresponding P2 detection head to the detection head. Its predicted feature map size is 160×160, which can better extract such small targets and preserve the position and feature information of small targets.

[0052] At the same time, removing the P5 detection head reduces the computational load of the model and improves the target recognition rate;

[0053] The final output feature map contains information from different scales, which can better adapt to target detection tasks of different sizes;

[0054] S3. Bird Relative Angle Calculation: Based on the spatial position and attitude parameters from step S1 and the bird's pixel coordinates from step S2, and considering the bird's center point's pixel position (X, Y) in the video, combined with the camera's horizontal field of view (HOV), vertical field of view (VOV), and the camera's own pitch angle... and azimuth angle The azimuth and pitch angles of the bird target relative to the camera were calculated. and pitch angle The calculation formula is:

[0055] The pitch angle of the bird relative to the center line of the camera is:

[0056] ;

[0057] The orientation angle relative to the center line of the camera is:

[0058] ;

[0059] By superimposing the camera's own azimuth and pitch angles, the bird's pitch angle relative to the camera can be obtained. and azimuth :

[0060] ;

[0061] ;

[0062] in, The pitch angle of the bird relative to the center line of the camera. The azimuth angle of the bird relative to the center line of the camera. This refers to the camera's own orientation and angle. The camera's own tilt angle. This refers to the absolute azimuth angle of the bird. The absolute pitch angle for birds; The width of the video image in pixels. X represents the pixel height of the video image, X and Y represent the pixel coordinates of the bird's center point in the video, HOV represents the horizontal field of view of the camera, and VOV represents the vertical field of view.

[0063] S4. Constructing the spatial observation line equation: Establish a spatial rectangular coordinate system for the camera and bird positions. Combining the angle of each camera to the bird target calculated in step S3, construct a spatial line equation pointing from each camera to the corresponding bird target, where the latitude, longitude, and altitude of the airport reference point are ( ). The latitude and longitude of the camera are ( ). Using the airport reference point as the center and the northeast-eastern sky as the XYZ axis, within a small area (less than 1 kilometer), assuming the Earth is a plane, and based on the Earth's average radius... The position of the camera in the Cartesian coordinate system can be calculated. )for:

[0064] ;

[0065] ;

[0066] ;

[0067] The equation of the straight line formed by the bird target and the camera is:

[0068] ;

[0069] The spatial coordinates of the airport reference point are ( The spatial coordinates of the camera are ( ); ); ( () represents the Cartesian coordinates of the camera; For birds, the Cartesian coordinates are given in a coordinate system with the airport reference point as the origin. This term is unknown and has the aforementioned constraints.

[0070] S5. Approximation and Filtering of Intersection Points of Skew Lines: For each bird location detected by the camera, there exists a three-dimensional spatial location as described in step S4. Calculate the minimum skew distance D between every two lines in all spatial line equations of step S4, and simultaneously calculate the coordinates P(X1, Y1, Z1) of the center point of the common perpendicular line containing the minimum distance. The calculation formula is:

[0071] ;

[0072] in and It is the vector corresponding to two skew lines. It is the vector formed by the line connecting the two cameras;

[0073] When D exceeds the set threshold, it is determined that the two lines above point to different individuals, and the calculation of the other two sets of lines continues; otherwise, the following steps are performed.

[0074] Any point on the common perpendicular can be represented as:

[0075] or ;

[0076] Based on the shortest distance condition (the common perpendicular is perpendicular to both lines):

[0077] ;

[0078] Solving for the results ; and then obtain .

[0079] Therefore, the coordinates of the midpoint of the common perpendicular are: ;

[0080] Where A0 ( ) and B0 ( ( ) are the coordinates of the perpendicular feet on the two skew lines, respectively;

[0081] Set a distance threshold and discard data whose minimum distance between opposite surfaces is greater than the distance threshold;

[0082] S6. Dynamic Clustering and Location Determination: A dynamic clustering algorithm is used to cluster the set A consisting of all P points obtained in step S5, forming multiple subsets. Subsets with fewer than a preset threshold of elements are discarded. The centroid of the sum of each subset is calculated, which represents the location of each bird target.

[0083] S7. Coordinate Output: Convert the position coordinates obtained in step S6 into geodetic coordinates and output them.

[0084] The present invention also provides a specific embodiment, as shown below:

[0085] This embodiment underwent field testing at a certain location from June 28, 2025 to June 30, 2025. Figure 4 As shown, cameras were installed at a height of 10 meters above the ground, with a total of 10 stations. Each station had 6 cameras, with a horizontal field of view of 55° and a vertical field of view of 33°. The 6 cameras were networked through a 4G network. After the cameras identified the location information of the birds, they were transmitted to the edge fusion device via the 4G network. The edge fusion device then calculated and displayed the absolute position of the target.

[0086] The position and orientation of each camera are shown in Table 1 below:

[0087] Table 1: Camera Position and Azimuth

[0088]

[0089] Bird target location information is shown in Tables 2 and 3: Table 2: Bird Target 1 ( Figure 6 Location information of the red track in the middle

[0090]

[0091] Table 3: Bird Targets 2 ( Figure 6 Location information (yellow track in the middle)

[0092]

[0093] Tables 2 and 3 contain 55 and 37 sets of "time-longitude-latitude" data, respectively. For example, bird target 1 has longitude of 139.3774791 and latitude of 35.09587967 at 07:43:21.318 and longitude of 139.3718786 and latitude of 35.10157481 at 07:46:03. The absolute coordinates of the bird in WGS-84 space are directly quantified and output, completely solving the blind spot of "no location information" in prevention and control.

[0094] Images of flocks of birds around the airport were captured using high-point cameras. An improved YOLOv8 image recognition algorithm was run on an NVIDIA Jetson Xavier NX environment. The training set contained 50,000 labeled bird flock images covering 10 common bird species, and the test set consisted of 5,000 real-time acquired images. Preprocessing was performed, including image enhancement and noise reduction, to improve image quality.

[0095] Tests conducted from June 28th to June 30th, 2025, showed that while meeting processing speed and latency requirements, the model improved the accuracy of bird identification by approximately 11.6% compared to the traditional YOLOv8 algorithm. It also demonstrated a significant improvement in the detection of small targets. The comparison results are shown below. Figure 5 As shown, Figure 5 The horizontal axis represents time (0-70 hours), and the vertical axis represents the number of recognitions. The improved YOLOv8 recognized more animals than the original YOLOv8 throughout the entire time period, with an average improvement of about 11.6% (e.g., the improved version recognized 15 animals in 10 hours, while the original version only recognized 12). Moreover, this data is based on testing with NVIDIA Jetson Xavier NX edge devices, proving that high accuracy and real-time processing can be achieved at the edge, solving the problem of the trade-off between accuracy and efficiency.

[0096] pass Figure 3 The model optimization logic can be displayed intuitively. Specifically, the P5 large target detection head is removed (to reduce redundant calculations), and a new P2 detection head (160×160 high resolution, suitable for small bird targets) is added. Only the three types of detection heads, P2, P3, and P4, are retained. This reduces the amount of computation from a structural perspective and lays the foundation for real-time performance.

[0097] Figure 6 The image shows two bird flight paths that were identified. The overall identification process achieved good results in terms of horizontal accuracy, trajectory continuity, and smoothness. The trajectory lines are smooth and without breaks, proving that the present invention can track bird movement paths in real time, rather than just identifying "whether there is danger" and can meet the bird control needs around airports.

[0098] This invention optimizes the YOLOv8 model by adding a P2 detector head for detecting small targets. Its predicted feature map size is 160×160, providing higher resolution. The elimination of the P5 detector head ensures detection efficiency. Furthermore, the bird target recognition results from multiple cameras are converted to the same coordinate system with the Airport Reference Point (ARP) as the origin. Spatial location calculation and dynamic clustering analysis algorithms are used to derive the absolute spatial location of the birds, enabling absolute bird flock positioning. This provides strong technical support for bird flock monitoring and control, effectively solving the problem of conventional single-camera positioning being insufficient. Compared to conventional binocular camera deployment, this invention leverages the existing high-point camera network built by domestic communication base stations, avoiding the cost of secondary deployment. It also uses conventional cameras to reduce deployment costs. Moreover, the number of cameras can be adjusted according to different bird flock control requirements around the airport to improve the bird flock detection and positioning capabilities in key areas.

[0099] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0100] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for identifying and analyzing the geographical location of bird flocks using a high-point camera, characterized by: Includes the following steps: S1. Data Acquisition and Transmission: Deploy several cameras around the monitoring area to collect video streams in real time and simultaneously record the spatial position parameters and attitude parameters of each camera, and send the spatial position parameters and attitude parameters to the processing terminal. S2. Feature Extraction and Target Detection: An improved YOLOv8 model is used to detect birds in the video, identify the birds in the image and obtain their pixel coordinates; the improvement includes introducing the P2 layer feature map from the Backbone into the Neck part of the model, adding a corresponding P2 detection head to the detection head, and removing the P5 detection head; S3. Bird relative angle calculation: Based on the spatial position parameters and attitude parameters in step S1 and the bird pixel coordinates in step S2, calculate the azimuth and pitch angles of the bird target relative to the camera. S4. Constructing the spatial observation line equation: Establish a spatial rectangular coordinate system for the camera and bird positions. Combine the azimuth and elevation angles of each camera to the bird target calculated in step S3 to establish a spatial line equation from each camera to the corresponding bird target. S5. Approximation and selection of intersection points of skew lines: Calculate the minimum skew distance D between every two lines in all spatial line equations in step S4, and at the same time calculate the coordinates P of the center point of the common perpendicular line where the minimum skew distance D is located. Set a distance threshold and discard data where the minimum distance D between opposite surfaces is greater than the distance threshold; S6. Dynamic clustering and location determination: The dynamic clustering algorithm is used to cluster the set A composed of all the centroids obtained in step S5 to form multiple subsets, and the subsets with fewer than the preset threshold number of elements are discarded. S7. Coordinate Output: Convert the position coordinates in the subset obtained in step S6 into geodetic coordinates and output them.

2. The method for identifying and analyzing the geographical location of bird flocks using a high-point camera according to claim 1, characterized in that: In step S1, the spatial position parameters of the camera include longitude, latitude, and altitude, and the attitude parameters include azimuth and pitch angles.

3. The method for identifying and analyzing the geographical location of bird flocks using a high-point camera according to claim 2, characterized in that: In step S3, the bird target's azimuth relative to the camera and pitch angle The calculation formula is: ; ; ; ; in, The pitch angle of the bird relative to the center line of the camera. The azimuth angle of the bird relative to the center line of the camera. This refers to the camera's own orientation and angle. The camera's own tilt angle. This refers to the absolute azimuth angle of the bird. The absolute pitch angle for birds. The width of the video image in pixels. X represents the pixel height of the video image, X and Y represent the pixel coordinates of the bird's center point in the video, HOV represents the horizontal field of view of the camera, and VOV represents the vertical field of view.

4. The method for identifying and analyzing the geographical location of bird flocks using a high-point camera according to claim 3, characterized in that: The specific calculation formula for the spatial straight line equation L of each camera pointing to the corresponding bird target in step S4 is as follows: ; ; ; ; The spatial coordinates of the airport reference point are ( The spatial coordinates of the camera are ( ); ); ( The coordinates of the camera are Cartesian coordinates in a coordinate system with the airport reference point as the origin. This refers to the absolute azimuth angle of the bird. The absolute pitch angle for birds; The coordinates of the bird are Cartesian coordinates in a coordinate system with the airport reference point as the origin. This term is unknown and has the aforementioned constraints.

5. The method for identifying and analyzing the geographical location of bird flocks using a high-point camera according to claim 4, characterized in that: In step S5, the minimum distance between every two straight lines is skew. The calculation formula is: ; in and It is the vector corresponding to two skew lines. It is a vector formed by the line connecting the two cameras.

6. The method for identifying and analyzing the geographical location of bird flocks using a high-point camera according to claim 5, characterized in that: The formula for calculating the coordinates P of the center point of the common perpendicular line containing the minimum distance in step S5 is as follows: ; Where A0 ( ) and B0 ( ) are the coordinates of the perpendicular feet on the two skew lines, respectively.

Images