A vision-based ultrasonic directional bird-repelling drone device and bird-repelling method

Abstract

The application discloses a visual-based ultrasonic directional bird repelling unmanned aerial vehicle device and a bird repelling method thereof. The ultrasonic directional bird repelling unmanned aerial vehicle device comprises a four-rotor unmanned aerial vehicle and a visual detection directional bird repelling module installed on the unmanned aerial vehicle. The visual detection directional bird repelling module comprises a master control module, a holder module, a holder attitude measurement module, a visual detection module, an ultrasonic module and an operation processing module. The bird repelling method using the device comprises the following steps: after obtaining bird images, performing convolutional neural network processing on the obtained images, distinguishing harmful birds and harmless birds, obtaining feature points of the birds, mapping the feature points of a two-dimensional plane to a three-dimensional space, estimating the coordinates of the birds in the three-dimensional space, estimating and solving the motion of the birds according to the change rule of the three-dimensional coordinates of the birds, obtaining the flight track of the birds, controlling the holder module, making the holder module face the motion direction of the birds and performing ultrasonic emission.

Description

Technical Field

[0001] The present invention belongs to the technical field of ultrasonic bird repelling technology and relates to a vision-based ultrasonic directional bird repelling drone device and a bird repelling method thereof. Background Art

[0002] With the continuous development of my country's social economy, the impact of birds on farmland has become increasingly serious. According to a survey by the "Granivorous Bird Working Group" of the International Biological Program (IBP), mountain birds around the world can consume 15 million tons of food in a year, enough to feed the world's 90 million people.

[0003] The history of bird repelling is also very long. The classic bird repelling methods mainly include sound repelling and visual repelling. The most common ones are as follows:

[0004] 1. Manual bird repelling: Birds are more harmful in the early morning, midday, and dusk. You can arrive at the fields in advance and repel the birds immediately. After 15 minutes, you should check and repel the birds again. Generally, you need to repel the birds 3-5 times during each period. This method is time-consuming and labor-intensive.

[0005] 2. Sound bird repellent: Record sounds such as hammering, firecrackers, hawk calls, and bird cries, and play them at high volume throughout the farmland at irregular intervals to deter birds. Sound devices should be placed around crops and at bird entry points to amplify the sound through wind direction and echoes. However, this will have a significant impact on nearby residents.

[0006] 3. Place objects to repel birds: Place dummies, fake eagles, etc. in the fields. This can prevent the invasion of harmful birds in the short term, but it will become ineffective in the long term.

[0007] Patent publication number CN107372455A "A kind of agricultural bird-repelling machinery" only uses the original bird-repelling machinery, which can only repel the surrounding birds, and it is difficult to repel the birds within the range.

[0008] Patent publication number CN112640884A, "An airport bird-repelling device and method thereof," uses a manually remote-controlled drone and uses a net-bomb method after discovering a bird. The net-bomb method is prone to causing harm to birds, and ammunition needs to be replenished after each use, making it impossible to repel birds continuously. Summary of the Invention

[0009] The purpose of the present invention is to provide a vision-based ultrasonic directional bird-repellent aerial robot, which can be used to solve the maintenance of related buildings such as urban relics, the protection of related power facilities and equipment, agriculture and other work scenarios related to bird-repellent needs, and reduce physiological damage to birds.

[0010] A first aspect of the present invention provides a vision-based ultrasonic directional bird-repelling drone device, comprising a quadrotor drone and a visual detection and directional bird-repelling module mounted on the drone; the visual detection and directional bird-repelling module comprises a main control module, a pan-tilt module, a pan-tilt attitude measurement module, a visual detection module, an ultrasonic module, and an operation processing module;

[0011] The pan / tilt module is connected to the main control module and can realize pitch and yaw movements;

[0012] The pan-tilt posture measurement module is connected to the calculation processing module and is used to accurately locate the orientation of the pan-tilt module;

[0013] The visual detection module is connected to the calculation processing module and is used to capture images of birds in high-speed motion and estimate the distance of the birds, accurately obtaining the position of the birds in the camera coordinate system;

[0014] The ultrasonic module is connected to the main control module and is used for directional bird repelling.

[0015] A second aspect of the present invention provides a method for repelling birds using the above-mentioned device, comprising the following steps:

[0016] Step 1: After the staff has planned the designated route, the quadcopter will cruise at a specified altitude;

[0017] Step 2: The gimbal module regularly pitches and yaws within a limited range to expand the range that the visual detection module can identify;

[0018] Step 3: After acquiring the bird image, the acquired image is processed by convolutional neural network to distinguish harmful birds from harmless birds and obtain the characteristic points of the birds;

[0019] By mapping the feature points of the two-dimensional plane to the three-dimensional space, the coordinates of the bird in the three-dimensional space are estimated;

[0020] The bird's motion is estimated and calculated based on the changing pattern of the bird's three-dimensional coordinates;

[0021] After obtaining the flight trajectory of the bird, the pan-tilt module is controlled so that it can be aligned with the direction of movement of the bird and the ultrasonic wave emission operation is performed.

[0022] The beneficial effects of the present invention are:

[0023] Compared with the blindness of the traditional bird driving method, the application utilizes visual detection technology to effectively and accurately distinguish different types of birds, identifies the target birds, and the two-axis gimbal carried by the unmanned aerial vehicle drives the target according to the flight trajectory of the target birds and the movement direction of the birds to align the birds, thereby improving the working efficiency and quality. Compared with the traditional driving method, the scheme utilizes the ultrasonic wave to trigger the defense reaction of the birds to achieve the driving effect, and almost does not cause damage to the environment, is environmentally friendly and efficient; the introduction of the visual detection technology can distinguish the target birds, estimate the flight trajectory of the birds, and provide the bird movement coordinates for the ultrasonic bird driving, thereby effectively improving the driving efficiency and effect, and is a specific and feasible environmentally friendly bird driving method. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 The application is an ultrasonic directional bird driving unmanned aerial vehicle device.

[0025] Figure 2 The application is an unmanned aerial vehicle flight control flowchart.

[0026] Figure 3 The application is a feature point coordinate on an image obtained by the visual detection module.

[0027] Figure 4 The application is a bird driving method flowchart. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0029] As shown in the drawings, Figure 1 The application provides an ultrasonic directional bird driving unmanned aerial vehicle device based on vision, which comprises:

[0030] A quadcopter unmanned aerial vehicle, a visual detection directional bird driving module installed on the unmanned aerial vehicle.

[0031] The unmanned aerial vehicle in the embodiment is composed of high-strength and lightweight carbon fiber, the main body is connected with the parking support and the wing carbon pipe 1 by an alloy device, and is connected with the bird driving gimbal by an aluminum column, and is assembled by combining a bolt sleeve set with a mortise and tenon joint.

[0032] Further, the quadcopter unmanned aerial vehicle comprises a flight control system, a moving module, and a positioning and patrolling module.

[0033] The flight control system is connected with the moving module and the positioning and patrolling module, and is used for controlling the normal work of each module through the flight control.

[0034] The movement module is connected to the flight control module, that is, the power system 2, and is used to drive the propeller to move through the motor.

[0035] The positioning patrol module is built into the flight control module and is used to locate the UAV through the locator and inertial navigation system and to make the UAV patrol along the specified route.

[0036] In a preferred example, the visual detection and directional bird-repelling module includes a main control module, a pan-tilt module, a pan-tilt attitude measurement module, a visual detection module, an ultrasonic module, and an operation processing module, wherein:

[0037] The main control module connects the pan / tilt module, ultrasonic module, and calculation processing module, and controls the normal operation of each module through the main control board.

[0038] The gimbal module is connected to the main control module and is a dual-axis gimbal 5, including two motion motors, which can enable the gimbal to achieve pitch and yaw movements.

[0039] The angle motor is limited by the electronic control program and the structural mechanical limit, which jointly determine its working viewing angle range in a two-way manner: +15° to -45°. This viewing angle range can well take into account the visual image requirements and the overall safety space of the aerial robot itself, so that it will not focus too much on downward tilting perspectives or upward high angles and forget its own environment, causing irreversible damage.

[0040] The main control module continuously receives the flag instructions sent by the operation processing module at a high frequency through the virtual serial port, ensuring that the motor can be driven to aim at the bird and drive the ultrasonic module when a bird is identified.

[0041] Further:

[0042] When no flag command is received, the main control module controls the gimbal module to pitch and yaw regularly within the limit within 2 seconds, striving to expand the range that the camera can recognize and increase the possibility of birds entering the camera's field of view.

[0043] When receiving the flag instruction, the operation processing module sends the calculated result to the main control module. The main control module obtains the pitch angle and yaw angle of the target, compares and calculates the expected pitch angle and yaw angle with the actual pitch angle and yaw angle of the motor rotation, and puts the error value into the controller of the position loop to calculate the expected motor rotation speed. The expected motor rotation speed is then compared and calculated with the actual motor rotation speed, and the calculated error is put into the controller of the speed loop to generate the corresponding PWM wave, so that the gimbal module drives the motor to reach the target pitch angle and yaw angle, that is, the gimbal module can rotate to the desired position at the desired speed to align with the target, thereby achieving the bird-repelling effect.

[0044] The computing and processing module, which connects the gimbal attitude measurement module, visual image module, and main control module, is an industrial control motherboard, that is, a lightweight computer with a core graphics card or a dedicated edge computing device with an independent graphics card. It is used to perform image processing and complete edge computing.

[0045] The gimbal attitude measurement module is connected to the calculation and processing module and is an inertial measurement unit. The gimbal is equipped with a high-precision inertial measurement unit. The calculation and processing module obtains the data of the inertial measurement unit to accurately locate the direction of the ultrasonic transmitting gimbal.

[0046] The visual detection module is connected to the processing module. The bird-repelling robot's visual image acquisition device is a binocular stereo depth camera 4 equipped with high resolution and a high shooting frequency. This device is used to capture images of birds in high-speed motion. This module's visual image acquisition device can estimate the distance to the bird and accurately determine its position in the camera coordinate system.

[0047] The ultrasonic module is connected to the main control module, and is a directional area bird-repelling structure equipment 3 composed of a buzzer and a closely connected outward-expanding sound-focusing speaker. It completes precise directional bird-repelling after receiving the main control command.

[0048] The main control board drives the buzzer, which operates on a 2-second pulse and then off-pulse 1-second pulse. The frequency fluctuates from 11kHz to 25kHz every minute, with a random frequency selected and remaining constant for 1 minute. When the bird repellent is functioning properly, the buzzer emits a strong sound to repel birds away from the ultrasonic coverage area.

[0049] like Figure 2 As shown, an embodiment of the present invention further provides a bird-repelling method using a vision-based ultrasonic directional bird-repelling drone device, and the bird-repelling method is performed using the above-mentioned bird-repelling device, including:

[0050] Step 1: After the staff has planned the designated route, the drone will cruise at a specified altitude.

[0051] Step 2: At the same time, the gimbal on the drone can regularly pitch and yaw completely within the limit, striving to expand the range that the camera can recognize and increase the possibility of birds entering the camera's field of view.

[0052] Step 3: During the shooting process, a global exposure visual image acquisition module with a high shooting frequency and high pixel count is used to capture bird images. The captured images are processed using a convolutional neural network (CNN) using the drone's onboard industrial control board to distinguish between harmful and harmless birds and to capture their characteristic features.

[0053] By mapping feature points on a two-dimensional plane into three-dimensional space, the bird's coordinates in that space are estimated. The bird's motion is estimated and solved based on the changing patterns of its three-dimensional coordinates. After determining the bird's flight path, the industrial control motherboard transmits data via USB, controls the pan-tilt head (PTZ) to align it with the bird's direction of movement, and instructs the control execution platform to initiate ultrasonic transmission.

[0054] The method further includes step 4: controlling the pan / tilt to rotate to a desired pitch angle and yaw angle at a desired speed and angle through a PID double loop, and driving a buzzer at a specified frequency and period to drive away birds.

[0055] The bird visual recognition in this embodiment includes three processes: "obtaining image data", "putting data into the inference engine" and "matching feature points".

[0056] The visual detection module is used to shoot and record the environmental information within the current camera field of view in the form of images. The number of images matches the camera shooting frame rate. The images obtained by the visual detection module are transmitted to the calculation processing module, and then the coordinates of the feature points on the image are obtained (the feature points we choose here are Figure 3 The geometric center of the red rectangle shown).

[0057] like Figure 4 As shown, this embodiment labels bird image data obtained from the China Birdwatching Record Center. Birds are divided into two categories: those that eat grain and those that do not, that is, those that are harmful to grain and those that are harmless to grain. Classes are created, categorized as [injurious, benefit]. The bird class labels are obtained in the format of [class, x, y, w, h]. The dataset is input into a mature CNN model, preferably the YOLOv5s model in this example.

[0058] Furthermore, the processing flow of the yolov5s model involves the forward propagation of a deep convolutional neural network.

[0059] First, the input image undergoes image preprocessing, which involves normalizing the image coordinates and adjusting the image size to match the model training size. After image preprocessing, the image is placed into the CSPDarknet53 backbone network to extract high-level semantic features of the image.

[0060] Subsequently, the feature pyramid network fuses features of different scales from the backbone network to effectively capture information about objects of different sizes. The detection head consists of a series of convolutional layers responsible for generating predictions for object detection, including the coordinates of the object bounding box and the probability score of the object category.

[0061] During training, the YOLOv5s model uses anchor box clustering to learn anchor boxes adapted to the dataset. These anchor boxes are used to generate model predictions. During inference, the model uses non-maximum suppression (NMS) to eliminate redundant detection results, ensuring that each object is labeled only once, and sorts the results based on detection confidence.

[0062] Finally, the model outputs object detection results, each of which contains the object category, the coordinates of the bounding box, and the confidence score of the detection.

[0063] The entire process is completed in a single forward propagation, achieving an end-to-end target detection task.

[0064] Furthermore, the input of the yolov5s model undergoes image preprocessing using the mosaic algorithm (i.e., randomly cropping and merging multiple images into one). The mosaic algorithm increases the complexity of the image background, enhances the diversity of the original dataset, and makes the network parameters more robust.

[0065] Furthermore, the network backbone of the yolov5s model reduces the accuracy loss caused by conventional image downsampling by downsampling the focus images, thereby improving the recognition accuracy of small targets such as birds.

[0066] The model also uses the CSPDarknet53 backbone network structure. CSPDarknet53 is a powerful backbone network designed for object detection tasks. By introducing the CSP structure, cross-stage connections, and channel attention mechanism, it improves the model's ability to express features of different scales and levels, thereby achieving good performance in object detection tasks.

[0067] Yolov5s strikes a good balance between low computational effort and high accuracy. The appropriate model size allows edge inference devices with lower hardware performance to achieve a good processing frequency while maintaining good recognition accuracy. Using the Yolov5s model, the lightweight edge inference devices onboard unmanned bird repellents can accurately determine the center point of a bird while ensuring high recognition accuracy, and then substitute the feature point (the bird's center) into the subsequent coordinate estimation solution.

[0068] The coordinate estimation solution in this embodiment includes two processes: "estimating the depth of feature points" and "obtaining the coordinates of the bird".

[0069] When using a binocular camera to calculate the 3D coordinates of feature points, the camera must first be calibrated. Calibration allows you to obtain the camera's internal parameters, external parameters, and distortion parameters. Internal parameters include focal length, principal point, and other camera internal parameters; external parameters include the camera's position and orientation. External parameters are valuable for calibration in multimodal environments that incorporate gyroscope sensors; distortion parameters can remove image distortion.

[0070] Subsequently, stereo matching is performed, and an algorithm is used to find corresponding feature point pairs in the images, that is, pixel points representing the same actual point in the two images. This usually involves disparity calculation, that is, the difference in pixel coordinates of the same actual point in the two images, and after calculation, depth information is obtained.

[0071] Next, using the camera geometry and parallax-depth relationship, the three-dimensional coordinates of the feature points are calculated through triangulation. The three-dimensional coordinates (X, Y, Z) can be obtained through the following relationship:

[0072] Z=f·T / d

[0073] Where Z is the depth of the feature point, f is the focal length of the camera, T is the baseline length of the camera, and d is the parallax. This process essentially uses the principle of similar triangles to map the parallax relationship in the image space to the depth of the actual object.

[0074] Finally, since the calculated 3D coordinates are usually in the camera coordinate system, a coordinate system transformation may be required, that is, converting them to the world coordinate system or other required coordinate system. This includes considering the camera's external parameter information, that is, the camera's position and orientation in the world coordinate system.

[0075] In this step, the unmanned bird repellent uses the geometric center point of the predicted frame obtained by yolov5s inference as the feature point, calculates the disparity of the feature points of the two images, and estimates the distance from the camera to the bird.

[0076] The bird's flight trajectory prediction in this embodiment includes two steps: "performing Kalman filtering on coordinates" and "calculating the bird's movement trajectory."

[0077] Due to issues such as sensor accuracy and sensor noise, the acquired and resolved data has very serious noise. However, this noise conforms to the normal distribution. Although the flight movement of birds is irregular, they will not suddenly change direction in a very short period of time. Therefore, through high-frequency calculation, a first-order linear Kalman filter can be used to converge the bird's position information to an optimal value, and the optimal estimated position of the bird is used as the final reliable data.

[0078] The observation matrix is ​​the matrix that converts the coordinates obtained by the binocular camera after estimating the depth through parallax into world coordinates.

[0079] Bird flight trajectory prediction is based on the information changes obtained after the bird is identified, such as the position transformation of the bird in the world coordinate system, to infer the bird's subsequent flight trajectory.

[0080] The Kalman filter process is as follows:

[0081] (1) State transfer equation:

[0082] x k =Ax k-1 +U k

[0083] Among them, x k with x k-1 is the pixel coordinate of the bird under the binocular camera; A is the state transfer matrix, which can be approximated by taking a unit matrix.

[0084] Since the Kalman state estimator is used for filtering, the control quantity U k is the zero vector.

[0085] (2) Observation equation:

[0086] z k =Hx k +V k

[0087] Among them, z k is the coordinate of the bird in the world coordinate system. H is the observation matrix of Kalman filter. V k is the vector related to the observation equation and the control quantity.

[0088] Let the projection matrix from the image coordinate system to the camera coordinate system be The homogeneous transformation matrix from the camera coordinate system to the world coordinate system is Then the observation matrix

[0089]

[0090] Among them, V k It is related to the control quantity, so the zero vector is selected.

[0091] The observation equation projects the pixel coordinate system onto the world coordinate system.

[0092] (3) Calculation of prediction error covariance:

[0093]

[0094] Where Q is the process noise matrix; P k is the covariance of the forecast errors.

[0095] (4) Kalman gain calculation:

[0096]

[0097] The R matrix is ​​the system noise, and the noise of the real sensor is simulated by adjusting the R matrix.

[0098] (5) Update state estimation:

[0099]

[0100] (6) Update error covariance:

[0101]

[0102] (7) Select the optimal estimate as the state output:

[0103]

[0104] (8) The bird's velocity vector is obtained from the difference equation:

[0105] v k =y k -y k-1

[0106] (9) Direction of bird’s velocity vector:

[0107]

[0108] (10) Bird's speed:

[0109] |vk|

[0110] So far, this application has used the Kalman state estimator to fully estimate the motion state of the bird in the world coordinate system, that is, the magnitude and direction of the velocity.

[0111] The predicted feature points of the bird at the next moment can be expressed as:

[0112]

[0113] in, is the position of the bird in the world coordinate system at the next moment, is the position of the bird in the world coordinate system at this moment, T is the flight time of the ultrasonic wave and the sum of the estimated computer processing time, H -1 is the inverse matrix of the observation matrix.

[0114] In this example, the "Communication with the PTZ" section includes the control parameters for the PTZ and the transmission flag:

[0115] The calculation results of the calculation processing module are transmitted to the pan / tilt module through the virtual serial port, allowing the microcontroller to rotate to a specific angle.

[0116] Get the field of view of the visual inspection module, including the horizontal field of view FOV Horizontal And vertical field of view FOV vertical .

[0117] The motion angle of the gimbal is approximately expressed by the following linear transformation:

[0118] ΔYaw=FOV Horizontal ·x correct / width

[0119] ΔPitch=FOV vertical ·y correct / height

[0120] Among them, ΔYaw and ΔPitch are the rotation angles of the bird-repelling drone based on the current gimbal angle to the predicted point, in radians; x correct 、y correct To predict the coordinates of the feature points after the image coordinate origin is corrected to the center of the image; width and height are the image width and height. The coordinate system of the image usually takes the upper left corner of the image as the origin, the horizontal right is the positive direction of the x-axis, and the vertical downward is the positive direction of the y-axis. In some graphics scenarios, the coordinate system takes the center of the image as the origin, the horizontal right is the positive direction of the x-axis, and the vertical upward is the positive direction of the y-axis. To transform the image coordinate origin to the image center, you can use the translation matrix R correct , which can be expressed as:

[0121]

[0122] For any two-dimensional coordinate point (x, y), you can use the form of homogeneous coordinates to convert it to a new coordinate system:

[0123]

[0124] When the processing module determines that ultrasonic waves can be emitted, it transmits a flag to the gimbal, instructing it to turn on ultrasonic waves to drive away birds.

[0125] The above-described embodiments and / or implementations are merely intended to illustrate the preferred embodiments and / or implementations of the present technology, and are not intended to limit the present technology in any form, and any person skilled in the art can make some changes or modifications as other equivalent embodiments without departing from the scope of the technology disclosed in the present disclosure, but should be considered as the same technology or embodiments as the present technology.

Claims

1. A vision-based ultrasonic directional bird-repelling drone device, characterized by: It includes a four-rotor drone and a visual detection and directional bird-repelling module installed on the drone; the visual detection and directional bird-repelling module includes a main control module, a pan-tilt module, a pan-tilt attitude measurement module, a visual detection module, an ultrasonic module and an operation processing module; The pan / tilt module is connected to the main control module and can realize pitch and yaw movements; The pan-tilt posture measurement module is connected to the calculation processing module and is used to accurately locate the orientation of the pan-tilt module; The visual detection module is connected to the calculation processing module and is used to capture images of birds in high-speed motion and estimate the distance of the birds, accurately obtaining the position of the birds in the camera coordinate system; The ultrasonic module is connected to the main control module and is used for directional bird repelling; The main control module continuously receives the flag bit instructions transmitted by the operation processing module at a high frequency through the virtual serial port, ensuring that when a bird is identified, the pan / tilt module can be driven to aim at the bird and the ultrasonic module can be driven at the same time; Upon receiving the flag instruction, the operation processing module sends the pitch angle and yaw angle relative to the bird to the main control module, and the main control module controls the pan / tilt module through the PID; The result of the calculation processing module is transmitted to the pan / tilt module through the virtual serial port, allowing the microcontroller to move to a specific angle; Get the field of view of the visual inspection module, including the horizontal field of view FOV Horizontal And vertical field of view FOV vertical ; The motion angle of the gimbal is approximately expressed by the following linear transformation: ΔYaw=FOV Horizontal ·x correct / width ΔPitch=FOV vertical ·y correct / height Among them, ΔYaw and ΔPitch are the rotation angles of the bird-repelling drone based on the current gimbal angle to the predicted point, in radians; x correct 、y correct The coordinates of the feature points after the image coordinate origin is corrected to the center of the picture; width and height are the width and height of the image; Transform the image coordinate origin to the image center using the translation matrix R correct , expressed as: For any two-dimensional coordinate point (x, y), use the form of homogeneous coordinates to convert it to a new coordinate system: When the processing module determines that ultrasonic waves can be emitted, it transmits a flag to the gimbal, instructing it to turn on ultrasonic waves to drive away birds.

2. The vision-based ultrasonic directional bird-repelling drone device according to claim 1, characterized in that: The ultrasonic module is a directional area bird-repelling structure equipment composed of a buzzer and a closely connected wide-type sound-focusing speaker. After receiving instructions from the main control module, it completes accurate directional bird repelling.

3. A method for repelling birds using the device according to claim 1 or 2, characterized in that: Step 1: After the staff has planned the designated route, the quadcopter will cruise at a specified altitude; Step 2: The gimbal module regularly pitches and yaws within a limited range to expand the range that the visual detection module can identify; Step 3: After acquiring the bird image, the acquired image is processed by convolutional neural network to distinguish harmful birds from harmless birds and obtain the characteristic points of the birds; By mapping the feature points of the two-dimensional plane to the three-dimensional space, the coordinates of the bird in the three-dimensional space are estimated; The bird's motion is estimated and calculated based on the changing pattern of the bird's three-dimensional coordinates; After obtaining the flight trajectory of the bird, the pan-tilt module is controlled so that it can be aligned with the direction of movement of the bird and the ultrasonic wave emission operation is performed.

4. The method according to claim 3, characterized in that: The convolutional neural network adopts the yolov5s model.

5. The method according to claim 3, characterized in that: The estimated coordinates of the bird in three-dimensional space are specifically: Calibrate the binocular camera in the visual inspection module; Perform stereo matching to find corresponding feature point pairs in the images, that is, pixels representing the same actual point in the two images, thereby obtaining depth information; Using the geometric relationship of the cameras and the parallax-depth relationship, the three-dimensional coordinates of the feature points are calculated through triangulation; Convert the 3D coordinates in the camera coordinate system to the world coordinate system.

6. The method according to claim 3, characterized in that: The estimation and solution of the bird's movement is specifically: by identifying the position transformation of the bird in the world coordinate system, inferring the bird's subsequent movement trajectory, using a first-order linear Kalman filter to converge the bird's position information to an optimal value, and using the bird's optimal estimated position as the final reliable data.

Images