Target relative pose solving method based on monocular vision depth estimation

Abstract

The application discloses a target relative pose solution method based on monocular vision depth estimation, first calibrates a monocular camera intrinsic parameter matrix, takes a target image to carry out pretreatment and target detection, extracts at least two feature point pixel coordinates, then obtains feature point depth through a monocular ranging algorithm based on a pinhole imaging model and a similarity principle, obtains flight control system attitude data and carries out air communication, finally combines a projection transformation relationship between 2D-3D point pairs established based on a camera mathematical model to solve the relative pose between aircrafts. The application only needs to adopt a monocular camera to process a single frame image and at least two feature points to solve the relative pose between aircrafts, overcomes the problems of high complexity and long time consumption of a traditional monocular vision positioning algorithm, reduces the detection requirement for feature points, directly measures the feature point depth through a monocular ranging algorithm, and further reduces the calculation complexity.

Description

Technical Field

[0001] This invention relates to the field of monocular vision localization technology, and in particular to a method for calculating the relative pose of a target based on monocular vision depth estimation. Background Technology

[0002] Aerial refueling technology is an effective means of enhancing combat range and mobility in modern warfare. For unmanned aerial vehicles (UAVs), an emerging air combat force in modern warfare, one of their most prominent weaknesses is their short endurance, requiring periodic returns to base for refueling. This significantly reduces their ability to perform long-endurance and complex missions, limiting their operational capabilities. To address this issue, agencies led by NASA have launched several Automated Aerial Refueling (AAR) demonstration programs. Applying autonomous aerial refueling technology to UAVs is both a trend and an inevitability.

[0003] Autonomous aerial refueling technology essentially involves the in-flight docking of unmanned aerial vehicles (UAVs). Relative pose estimation is a crucial step in this process. Currently, sensors used to obtain relative pose primarily include Inertial Navigation Systems (INS), Differential Global Positioning Systems (DGPS), and visual sensors. INS accumulates errors over time, making it unsuitable for the accuracy requirements of the docking process. While DGPS can meet the accuracy requirements, its signal is easily blocked during docking, leading to a decrease in accuracy. Visual sensors, with their high accuracy and low cost, are widely used in autonomous aerial refueling.

[0004] In their paper "Visual Navigation Method in UAV Aerial Docking," Liu Aichao et al. proposed a relative pose estimation method based on ellipse detection and contour detection for UAV docking. This method first uses a moving docking icon and the UAV's GPS / INS information for coarse navigation to complete the rendezvous process. Then, using the color and shape information of the moving docking icon, a color segmentation method is used to select potential target regions. Fast ellipse detection and contour detection are performed in these regions to obtain accurate icon edge description information. Finally, an improved Orthogonal Iteration (OI) algorithm is used to estimate the relative pose, achieving precise navigation at close range. The drawback of this method is that it uses an iterative solution method to solve for the relative pose. This method has high computational complexity, consumes many resources, and has a long running time, which may affect the real-time performance of the aircraft's relative pose estimation to some extent.

[0005] Xi'an University of Electronic Science and Technology disclosed a target pose estimation method based on monocular vision and inertial navigation fusion in its patent application, "Target Pose Estimation Method Based on Monocular Vision and Inertial Navigation Fusion" (Patent Application No. CN202410265306.3, Patent Publication No. CN117968640A). This method obtains the relative attitude between the aircraft and a identifiable ground target using the aircraft's inertial navigation data; it calculates the relative position between the aircraft by measuring the pixel coordinates of two feature points, combining the distance formula with the projection transformation relationship between 2D-3D point pairs established based on a camera mathematical model. The shortcomings of this method are as follows: First, because it relies on inertial navigation to acquire attitude data and lacks communication, it is only suitable for measuring the relative attitude between the aircraft and a fixed ground target, and cannot measure the relative attitude between the aircraft and another movable target. Second, this method applies the projection transformation relationship between 2D-3D point pairs established based on the mathematical model of the camera. In this projection transformation relationship, since there are more unknowns than equations, it is impossible to calculate the relative position when only the pixel coordinates of the feature points are obtained. To solve this problem, the method introduces a distance formula as a supplement to the equations. Although this solves the problem, the introduced distance formula is a nonlinear equation, which undoubtedly increases the computational complexity. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to address the deficiencies involved in the background technology by providing a target relative pose calculation method based on monocular vision depth estimation.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0008] The target relative pose calculation method based on monocular vision depth estimation includes the following steps:

[0009] Step 1), perform monocular camera calibration;

[0010] Step 2) Use the calibrated monocular camera to capture the target image and perform preprocessing. Then, perform target detection on the preprocessed image to obtain a target image containing at least two feature points.

[0011] Step 3) Extract the pixel coordinates of each feature point in the target image;

[0012] Step 4) Obtain the depth information of each feature point using a monocular ranging algorithm based on the pinhole imaging model and the similarity principle;

[0013] Step 5) Acquire attitude data from the flight control system and conduct air communication to obtain the relative attitude between the aircraft.

[0014] Step 6) Using the pixel coordinates and depth information of each feature point, combined with the projection transformation relationship between the 2D-3D point pairs established based on the camera mathematical model, the relative positions between the aircraft are calculated.

[0015] Step 7) Organize the relative attitudes and relative positions between the aircraft into a homogeneous transformation matrix to obtain the target relative pose.

[0016] As a further optimization of the target relative pose calculation method based on monocular vision depth estimation in this invention, the method for calibrating the monocular camera in step 1) can be any of the following: Tsai two-step method, Zhang Zhengyou calibration method, active vision camera calibration method, camera self-calibration method, calibration based on genetic algorithm, calibration based on neural network, and calibration based on error backpropagation.

[0017] As a further optimization of the target relative pose calculation method based on monocular vision depth estimation in this invention, the monocular camera intrinsic parameter matrix parameters in step 1) include: the focal length of the monocular camera, the horizontal pixel offset, the vertical pixel offset, and the pixel coordinates of the image center of the monocular camera.

[0018] As a further optimization of the target relative pose calculation method based on monocular visual depth estimation in this invention, the method for target image preprocessing in step 2) can be any one of the following: binarization, grayscale conversion, filtering, histogram equalization, image enhancement, geometric transformation, image interpolation, image segmentation, or image morphological operations.

[0019] As a further optimization of the target relative pose calculation method based on monocular visual depth estimation in this invention, the target detection method for the preprocessed image described in step 2) can be any one of the following: contour detection, edge detection, key point detection, interest point detection, texture analysis, or feature matching.

[0020] As a further optimization of the target relative pose calculation method based on monocular vision depth estimation in this invention, the method for extracting the pixel coordinates of feature points in the target image in step 3) can be any one of the following: image moment calculation, fast corner detection, Harris corner detection, or Hough transform method.

[0021] As a further optimization of the target relative pose calculation method based on monocular vision depth estimation in this invention, the formula for the monocular ranging algorithm based on the pinhole imaging model and similarity principle used to obtain feature point depth information in step 4) is as follows:

[0022]

[0023] In the formula, z c is the depth of the feature point, which is also the z-axis coordinate of the feature point in the camera coordinate system; f is the focal length of the monocular camera; w is the actual length of the line segment formed by the two feature points as endpoints; P is the Euclidean distance between the two feature points in the pixel coordinate system.

[0024] As a further optimization of the target relative pose calculation method based on monocular vision depth estimation in this invention, the detailed steps of step 5) are as follows:

[0025] Step 5.1): During flight, the pitch angle, yaw angle, and roll angle of the aircraft at any given moment are obtained through the flight control system. The aircraft attitude matrix is ​​then obtained by combining the coordinate transformation rules. The expression for the aircraft attitude matrix is ​​as follows:

[0026]

[0027] In the formula, R is the aircraft attitude matrix; θ is the aircraft pitch angle; ψ is the aircraft yaw angle; and φ is the aircraft roll angle.

[0028] Step 5.2): Transmit the aircraft attitude matrix via air communication, and calculate the relative attitude between the aircraft using the three-dimensional coordinate system transformation relationship. The expression for the three-dimensional coordinate system transformation relationship is as follows:

[0029]

[0030] In the formula, R BA Let be the attitude matrix of aircraft B relative to aircraft A; R is the transpose of the attitude matrix of aircraft A; B Let be the attitude matrix of aircraft B.

[0031] As a further optimization of the target relative pose calculation method based on monocular vision depth estimation in this invention, the projection transformation relationship between 2D-3D point pairs established based on the camera mathematical model in step 6) is as follows:

[0032]

[0033] In the formula, x c y c z c , , and z are the x-axis, y-axis, and z-axis coordinates of the feature point in the camera coordinate system, respectively; u and v are the x-axis and y-axis coordinates of the feature point in the pixel coordinate system, respectively; f is the focal length of the monocular camera; dx is the horizontal pixel offset; dy is the vertical pixel offset; u0 and v0 are the x-axis and y-axis coordinates of the image center of the monocular camera in the pixel coordinate system, respectively.

[0034] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0035] 1. The relative pose between aircraft can be obtained by processing a single frame image using only a monocular camera, overcoming the problems of high complexity and long time consumption of traditional monocular vision positioning algorithms, and ensuring the real-time performance and accuracy of relative pose estimation during aircraft docking.

[0036] 2. Only at least two feature points are needed to estimate the relative pose between aircraft, solving the problem that traditional monocular vision localization algorithms require at least three feature points to estimate the relative pose between aircraft, thus reducing the detection requirements for feature points in relative pose estimation.

[0037] 3. By directly measuring the depth information of feature points using a monocular ranging algorithm, the number of unknowns in the projection transformation relationship between 2D-3D point pairs established based on the camera's mathematical model is reduced. This allows for the direct calculation of a unique relative position. Compared to solving the problem by supplementing equations, this invention has lower computational complexity and shorter running time. Attached Figure Description

[0038] Figure 1 This is a flowchart of the present invention;

[0039] Figure 2 (a) Figure 2 (b) are respectively the scene simulation diagram established in the simulation experiment of the present invention and the visual sensor image containing feature points captured;

[0040] Figure 3 (a) Figure 3 (b) Figure 3 (c) These are the original target images captured by the calibrated monocular camera, and the images of the target target captured by the calibrated monocular camera. Figure 3 (a) The result image after image segmentation and morphological operations based on the HSV color space; Figure 3 (b) The result after binarization;

[0041] Figure 4 This is a schematic diagram illustrating the principle of the monocular ranging algorithm based on the pinhole imaging model and the similarity principle in this invention.

[0042] Figure 5 This is a graph showing the test results of the simulation experiment of this invention. Detailed Implementation

[0043] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings:

[0044] This invention can be implemented in many different forms and should not be considered limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully express the scope of the invention to those skilled in the art. In the drawings, components are enlarged for clarity.

[0045] The idea behind this invention is to utilize real-time communication between the flight control system and the aircraft to obtain the relative attitude between them, eliminating the need for multi-view cameras or processing multiple frames of images. This solves the problems of long running time and high resource consumption associated with conventional visual positioning algorithms. After preprocessing and detecting the target image, this invention first obtains the depth information of feature points in the target image using a monocular ranging algorithm. Then, using the pixel coordinates of at least two feature points, combined with the projection transformation relationship between 2D-3D point pairs established based on a camera mathematical model, the relative position between the aircraft is calculated. This solves the problem that traditional monocular visual positioning algorithms require at least three feature points for relative pose estimation between aircraft, reducing the detection requirements for feature points in relative pose estimation. In terms of pose calculation, this invention utilizes the flight control system to obtain the current pitch, yaw, and roll angles of the aircraft, and obtains the relative attitude between the aircraft through real-time communication. By using the depth information and pixel coordinates of feature points in the target image, the relative position between the aircraft is directly calculated through the projection transformation relationship between 2D-3D point pairs established based on the camera's mathematical model. This eliminates the need for complex calculation methods such as neural networks and iterations, thus solving the problems of high computational complexity and long algorithm running time in traditional monocular vision positioning algorithms when solving pose.

[0046] The implementation steps of this invention are as follows:

[0047] Step 1), perform monocular camera calibration;

[0048] Step 2) Use the calibrated monocular camera to capture the target image and perform preprocessing. Then, perform target detection on the preprocessed image to obtain a target image containing at least two feature points.

[0049] Step 3) Extract the pixel coordinates of each feature point in the target image;

[0050] Step 4) Obtain feature point depth information using a monocular ranging algorithm based on the pinhole imaging model and similarity principle;

[0051] Step 5) Acquire attitude data from the flight control system and conduct air communication to obtain the relative attitude between the aircraft.

[0052] Step 6) Using the pixel coordinates and depth information between each pair of feature points, combined with the projection transformation relationship between the 2D-3D point pairs established based on the camera mathematical model, the relative positions between the aircraft are calculated.

[0053] Step 7) Organize the relative attitudes and positions between the aircraft into a homogeneous transformation matrix to obtain the target's relative pose. (Refer to...) Figure 1 The implementation steps of the present invention will be further described below.

[0054] Step 1: Perform monocular camera calibration.

[0055] The calibration methods for monocular cameras include, but are not limited to: Tsai's two-step method, Zhang Zhengyou's calibration method, active vision camera calibration method, camera self-calibration method, calibration based on genetic algorithm, calibration based on neural network, and calibration based on error backpropagation.

[0056] In this embodiment of the invention, the Zhang Zhengyou calibration method is used to calibrate the monocular camera. Specifically, the black and white checkerboard pattern for calibration is placed in a suitable position, and 23 images with complete checkerboard patterns are taken from different camera poses. The captured images are then imported into the Camera Calibrator APP program of MATLAB (Matrix Laboratory) software. After inputting the specifications of the black and white checkerboard pattern, calibration is performed. If the reprojection error of some images is too large during the calibration process, they can be deleted and recalibrated, ensuring that the final number of images used for calibration is between 15 and 20. If the reprojection error is no greater than 0.3, the calibration is considered successful. The intrinsic parameter matrix of the calibrated monocular camera is obtained from the output of the MATLAB software. The standard format of the intrinsic parameter matrix is ​​as follows:

[0057]

[0058] In the formula, I is the intrinsic parameter matrix of the monocular camera; f is the focal length of the monocular camera; dx is the horizontal pixel offset; dy is the vertical pixel offset; u0 and v0 are the x-axis and y-axis coordinates of the image center of the monocular camera in the pixel coordinate system, respectively.

[0059] Step 2: Use the calibrated monocular camera to capture target images and perform preprocessing, then perform target detection on the preprocessed images.

[0060] Figure 2 (a) is a simulation diagram of an aerial docking scenario for UAVs created in CoppeliaSim Edu 4.6 simulation software. Figure 2 (b) is an image containing feature points captured by a vision sensor in the simulation software.

[0061] Image preprocessing methods include, but are not limited to: binarization, grayscale conversion, filtering, histogram equalization, image enhancement, geometric transformation, image interpolation, image segmentation, and image morphological operations.

[0062] The embodiments of the present invention use image segmentation, image morphological operations, and binarization to preprocess images from a monocular camera in sequence. Figure 3 (a) is the original target image captured by the calibrated monocular camera. Figure 3 (b) is... Figure 3 (a) The result of image segmentation and morphological operations based on the HSV color space. Figure 3 (c) is correct. Figure 3 (b) The result after binarization.

[0063] For object detection in preprocessed images, the methods that can be selected include, but are not limited to: contour detection, edge detection, key point detection, interest point detection, texture analysis, and feature matching.

[0064] In embodiments of the present invention, a contour detection algorithm is used to find the contours of all feature points in the preprocessed image.

[0065] Step 3: Extract the pixel coordinates of feature points in the target image.

[0066] To extract the pixel coordinates of feature points in a target image, the methods that can be selected include, but are not limited to: image moment calculation, fast corner detection, Harris corner detection, and Hough transform.

[0067] In embodiments of the present invention, by calculating the image moments of all identified feature point contours and combining the relationship between the image moments and the image centroid pixel coordinates, the pixel coordinates of the centroids of all feature points in the target image are obtained. The relationship between the image moments and the image centroid pixel coordinates is as follows:

[0068]

[0069] In the formula, C x M y M represents the x-axis and y-axis coordinates of the image centroid in the pixel coordinate system, respectively; 10 and M 01 M is the first moment of the image, representing the image's quality distribution; 00It is the zeroth moment of the image, representing the overall quality of the image.

[0070] Step 4: Obtain feature point depth information through a monocular ranging algorithm based on the pinhole imaging model and similarity principle.

[0071] The monocular ranging algorithm used in this invention is based on the pinhole imaging model and the similarity principle, as illustrated in the schematic diagram below. Figure 4 As shown, the feature point depth z can be derived from the similarity principle. c As shown in the following formula:

[0072]

[0073] In the formula, z c is the depth of the feature point, which is also the z-axis coordinate of the feature point in the camera coordinate system; f is the focal length of the monocular camera; w is the actual length of the line segment formed by the two feature points as endpoints; P is the Euclidean distance between the two feature points in the pixel coordinate system.

[0074] Step 5: Acquire attitude data from the flight control system and conduct air communication to calculate the relative attitude between the aircraft.

[0075] In embodiments of the present invention, the pitch angle, yaw angle and roll angle of the aircraft at any time can be obtained by using the flight control system. The aircraft attitude matrix can be obtained by combining the coordinate transformation law. The aircraft attitude matrix is ​​then transmitted through air communication. Finally, the relative attitude between the aircraft can be calculated by using the three-dimensional coordinate system transformation relationship.

[0076] The aircraft attitude matrix is ​​expressed as follows:

[0077]

[0078] In the formula, R is the aircraft attitude matrix; θ is the aircraft pitch angle; ψ is the aircraft yaw angle; and φ is the aircraft roll angle.

[0079] Specifically, suppose there are two aircraft A and B in the air, and their attitude matrices obtained by their respective flight control systems are R and R, respectively. A and R B If we want to find the attitude matrix R of aircraft B relative to aircraft A... BA Then, aircraft B will set its own attitude matrix R B By transmitting the data to aircraft A via communication and combining it with the three-dimensional coordinate system transformation relationship, we can obtain:

[0080]

[0081] In the formula, R BA Let be the attitude matrix of aircraft B relative to aircraft A; R is the transpose of the attitude matrix of aircraft A; B Let be the attitude matrix of aircraft B.

[0082] Step 6: Using the pixel coordinates and depth information of the two feature points, combined with the projection transformation relationship between the 2D-3D point pairs established based on the camera mathematical model, the relative positions between the aircraft are calculated.

[0083] The projection transformation relationship between 2D-3D point pairs established based on the camera's mathematical model is shown in the following formula:

[0084]

[0085] In the formula, x c y c z c , , and z are the x-axis, y-axis, and z-axis coordinates of the feature point in the camera coordinate system, respectively; u and v are the x-axis and y-axis coordinates of the feature point in the pixel coordinate system, respectively; f is the focal length of the monocular camera; dx is the horizontal pixel offset; dy is the vertical pixel offset; u0 and v0 are the x-axis and y-axis coordinates of the image center of the monocular camera in the pixel coordinate system, respectively.

[0086] In the embodiments of the present invention, since the pixel coordinates (u,v) of the feature point have been obtained by calculating image moments, and the depth z of the feature point has been obtained by a monocular ranging algorithm, c At this point, the pixel coordinates (u,v) and the feature point depth z are directly used. c Substituting into the above equation, x can be solved by solving a linear equation. c y c This means that the relative positions between the aircraft were calculated.

[0087] The final calculated relative positions between the aircraft are compared with those given by the simulation software. The comparison results are as follows: Figure 5 As shown, when the lateral distance between the aircraft is not too large, the algorithm results and simulation results are almost identical, with an error at the millimeter level. When the lateral distance is large, although there is some error between the algorithm results and simulation results, the error still remains at the centimeter level, meeting the accuracy requirements for aerial docking of aircraft. Furthermore, since this invention only needs to solve linear equations, the algorithm runs very quickly and consumes very few computational resources. Therefore, this invention demonstrates that it can quickly and accurately calculate the relative pose between aircraft using at least two feature points through monocular visual depth estimation and aerial communication.

[0088] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0089] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for solving relative pose of a target based on monocular vision depth estimation, characterized in that, The method comprises the following steps: Step 1), calibrating a monocular camera; Step 2), taking a target image by using the calibrated monocular camera and performing preprocessing, and then performing target detection on the preprocessed image to obtain a target image containing at least two feature points; Step 3), extracting pixel coordinates of each feature point in the target image; Step 4), obtaining depth information of each feature point by using a monocular ranging algorithm based on a pinhole imaging model and a similarity principle, and the calculation formula is as follows: ； In the formula, is the depth of the feature point, which is also the z-axis coordinate of the feature point in the camera coordinate system; is the focal length of the monocular camera; is the actual length of the line segment with the two feature points as endpoints; is the Euclidean distance of the two feature points in the pixel coordinate system; Step 5), obtaining attitude data of a flight control system and performing air communication to obtain relative attitudes between aircrafts; Step 6), using pixel coordinates and depth information of each feature point, and combining a projection transformation relationship between 2D-3D point pairs established based on a camera mathematical model, solving relative positions between the aircrafts; Step 7), arranging the relative attitudes and the relative positions between the aircrafts into a homogeneous transformation matrix form to obtain a target relative pose. 2.The monocular vision depth estimation based target relative pose solving method according to claim 1, characterized in that, In step 1), the method for calibrating the monocular camera adopts any one of the following methods: Tsai two-step method, Zhang Zhengyou calibration method, active vision camera calibration method, camera self-calibration method, calibration based on a genetic algorithm, calibration based on a neural network, and calibration based on error back propagation. 3.The monocular vision depth estimation based target relative pose solving method according to claim 1, characterized in that, In step 1), the monocular camera intrinsic matrix parameters include a focal length of the monocular camera, a horizontal pixel offset, a vertical pixel offset, and pixel coordinates of a center of the monocular camera image. 4.The monocular vision depth estimation based target relative pose solving method of claim 1, wherein, In step 2), the method for target image preprocessing adopts any one of the following methods: binarization, grayscale, filtering, histogram equalization, image enhancement, geometric change, image interpolation, image segmentation, and image morphological operation.

5. The monocular vision depth estimation based target relative pose solving method according to claim 1, characterized in that, In step 2), the target detection method for the preprocessed image adopts any one of the following methods: contour detection, edge detection, key point detection, interest point detection, texture analysis, and feature matching. 6.The monocular vision depth estimation based target relative pose solving method of claim 1, wherein, In step 3), the method for extracting pixel coordinates of feature points in the target image adopts any one of the following methods: image moment calculation, fast corner detection, Harris corner detection, and Hough transformation method.

7. The monocular vision depth estimation based target relative pose solving method according to claim 1, characterized in that, The detailed steps of step 5) are as follows: Step 5.1), during flight, obtaining a pitch angle, a yaw angle and a roll angle of an aircraft at any time through a flight control system, and combining a coordinate conversion rule to obtain an aircraft attitude matrix, and the expression of the aircraft attitude matrix is as follows: ； wherein is the aircraft attitude matrix; is the aircraft pitch angle; is the aircraft yaw angle; is the aircraft roll angle; Step 5.2), transmitting the aircraft attitude matrix through air communication, and using a three-dimensional coordinate system conversion relationship to solve relative attitudes between the aircrafts, and the expression of the three-dimensional coordinate system conversion relationship is as follows: ； wherein is the attitude matrix of the aircraft B with respect to the aircraft A; is the transpose of the attitude matrix of the aircraft A; is the attitude matrix of the aircraft B. 8.The monocular vision depth estimation based target relative pose solving method of claim 1, wherein, In step 6), the projection transformation relationship between 2D-3D point pairs established based on the camera mathematical model is as follows: ； In the formula, , , are the x-axis, y-axis, and z-axis coordinates of the feature point in the camera coordinate system, respectively; u and v are the x-axis and y-axis coordinates of the feature point in the pixel coordinate system, respectively; is the focal length of the monocular camera; is the horizontal pixel offset; is the vertical pixel offset; , are the x-axis and y-axis coordinates of the center of the monocular camera image in the pixel coordinate system, respectively.

Images