An unmanned aerial vehicle deck transfer guiding method based on guide line relative measurement

Abstract

This application belongs to the field of airborne navigation visual relative measurement, and specifically relates to a method for UAV deck transfer guidance based on guide line relative measurement. The method involves preprocessing the original image using filtering and histogram analysis to remove noise, resulting in a preprocessed image. Image correction is then performed on the preprocessed image, including intrinsic distortion correction and platform extrinsic correction, yielding a lens-distortion-free top-view corrected image. The guide line coordinates of the top-view corrected image are extracted, and these coordinates are extracted and fitted to obtain a fitted guide curve equation. The fitted guide curve equation is then calculated to obtain the UAV platform's heading offset angle and offset distance. By capturing the deck guide line with an airborne downward-facing camera, the lateral position deviation and heading deviation of the UAV relative to the guide line are calculated, enabling efficient UAV transfer on the deck and improving UAV mission capabilities.

Description

Technical Field

[0001] This application belongs to the field of airborne navigation visual relative measurement, and specifically relates to a method for unmanned aerial vehicle deck transfer guidance based on guide line relative measurement. Background Technology

[0002] The importance of drones in modern times is constantly increasing. The ability to transfer drones on the deck is an important factor affecting the efficiency of aircraft missions. Providing safe and efficient transfer guidance for drones (such as guiding mission aircraft to the takeoff position) is of great significance for improving the overall performance of drones.

[0003] Visual relative measurement technology uses an image sensor (industrial camera) to capture the image information of the object under test. Based on the image information and the imaging principle of the image sensor, it can accurately measure the position and orientation of the object under test relative to the image sensor.

[0004] Therefore, how to guide drones more effectively is a problem that needs to be solved. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a method for guiding the transfer of unmanned aerial vehicles (UAVs) on decks based on relative measurement of guide lines, thereby resolving the problem of low transfer accuracy of UAVs on decks in existing technologies.

[0006] The technical solution of this application is: a method for guiding the transfer of unmanned aerial vehicles (UAVs) on a deck based on relative measurement of guide lines, comprising:

[0007] A monocular camera is mounted on the drone's image stabilization platform and a supplementary light source is set up. The monocular camera collects the imaging data of the runway line in real time to obtain the raw image output by the drone.

[0008] The original image is preprocessed by filtering and histogram analysis to remove noise, resulting in a preprocessed image.

[0009] The preprocessed image is then corrected, including intrinsic distortion correction and extrinsic platform correction, to obtain a top-view corrected image without lens distortion.

[0010] Extract the coordinates of the guide line from the top-view corrected image, extract and fit the guide line coordinates to obtain the fitted guide curve equation;

[0011] The heading offset angle and offset distance of the UAV platform are obtained by calculating the fitted guide curve equation.

[0012] Preferably, the original image is preprocessed using filtering and histogram analysis, specifically as follows:

[0013] Gaussian filtering is applied to the original image output by the UAV, and the image is smoothed using the Gaussian function to obtain the Gaussian filtered image.

[0014] Histogram equalization is performed on the Gaussian filtered image to adjust the image pixel values ​​and obtain the histogram equalized image.

[0015] The image with histogram equalization is then subjected to image contrast enhancement and sharpening to obtain the preprocessed image.

[0016] Preferably, the Gaussian function is:

[0017] ;

[0018] In the formula, Let x and y be the standard deviation, and x and y be the coordinates on the plane.

[0019] The formula for adjusting the image pixel values ​​is:

[0020] ;

[0021] Where s is the pixel value of the output image, and r is the pixel value of the input image. It is the normalized histogram of the input image;

[0022] The formula for enhancing image contrast is:

[0023] ;

[0024] Where c is a constant used to control the degree of transformation, s is the output pixel value, and r is the input pixel value;

[0025] The gradient of the image is calculated using the Sobel operator and then fused with the contrast-enhanced image to sharpen the image edges. The sharpening formula is as follows:

[0026] ;

[0027] ;

[0028] Where s is the output pixel value, r is the input pixel value, and G is the result of the Sobel operator. It is the normalization function for the Sobel gradient result, and c is a constant.

[0029] Preferably, the original image is preprocessed using filtering and histogram analysis, specifically as follows:

[0030] Bilateral filtering is applied to the original image output by the UAV to obtain an image that retains the image edge information;

[0031] An adaptive histogram equalization method is used to correct the brightness of an image that preserves its edge information. The image is divided into multiple sub-blocks, and histogram equalization is performed on each sub-block. The images are then combined to obtain the equalized image.

[0032] The equalized image is then subjected to image contrast enhancement and sharpening to obtain the preprocessed image.

[0033] Preferably, the formula for calculating bilateral filtering is:

[0034] ;

[0035] in, and These are the spatial standard deviation and the pixel value standard deviation, respectively.

[0036] The formula for brightness correction of an image that preserves its edge information is:

[0037] ;

[0038] Where s is the pixel value of the output image, and r is the pixel value of the input image. It is the normalized histogram of the local area obtained from the input image;

[0039] The formula for enhancing and sharpening the contrast of the equalized image is:

[0040] ;

[0041] Where s is the pixel value of the output image, and r is the pixel value of the input image. This is the smoothed image, and k is the enhancement coefficient.

[0042] Preferably, image correction is performed on the preprocessed image, specifically as follows:

[0043] Before use, the monocular camera is calibrated to obtain its model and parameters. The monocular camera is a pinhole camera model, and the camera's focal length, principal point, radial distortion, and tangential distortion parameters are obtained using the checkerboard calibration method. The pinhole camera model formula is as follows:

[0044] , ;

[0045] ;

[0046] in, and For camera focal length, and As the camera's main point, For camera distortion parameters, It is radial distortion. Tangential distortion;

[0047] The camera focal length, principal point, radial distortion, and tangential distortion parameters are input into the distortion formula to correct the preprocessed image, resulting in the corrected image. The distortion formula is:

[0048] Radial distortion:

[0049] ;

[0050] Tangential distortion:

[0051] ;

[0052] in and These are the distorted pixel coordinates. and For denormalized pixel coordinates;

[0053] Using the camera's mounted extrinsic parameters and the UAV platform's real-time attitude angles, the camera's real-time extrinsic parameters are calculated using the following formula:

[0054] ;

[0055] in, External parameters are installed on the camera and calibrated during camera installation. The real-time external parameters of the camera carrier platform are calculated using the attitude angles output in real time by the UAV's inertial navigation system. Provides real-time external parameters for the camera;

[0056] The offset coordinates are calculated based on the pitch angle, and the relative pitch angle of the camera relative to the absolute horizontal plane is obtained by combining the real-time extrinsic parameters of the camera. The compensation offset y-coordinate of each pixel coordinate in the image is calculated based on the relative pitch angle.

[0057] The offset coordinates are calculated based on the pitch angle, and the relative roll angle of the camera relative to the absolute horizontal plane is obtained by combining the real-time extrinsic parameters of the camera. The compensation offset x-coordinate of each pixel coordinate in the image is calculated based on the roll angle.

[0058] The image is remapped based on the compensated offset x-coordinate and compensated offset y-coordinate, as shown in the following formula:

[0059]

[0060] in, This is a top-down view image after mapping, without pitch or roll, and without lens distortion. This is the image after internal parameter distortion correction. This is a pixel mapping function.

[0061] Preferably, the coordinates of the guide line are extracted and fitted, specifically as follows:

[0062] Design a neural network based on an encoder-decoder structure. The encoder part extracts and encodes features from the input image and consists of 10 fully convolutional networks. The decoder part transforms the output features of the encoder to obtain the required pixel classification matrix features. The encoder mainly consists of 10 deconvolutional modules.

[0063] A segmentation module is connected after the encoder. The segmentation module consists of two 3◊3 convolutional networks and two 1◊1 convolutional networks, and finally outputs the segmentation result of W◊H◊2.

[0064] In the segmentation results of W◊H◊2, filter the pixels that are classified as positive (+) by the guide line to obtain the pixel coordinates of the guide line;

[0065] Based on the segmented ground marker image and guide line pixel coordinates, the coordinates of each marker point are recorded; and the least squares method is used to perform high-order fitting of the marker lines to obtain the center position of the marker lines.

[0066] Preferably, the coordinates of the guide line are extracted and fitted, specifically as follows:

[0067] Design a lightweight neural network structure based on residual neural networks. The network has 11 layers, including 2 regular 3◊3 convolutional layers, 7 residual convolutional network layers, and 2 regular 1◊1 convolutional layers. The input is a corrected runway line image of size W◊H◊3, and the output is a guide line segmentation result image of size W / 2◊H / 2◊2.

[0068] Upsample the segmentation result image of W / 2◊H / 2◊2 guide line to obtain the segmentation result. The calculation formula is:

[0069] ;

[0070] in, For output high-resolution images, For low-resolution images, This is a rounding function, where M and N are the low-resolution image sizes, and m and n are the high-resolution image sizes;

[0071] In the segmentation results, the pixels classified as positive (+) by the guide line are selected to obtain the pixel coordinates of the guide line;

[0072] Based on the segmented ground marker image and the pixel coordinates of the guide line, the coordinates of each marker point are recorded; and the least squares method is used to perform high-order fitting of the marker line to obtain the center position of the marker line.

[0073] Preferably, the fitted guiding curve equation is calculated, specifically as follows:

[0074] The slope of the guide curve at the center of the image pixel is calculated to obtain the heading offset angle of the UAV platform, as shown in the following formula:

[0075] ;

[0076] ;

[0077] Where k is the slope of the leader curve, and f is the equation of the leader curve. Let these be the coordinates of the camera's principal point. This refers to the heading offset angle of the unmanned aerial vehicle (UAV) platform.

[0078] Calculate the pixel distance from the image center point to the guide line, and calculate the actual offset distance of the drone based on the height of the monocular camera, using the following formula:

[0079] ;

[0080] ;

[0081] in, The pixel distance from the image center point to the guide line. The center point of the image. The equation for the leader curve. Let be the equation for the distance from a point to a curve. For the camera platform height, and For camera focal length, This represents the actual deviation distance of the drone.

[0082] The UAV deck transfer guidance method based on guideline relative measurement in this application has the following advantages:

[0083] By using an airborne downward-facing camera to photograph the deck guide line, and calculating the lateral position and heading deviation of the computer relative to the guide line, the efficient transfer of the UAV on the deck is achieved, thus improving the UAV's mission capabilities. Attached Figure Description

[0084] Figure 1 This is a flowchart of the image preprocessing process for this application;

[0085] Figure 2 This is a flowchart of the image correction process for this application;

[0086] Figure 3 This is a flowchart of the guide line identification and fitting process in this application;

[0087] Figure 4 This is a flowchart of the navigation parameter calculation process for this application. Detailed Implementation

[0088] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are only some, not all, of the embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0089] The first aspect of this application provides a method for guiding the transfer of unmanned aerial vehicles (UAVs) on a deck based on relative measurement of guide lines, comprising the following steps:

[0090] In step S100, a monocular camera is installed on the drone's anti-shake platform, and a supplementary light source is set. The monocular camera collects the imaging data of the runway line in real time to obtain the original image output by the drone.

[0091] Step S200 involves preprocessing the original image using filtering and histogram analysis to remove noise, resulting in a preprocessed image. For example... Figure 1 As shown, image preprocessing using Gaussian filtering and histogram equalization can improve image smoothness, reduce image noise, reduce target jitter in the image, reduce illumination changes in the image, enhance image contrast, and make image pixels more conducive to the extraction and measurement of guide lines.

[0092] The specific steps are as follows:

[0093] Step S210: Perform Gaussian filtering on the original image output by the UAV. Use the Gaussian function to smooth the image and effectively remove Gaussian noise. The Gaussian filtering formula is as follows:

[0094] ;

[0095] In the formula, Let x and y be the standard deviation, and x and y be the coordinates on the plane.

[0096] Step S220: Perform histogram equalization on the Gaussian-filtered image. By adjusting the image pixel values, the histogram distribution of the output image pixel values ​​is made more uniform, which can effectively prevent underexposure or overexposure of the image. The specific formula is as follows:

[0097] ;

[0098] Where s is the pixel value of the output image, and r is the pixel value of the input image. It is the normalized histogram of the input image;

[0099] Step S230: Perform image contrast enhancement and sharpening processing on the histogram-equalized image. The image contrast is enhanced by a non-linear contrast stretching method, and the specific formula is as follows:

[0100] ;

[0101] Where c is a constant used to control the degree of transformation, s is the output pixel value, and r is the input pixel value.

[0102] Then, the gradient of the image is calculated using the Sobel operator and fused with the contrast-enhanced image to sharpen the image edges. The relevant formulas for sharpening are as follows:

[0103] ;

[0104] ;

[0105] Where s is the output pixel value, r is the input pixel value, and G is the result of the Sobel operator. It is the normalization function for the Sobel gradient result, and c is a constant.

[0106] In a specific example, the second method is used to preprocess the original image through filtering and histogram analysis, specifically as follows:

[0107] Bilateral filtering is applied to the original image output by the UAV to obtain an image that retains its edge information; the formula for bilateral filtering is:

[0108] ;

[0109] in, and These are the spatial standard deviation and the pixel value standard deviation, respectively.

[0110] An adaptive histogram equalization method is used to correct the brightness of an image that preserves its edge information. The image is divided into multiple sub-blocks, and histogram equalization is performed on each sub-block before combining them to obtain the equalized image. The formula for brightness correction of the image that preserves its edge information is as follows:

[0111] ;

[0112] Where s is the pixel value of the output image, and r is the pixel value of the input image. It is the normalized histogram of the local area obtained from the input image.

[0113] The equalized image is then subjected to image contrast enhancement and sharpening to obtain the preprocessed image. The formulas for image contrast enhancement and sharpening of the equalized image are as follows:

[0114] ;

[0115] Where s is the pixel value of the output image, and r is the pixel value of the input image. This is the smoothed image, and k is the enhancement coefficient.

[0116] Step S300, as follows Figure 2 As shown, image correction is performed on the preprocessed image. Image correction mainly consists of intrinsic distortion correction and extrinsic correction of the camera and UAV platform. Through image correction, the image captured by the UAV can be converted into a top-down view image without pitch and roll distortion, facilitating subsequent extraction and high-precision measurement of runway lines. The specific steps are as follows:

[0117] Step S310: Before use, calibrate the monocular camera to obtain its model and parameters. The monocular camera is a pinhole imaging model. The camera focal length, principal point, radial distortion, and tangential distortion parameters are obtained using the checkerboard calibration method. The pinhole imaging model formula is as follows:

[0118] , ;

[0119] ;

[0120] in, and For camera focal length, and As the camera's main point, For camera distortion parameters, It is radial distortion. Tangential distortion;

[0121] Step S320: Input the camera focal length, principal point, radial distortion, and tangential distortion parameters into the distortion formula to correct the preprocessed image, obtaining the corrected image; the distortion formula is:

[0122] Radial distortion:

[0123] ;

[0124] Tangential distortion:

[0125] ;

[0126] in and These are the distorted pixel coordinates. and For denormalized pixel coordinates;

[0127] Step S330: Calculate the camera's real-time extrinsic parameters using the camera's mounted extrinsic parameters and the UAV platform's real-time attitude angles, as shown in the following formula:

[0128] ;

[0129] in, External parameters are installed on the camera and calibrated during camera installation. The real-time external parameters of the camera carrier platform are calculated using the attitude angles output in real time by the UAV's inertial navigation system. Provides real-time external parameters for the camera;

[0130] Step S340: Calculate the offset coordinates based on the pitch angle, combine the camera's real-time extrinsic parameters to obtain the relative pitch angle of the camera relative to the absolute horizontal plane, and calculate the compensation offset y coordinate of each pixel coordinate in the image based on the relative pitch angle.

[0131] Step S350: Calculate the offset coordinates based on the pitch angle, combine the camera's real-time extrinsic parameters to obtain the relative roll angle of the camera relative to the absolute horizontal plane, and calculate the compensation offset x-coordinate of each pixel coordinate in the image based on the roll angle.

[0132] Step S360: Remap the image based on the compensated offset x-coordinate and compensated offset y-coordinate, using the following formula:

[0133]

[0134] in, This is a top-down view image after mapping, without pitch or roll, and without lens distortion. This is the image after internal parameter distortion correction. This is a pixel mapping function.

[0135] Step S400: Extract the coordinates of the guide line from the top-view corrected image, extract and fit the guide line coordinates to obtain the fitted guide curve equation.

[0136] like Figure 3 As shown, in this example, the guide line classification matrix in the image is obtained through a neural network, thereby realizing guide line extraction and fitting. The specific steps are as follows:

[0137] Step S410: Design the structure of the neural network for segmentation. In this example, a neural network based on an encoder-decoder structure is designed. The encoder part mainly performs feature extraction and encoding on the input image and consists of 10 fully convolutional networks. It downsamples the image and increases the feature dimension of the neural network. The decoder part mainly transforms the output features of the encoder to obtain the pixel classification matrix features required for this task. The encoder mainly consists of 10 deconvolutional modules to realize upsampling and dimensionality reduction of the features.

[0138] Step S420: Connect the segmentation module after the encoder to obtain the image segmentation result. In this example, the segmentation module consists of two 3◊3 convolutional networks and two 1◊1 convolutional networks, and finally outputs the segmentation result of W◊H◊2.

[0139] Step S430: In the segmentation results of step S420, filter the pixels that are classified as positive (+) by the guide line to obtain the pixel coordinates of the guide line;

[0140] Based on the segmented ground marker image, curve fitting of the ground markers is achieved using high-order equations based on pixel coordinates to obtain the center position of the markers. Through the fitting operation on the markers, a functional representation of the UAV path is obtained, making the ground markers smoother. Specifically, the coordinates of each marker point are recorded in the segmented ground marker image; and a least-squares method is used for high-order curve fitting of the markers, with third-order curve fitting employed in this method.

[0141] In one specific implementation, the guide line coordinates are extracted and fitted using a second method, specifically as follows:

[0142] Design a lightweight neural network structure based on residual neural networks. The network has 11 layers, including 2 regular 3◊3 convolutional layers, 7 residual convolutional network layers, and 2 regular 1◊1 convolutional layers. The input is a corrected runway line image of size W◊H◊3, and the output is a guide line segmentation result image of size W / 2◊H / 2◊2.

[0143] Upsample the segmentation result image of W / 2◊H / 2◊2 guide line to obtain the segmentation result. The calculation formula is:

[0144] ;

[0145] in, For output high-resolution images, For low-resolution images, This is a rounding function, where M and N are the low-resolution image sizes, and m and n are the high-resolution image sizes;

[0146] In the segmentation results, the pixels classified as positive (+) by the guide line are selected to obtain the pixel coordinates of the guide line;

[0147] Based on the segmented ground marker image and the pixel coordinates of the guide line, the coordinates of each marker point are recorded; and the least squares method is used to perform high-order fitting of the marker line to obtain the center position of the marker line.

[0148] Step S500, as follows Figure 4 The fitted guide curve equation is calculated to obtain the heading offset angle and offset distance of the UAV platform.

[0149] The specific steps are as follows:

[0150] Step S510: Calculate the slope of the guide line curve at the center of the image pixel to obtain the UAV platform's heading offset angle, as shown in the following formula:

[0151] ;

[0152] ;

[0153] Where k is the slope of the leader curve, and f is the equation of the leader curve. Let these be the coordinates of the camera's principal point. This refers to the heading offset angle of the unmanned aerial vehicle (UAV) platform.

[0154] Step S520: Calculate the pixel distance from the image center point to the guide line. Based on the monocular camera height, calculate the actual offset distance of the UAV, using the following formula:

[0155] ;

[0156] ;

[0157] in, The pixel distance from the image center point to the guide line. The center point of the image. The equation for the leader curve. Let be the equation for the distance from a point to a curve. For the camera platform height, and For camera focal length, This represents the actual deviation distance of the drone.

[0158] In summary, this application has the following advantages:

[0159] By acquiring deck guideline images through an airborne downward-looking camera, and through image preprocessing, image correction, guideline recognition and fitting, and navigation parameter calculation, the lateral position deviation and heading deviation of the UAV relative to the guideline can be obtained in real time, enabling autonomous transfer of the UAV on the deck without the need for manual command and remote control, effectively reducing human error and improving transfer safety and operational efficiency.

[0160] By employing preprocessing techniques such as Gaussian filtering or bilateral filtering, histogram equalization, contrast enhancement, and sharpening, the adverse effects of factors such as imaging jitter, uneven lighting, low light and backlight, and image noise can be suppressed. Combined with camera intrinsic distortion correction and UAV platform extrinsic attitude correction, imaging deviations caused by lens distortion, deck sway, and fuselage pitch and roll can be eliminated, and a standard orthographic bottom view image can be obtained, significantly improving measurement stability and accuracy in complex environments.

[0161] The above description is merely a specific embodiment 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 technical scope 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 guiding the transfer of unmanned aerial vehicles (UAVs) on deck based on relative measurement of guide lines, characterized in that, include: A monocular camera is mounted on the drone's image stabilization platform and a supplementary light source is set up. The monocular camera collects the imaging data of the runway line in real time to obtain the raw image output by the drone. The original image is preprocessed by filtering and histogram analysis to remove noise, resulting in a preprocessed image. The preprocessed image is then corrected, including intrinsic distortion correction and extrinsic platform correction, to obtain a top-view corrected image without lens distortion. Extract the coordinates of the guide line from the top-view corrected image, extract and fit the guide line coordinates to obtain the fitted guide curve equation; The heading offset angle and offset distance of the UAV platform are obtained by calculating the fitted guide curve equation.

2. The UAV deck transfer guidance method based on guide line relative measurement as described in claim 1, characterized in that, The original image is preprocessed using filtering and histogram analysis, specifically as follows: Gaussian filtering is applied to the original image output by the UAV, and the image is smoothed using the Gaussian function to obtain the Gaussian filtered image. Histogram equalization is performed on the Gaussian filtered image to adjust the image pixel values ​​and obtain the histogram equalized image. The image with histogram equalization is then subjected to image contrast enhancement and sharpening to obtain the preprocessed image.

3. The UAV deck transfer guidance method based on guide line relative measurement as described in claim 3, characterized in that, The Gaussian function is: ； In the formula, Let x and y be the standard deviation, and x and y be the coordinates on the plane. The formula for adjusting the image pixel values ​​is: ； Where s is the pixel value of the output image, and r is the pixel value of the input image. It is the normalized histogram of the input image; The formula for enhancing image contrast is: ； Where c is a constant used to control the degree of transformation, s is the output pixel value, and r is the input pixel value; The gradient of the image is calculated using the Sobel operator and then fused with the contrast-enhanced image to sharpen the image edges. The sharpening formula is as follows: ； ； Where s is the output pixel value, r is the input pixel value, and G is the result of the Sobel operator. It is the normalization function for the Sobel gradient result, and c is a constant.

4. The UAV deck transfer guidance method based on guide line relative measurement as described in claim 1, characterized in that, The original image is preprocessed using filtering and histogram analysis, specifically as follows: Bilateral filtering is applied to the original image output by the UAV to obtain an image that retains the image edge information; An adaptive histogram equalization method is used to correct the brightness of an image that preserves its edge information. The image is divided into multiple sub-blocks, and histogram equalization is performed on each sub-block. The images are then combined to obtain the equalized image. The equalized image is then subjected to image contrast enhancement and sharpening to obtain the preprocessed image.

5. The UAV deck transfer guidance method based on guide line relative measurement as described in claim 4, characterized in that, The formula for bilateral filtering is: ； in, and These are the spatial standard deviation and the pixel value standard deviation, respectively. The formula for brightness correction of an image that preserves its edge information is: ； Where s is the pixel value of the output image, and r is the pixel value of the input image. It is the normalized histogram of the local area obtained from the input image; The formula for enhancing and sharpening the contrast of the equalized image is: ； Where s is the pixel value of the output image, and r is the pixel value of the input image. This is the smoothed image, and k is the enhancement coefficient.

6. The UAV deck transfer guidance method based on guide line relative measurement as described in claim 1, characterized in that, Image correction is performed on the preprocessed image, specifically as follows: Before use, the monocular camera is calibrated to obtain its model and parameters. The monocular camera is a pinhole camera model, and the camera's focal length, principal point, radial distortion, and tangential distortion parameters are obtained using the checkerboard calibration method. The pinhole camera model formula is as follows: ， ； ； in, and For camera focal length, and As the camera's main point, These are the camera's distortion parameters. Radial distortion, Tangential distortion; The camera focal length, principal point, radial distortion, and tangential distortion parameters are input into the distortion formula to correct the preprocessed image, resulting in the corrected image. The distortion formula is: Radial distortion: ； Tangential distortion: ； in and These are the distorted pixel coordinates. and For denormalized pixel coordinates; Using the camera's mounted extrinsic parameters and the UAV platform's real-time attitude angles, the camera's real-time extrinsic parameters are calculated using the following formula: ； in, External parameters are installed on the camera and calibrated during camera installation. The real-time external parameters of the camera carrier platform are calculated using the attitude angles output in real time by the UAV's inertial navigation system. Provides real-time external parameters for the camera; The offset coordinates are calculated based on the pitch angle, and the relative pitch angle of the camera relative to the absolute horizontal plane is obtained by combining the real-time extrinsic parameters of the camera. The compensation offset y-coordinate of each pixel coordinate in the image is calculated based on the relative pitch angle. The offset coordinates are calculated based on the pitch angle, and the relative roll angle of the camera relative to the absolute horizontal plane is obtained by combining the real-time extrinsic parameters of the camera. The compensation offset x-coordinate of each pixel coordinate in the image is calculated based on the roll angle. The image is remapped based on the compensated offset x-coordinate and compensated offset y-coordinate, as shown in the following formula: in, This is a top-down view image after mapping, without pitch or roll, and without lens distortion. This is the image after internal parameter distortion correction. This is a pixel mapping function.

7. The UAV deck transfer guidance method based on guide line relative measurement as described in claim 1, characterized in that, The coordinates of the guide line are extracted and fitted, specifically as follows: Design a neural network based on an encoder-decoder structure. The encoder part extracts and encodes features from the input image and consists of 10 fully convolutional networks. The decoder part transforms the output features of the encoder to obtain the required pixel classification matrix features. The encoder mainly consists of 10 deconvolutional modules. A segmentation module is connected after the encoder. The segmentation module consists of two 3◊3 convolutional networks and two 1◊1 convolutional networks, and finally outputs the segmentation result of W◊H◊2. In the segmentation results of W◊H◊2, filter the pixels that are classified as positive (+) by the guide line to obtain the pixel coordinates of the guide line; Based on the segmented ground marker image and guide line pixel coordinates, the coordinates of each marker point are recorded; and the least squares method is used to perform high-order fitting of the marker lines to obtain the center position of the marker lines.

8. The UAV deck transfer guidance method based on guide line relative measurement as described in claim 1, characterized in that, The coordinates of the guide line are extracted and fitted, specifically as follows: Design a lightweight neural network structure based on residual neural networks. The network has 11 layers, including 2 regular 3◊3 convolutional layers, 7 residual convolutional network layers, and 2 regular 1◊1 convolutional layers. The input is a corrected runway line image of size W◊H◊3, and the output is a guide line segmentation result image of size W / 2◊H / 2◊2. Upsample the segmentation result image of W / 2◊H / 2◊2 guide line to obtain the segmentation result. The calculation formula is: ； in, For output high-resolution images, For low-resolution images, This is a rounding function, where M and N are the low-resolution image sizes, and m and n are the high-resolution image sizes; In the segmentation results, the pixels classified as positive (+) by the guide line are selected to obtain the pixel coordinates of the guide line; Based on the segmented ground marker image and the pixel coordinates of the guide line, the coordinates of each marker point are recorded; and the least squares method is used to perform high-order fitting of the marker line to obtain the center position of the marker line.

9. The UAV deck transfer guidance method based on guide line relative measurement as described in claim 1, characterized in that, The equation of the fitted guiding curve is calculated as follows: The slope of the guide curve at the center of the image pixel is calculated to obtain the heading offset angle of the UAV platform, as shown in the following formula: ； ； Where k is the slope of the leader curve, and f is the equation of the leader curve. Let these be the coordinates of the camera's principal point. This refers to the heading offset angle of the unmanned aerial vehicle (UAV) platform. Calculate the pixel distance from the image center point to the guide line, and calculate the actual offset distance of the drone based on the height of the monocular camera, using the following formula: ； ； in, The pixel distance from the image center point to the guide line. The center point of the image. The equation for the leader curve. Let be the equation for the distance from a point to a curve. For the camera platform height, and For camera focal length, This represents the actual deviation distance of the drone.

Images