A real-time target attitude measurement method based on three-dimensional modeling

Abstract

The present invention provides a real-time target attitude measurement method based on three-dimensional modeling, including: establishing a three-dimensional model of a flying target according to close-range shooting or other prior conditions; using trajectory prediction to provide an initial attitude value, driving the three-dimensional model with the initial attitude value to obtain an initial simulated image; establishing 3D-2D feature associations between the three-dimensional model and the real image through projective geometric relationships; solving the tentative target attitude through the EPnP algorithm, and performing iterative optimization in the rough matching stage based on the proxy loss; defining a coincidence deviation distance based on the general contour points, performing iterative optimization in the fine matching stage, and outputting the optimal attitude value of the flying target relative to the camera coordinate system; and further obtaining the attitude information of the flying target in the measuring station coordinate system. The present invention gives a mathematical model for solving the three-dimensional attitude of an unlabeled flying target, which has the characteristics of high accuracy, strong real-time performance, and high robustness. When this method is extended from monocular vision to binocular or multiocular vision, higher-accuracy attitude information will be obtained.

Description

Technical Field

[0001] The present invention belongs to the field of computer vision, and particularly relates to a real-time target attitude measurement method based on three-dimensional modeling. Background Art

[0002] The detection and tracking technology of flying targets is a key technology in the fields of aerospace, transportation, range measurement, etc. It is widely used in aspects such as space optical communication, aircraft tracking control, and air traffic control. In order to meet the tracking requirements for reference points that are difficult to directly measure on the target, the tracking scenario has changed from point targets to extended targets. Based on the research progress of point target tracking, the measurement of the position parameters (X, Y, Z) τ during the target flight process has been developed and matured, but the attitude measurement of flying targets has always been a difficult problem in photogrammetry. However, the measurement and estimation of three-dimensional attitudes are the keys to achieving high-precision tracking and measurement of extended targets.

[0003] The serial images of an optoelectronic theodolite record the motion, shape, and attitude of the target, and have the characteristics of being able to represent more detailed information of the target and strong correlation between front and rear images (>20 frames / second), providing effective data support for the detection, recognition, and measurement and estimation of motion parameters of flying targets. Therefore, there is a practical task drive to determine the three-dimensional attitude of the target by fully mining the hidden information in the images.

[0004] Currently, the methods for solving the three-dimensional attitude of unlabeled flying targets based on a single image can be roughly divided into three categories. One is the method based on basic geometric structures, which is only applicable to targets with cylindrical features. The second is the method based on a truth library, whose accuracy is limited by the number of templates, and the increase in the number of templates will significantly reduce the algorithm speed. Moreover, this method has no strict mathematical relationship and can only obtain an estimated value of the target attitude. The third is the method based on three-dimensional models. By establishing a local three-dimensional model, non-cooperative targets are converted into cooperative targets, and based on the theory of central projection, relatively high-precision attitude values can be obtained. Therefore, researching a real-time target attitude measurement method based on three-dimensional modeling has important research value. Summary of the Invention

[0005] The present invention provides a real-time target attitude measurement method based on three-dimensional modeling, specifically solving the problem of attitude measurement and estimation of unlabeled flying targets. The basic solution idea of classical attitude measurement and estimation is: first, determine the initial value of target attitude estimation, and then obtain the most accurate attitude value based on the reprojection error. However, for pixel-by-pixel error solving and iteration, it requires a large time cost. Therefore, in view of the real-time problem, the present invention proposes a double-iteration loop strategy of rough matching iteration and fine matching iteration. The proposed flying target attitude measurement method has the characteristics of high accuracy, strong real-time performance, and high robustness.

[0006] The technical solution proposed by the present invention is as follows: A real-time target attitude measurement method based on 3D modeling, including:

[0007] Step 1: Provide an initial attitude value based on trajectory prediction;

[0008] Obtain the real image sequence captured by the photoelectric theodolite. According to the real image sequence, the spatial position of the flying target is provided by the photoelectric theodolite and the laser rangefinder. Based on trajectory prediction, the direction of the speed is used to replace the direction of the central axis of the flying target, and the initial attitude value A of the flying target in the camera coordinate system is obtained;

[0009] Step 2: Obtain the initial image of the 3D model with the initial attitude value;

[0010] According to the initial attitude value, drive the 3D model in the simulation imaging system, obtain the simulated image data captured by the simulation imaging system, and obtain the image of the 3D model in the specified attitude from the simulated image data;

[0011] Step 3: Establish the feature correlation between the 3D model and the real image;

[0012] According to the image (real image) of the flying target and the image (simulated image) of the 3D model, determine the contour features of the flying target and the contour features of the 3D model through image segmentation and contour extraction techniques, and further use the image of the 3D model as a medium to construct the 3D-2D feature correlation of the general contour points in the model body coordinate system and the real image coordinate system;

[0013] Step 4: Iteratively optimize in the rough matching stage;

[0014] According to the 3D-2D feature correlation of the general contour points, establish a central projection equation based on 4 feature point pairs, and obtain the tentative attitude value B of the 3D model relative to the camera coordinate system through the EPnP algorithm; According to the initial attitude value A and the tentative attitude value B, obtain the deviation between the real image and the simulated image; Iterate on the deviation until the deviation value is less than the set threshold, and output the attitude estimation value C in the rough matching stage;

[0015] Step 5: Define the coincidence deviation distance in the fine matching stage;

[0016] According to the attitude estimation value C in the rough matching stage, obtain the simulated image that best matches the real image, and define the distance between the general contour points of the real image and the simulated image as the coincidence deviation distance; Among them, the smaller the distance, the more accurate the attitude value in the fine matching stage; Iterate on the distance until the distance value is less than the set threshold, and obtain the optimal attitude value D of the flying target relative to the camera coordinate system;

[0017] Step 6. Calculate the attitude value in the theodolite coordinate system;

[0018] According to the optimal attitude value D of the flight target relative to the camera coordinate system and the rotation parameter information of the camera with the photoelectric theodolite, obtain the optimal attitude information E of the flight target relative to the theodolite coordinate system.

[0019] In addition, in one application example, it further includes: acquiring a sequence of images of the aircraft's navigation, establishing a high-precision three-dimensional model of the aircraft, and acquiring a projection image of the three-dimensional model;

[0020] The sequence of images of the aircraft's navigation is obtained by shooting with a photoelectric theodolite and is a real image collected during the aircraft's navigation; the high-precision three-dimensional model is obtained through close-range shooting or other prior knowledge; the projection image of the three-dimensional model is obtained by building a simulation imaging system in OpenGL and is a simulation image in aircraft attitude measurement and estimation;

[0021] Using a trajectory prediction method, based on the Kalman filtering principle, calculate the velocity vector of the aircraft through the position information of the tracking points. Replace the aircraft's central axis direction with the velocity vector to obtain the initial attitude value A of the aircraft in the camera coordinate system, and the accuracy of the initial attitude value is within 10°. According to the initial attitude value A, drive the three-dimensional model in the simulation imaging system to obtain the image of the three-dimensional model at the initial attitude;

[0022] According to the point pair relationship between the real image, the simulation image, and the three-dimensional model, and using the simulation image as a medium, establish the feature association between the real image and the three-dimensional model, including:

[0023] Extract the image contour according to the two-dimensional feature extraction algorithm, and the algorithm is the Sobel operator or the Canny operator. Extract the ORB features of the contour points in the two images and perform feature matching to establish the 2D rel -2D imag conversion relationship. The aircraft three-dimensional model is composed of a triangular mesh. According to the back-projection relationship, establish the 2D imag -3D imag corresponding relationship; the back-projection relationship is a fast hashing algorithm for color indexing and a fast intersection algorithm for light and spatial triangles. And using the simulation image as a medium, establish the 2D rel -3D imag feature association between the real image and the three-dimensional model.

[0024] According to the 2D rel -3D imag feature association between the real image and the three-dimensional model, establish the central projection equation Where f represents the focal length of the camera, (x, y) represents the two-dimensional image coordinates, (X, Y, Z) represents the three-dimensional coordinates in the camera coordinate system, and τ represents the transpose. The projection equations corresponding to multiple feature points are combined to obtain the tentative pose value B of the three-dimensional model relative to the camera coordinate system through the EPnP algorithm;

[0025] According to the initial attitude value A and the tentative attitude value B, the deviation between the real image and the simulated image is obtained as follows: Where ω is the roll angle, κ is the pitch angle, is the yaw angle; is the initial value of attitude A, is the tentative posture value B. According to the iterative optimization of the deviation value, when When the minimum value is taken, the pose estimation value C of the coarse matching stage is obtained, which is recorded as

[0026] According to the pose estimation value in the coarse matching stage, a simulated image that best matches the real image is obtained. The coordinates of the two-dimensional contour points of the simulated image are:

[0027]

[0028] in represents the similarity ratio, f represents the camera focal length, (x0, y0) τ Indicates the eccentricity of the principal point; (x, y) τ Represents the observation value of the two-dimensional image; (X C ,Y C ,Z C ) τ Indicates the coordinates of any point on the aircraft in the camera coordinate system; (X T ,Y T ,Z T ) τ represents the coordinates of any point on the aircraft in the target coordinate system; F represents the nonlinear model;

[0029] The R A2C Represents the rotation matrix of the photoelectric theodolite coordinate system converted to the camera coordinate system. The azimuth angle α and pitch angle β are recorded by the encoder in the photoelectric theodolite to obtain the R A2C The rotation matrix is:

[0030]

[0031] The R T2A Represents the rotation matrix of the target coordinate system to the camera coordinate system, which is composed of the parameters to be determined ω, κ, The direction cosines and coordinate axis transformation matrix are composed of the following mathematical model:

[0032]

[0033] The (X T0 , Y T0 , Z T0 ) τ represents the coordinates of the center of the target coordinate system in the photoelectric theodolite coordinate system. According to the azimuth angle α and elevation angle β recorded by the encoder and the distance value D measured by the laser rangefinder, the coordinates of the aircraft tracking point in the photoelectric theodolite coordinate system are:

[0034]

[0035] After linearization according to the analytical formula of the two-dimensional contour point coordinates of the simulated image, it is obtained:

[0036]

[0037] According to the distance between the linearized simulated image contour points and the real image contour points, let the abscissa of the i-th point and the ordinate of the j-th point on the real contour be denoted as and Then the horizontal distance and vertical distance between the m-th point on the simulated image contour and the real image contour points are defined as:

[0038]

[0039] According to the horizontal distance and vertical distance described above, the maximum value of the two is defined as the distance between the m-th point and the real image contour point, denoted as According to the distance between the m-th point and the real image contour point, the coincidence deviation distance of the contour is defined as Iterate the coincidence deviation until the iteration convergence condition is satisfied, and output the optimal attitude value at this time Perform coordinate transformation to obtain the optimal attitude value E in the station coordinate system.

[0040] The advantages of the present invention compared with the prior art are as follows:

[0041] (1) The present invention has the characteristics of high precision and strong robustness. Specifically, the present invention provides a complete analytical form for solving the three-dimensional attitude based on the real image, which is different from the attitude detection and estimation based on pattern recognition, and solves the limitation that the attitude measurement accuracy is limited by the size of the template library. In addition, the present invention defines the coincidence deviation distance based on the contour pixel points, and uses the overall contour points to smooth the extraction and matching errors brought by individual contour points, providing a high-precision and high-robustness evaluation function for attitude optimization. Compared with the method that only uses the proxy loss function to evaluate the optimization effect, the accuracy and stability of the attitude measurement result are improved.

[0042] (2) The present invention features strong real-time performance. On the one hand, the Kalman filtering algorithm is adopted to provide an initial attitude value within 10° for the attitude measurement of the flying target, effectively reducing the optimization space for attitude solution. On the other hand, the attitude solution algorithm based on EPnP with four pairs of feature points is adopted to construct a proxy loss function based on the attitude angle residual value between the real image and the simulated image. Compared with the method of directly using the reprojection error to construct the optimization objective function, the calculation efficiency of attitude measurement is effectively improved.

[0043] (3) The present invention features strong applicability. On the one hand, different from the ellipticity method, the aspect ratio method, and the helix method, the present invention does not require manual addition of marking lines for the attitude solution of the flying target and does not limit the geometric shape of the target, improving the adaptability of the attitude solution algorithm. On the other hand, when this method is extended from monocular vision to binocular or multiocular vision, higher-precision attitude information will be obtained. Therefore, this method has a wide application prospect for attitude measurement. Description of the Drawings

[0044] Figure 1 It is a schematic diagram of the scene of an application example for the specific implementation introduction of the present invention.

[0045] Figure 2 It is a schematic diagram of the coordinate system and motion parameters defined in this method in an application example.

[0046] Figure 3 It is a flowchart of the attitude measurement method proposed by the present invention in an application example.

[0047] Figure 4 It is a schematic diagram of the coincidence deviation distance defined in this method in an application example.

[0048] Figure 5 It is a graph showing the change of the coincidence deviation distance when the pitch angle is near the true value of 15°. Detailed Embodiment

[0049] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and application examples. It should be understood that the application examples described herein are only used to explain the present application and are not used to limit the present invention.

[0050] The real-time target attitude measurement method for three-dimensional modeling provided by the present invention can be applied to an application environment as Figure 1 shown. Among them, a camera and a laser rangefinder are mounted on the photoelectric theodolite to obtain the tracking point position and sequence images of the target aircraft. The three-dimensional attitude information of the target aircraft relative to the coordinate system of the photoelectric theodolite is obtained through this method. The flowchart of this method is as Figure 3 shown.

[0051] The specific implementation steps of the present invention are as follows:

[0052] Step 1: Use trajectory prediction to provide an initial attitude value.

[0053] Obtain a real image sequence captured by the monocular camera 102 - A of the photoelectric theodolite. According to the real image sequence, the spatial position of the flying target is provided by the horizontal axis encoder 102 - B, vertical axis encoder 102 - C, and laser rangefinder 103 of the photoelectric theodolite. Based on trajectory prediction, using the velocity direction to replace the direction of the central axis of the flying target, obtain the initial attitude value A of the flying target in the camera coordinate system.

[0054] Step 2: Use the initial attitude value to obtain the initial image of the 3D model.

[0055] According to the initial attitude value, drive the 3D model in the simulation imaging system, obtain the simulated image data captured by the simulation imaging system, and obtain the image of the 3D model at a specified attitude from the simulated image data.

[0056] Step 3: Establish the feature correlation between the 3D model and the real image.

[0057] According to the image of the flying target (real image) and the image of the 3D model (simulated image), determine the contour features of the flying target and the contour features of the 3D model through image segmentation and contour extraction techniques. Further, using the image of the 3D model as an intermediary, construct the 3D - 2D feature correlation of the general contour points in the model body coordinate system and the real image coordinate system.

[0058] Step 4: Iteratively optimize in the rough matching stage.

[0059] According to the 3D - 2D feature correlation of the general contour points, establish a central projection equation based on 4 feature point pairs, and obtain a tentative attitude value B of the 3D model relative to the camera coordinate system through the EPnP algorithm. According to the initial attitude value A and the tentative attitude value B, obtain the deviation between the real image and the simulated image. Iterate the deviation until the deviation value is less than the set threshold, and output the attitude estimation value C in the rough matching stage.

[0060] Step 5: Define the coincidence deviation distance in the fine matching stage.

[0061] According to the attitude estimation value C in the rough matching stage, obtain the simulated image that best matches the real image. As Figure 4 shown, define the distance between the general contour points of the real image and the simulated image as the coincidence deviation distance. Among them, the smaller the distance, the more accurate the attitude value in the fine matching stage. Iterate the distance until the distance value is less than the set threshold, and obtain the optimal attitude value D of the flying target relative to the camera coordinate system.

[0062] Step 6: Calculate the attitude value in the measuring station coordinate system;

[0063] According to the optimal attitude value D of the flying target relative to the camera coordinate system and the rotation parameter information of the camera with the photoelectric theodolite, the optimal attitude information E of the flying target relative to the measuring station coordinate system is obtained.

[0064] In addition, one of the application examples also includes: acquiring sequential images of an aircraft in flight, establishing a high-precision three-dimensional model of the aircraft, and acquiring projection images of the three-dimensional model;

[0065] The sequence of images of the aircraft in flight is obtained by taking photos with a photoelectric theodolite and is a real image collected during the aircraft's flight. The high-precision three-dimensional model is obtained by close-up photography or other prior knowledge. The projection image of the three-dimensional model is obtained by building a simulated imaging system in OpenGL and is a simulated image used in aircraft attitude measurement and estimation.

[0066] Using a trajectory prediction method and the Kalman filter principle, the aircraft's velocity vector is calculated from the position information of the tracking points. The velocity vector is used to replace the aircraft's central axis direction to obtain the initial attitude value A of the aircraft in the camera coordinate system, with an accuracy of within 10°. Based on the initial attitude value A, the 3D model in the simulated imaging system is driven to obtain an image of the 3D model at the initial attitude value.

[0067] Based on the point-to-point relationship among the real image, the simulated image, and the three-dimensional model, a feature association between the real image and the three-dimensional model is established with the simulated image as an intermediary, including:

[0068] The image contour is extracted according to a two-dimensional feature extraction algorithm, which is a Sobel operator or a Canny operator. The ORB features of the contour points in the two images are extracted and feature matching is performed to establish a 2D image of the real image and the simulated image. rel -2D imag The three-dimensional model of the aircraft is composed of a triangulated network. According to the back-projection relationship, a 2D image and a three-dimensional model are established. imag -3D imag The back-projection relationship is a fast hash algorithm of color index and a fast intersection algorithm of light and space triangles. And the 2D relationship between the real image and the 3D model is established with the simulated image as the medium. rel -3D imag Feature association.

[0069] According to the 2D rel -3D imag Feature association, establishing central projection equation Where f represents the focal length of the camera, (x, y) represents the two-dimensional image coordinates, (X, Y, Z) represents the three-dimensional coordinates in the camera coordinate system, and τ represents the transpose. The projection equations corresponding to multiple feature points are combined to obtain the tentative pose value B of the three-dimensional model relative to the camera coordinate system through the EPnP algorithm;

[0070] Figure 2 The coordinate system and motion parameters involved in this method are defined. According to the initial attitude value A and the tentative attitude value B, the deviation between the real image and the simulated image is obtained as Where ω is the roll angle, κ is the pitch angle, is the yaw angle; is the initial value of attitude A, is the tentative posture value B. According to the iterative optimization of the deviation value, when When the minimum value is taken, the pose estimation value C of the coarse matching stage is obtained, which is recorded as

[0071] According to the pose estimation value in the coarse matching stage, a simulated image that best matches the real image is obtained. The coordinates of the two-dimensional contour points of the simulated image are:

[0072]

[0073] in, represents the similarity ratio, f represents the camera focal length, (x0, y0) τ Indicates the eccentricity of the principal point; (x, y) τ Represents the observation value of the two-dimensional image; (X C ,Y C ,Z C ) τ Indicates the coordinates of any point on the aircraft in the camera coordinate system; (X T ,Y T ,Z T ) τ represents the coordinates of any point on the aircraft in the target coordinate system; F represents the nonlinear model;

[0074] The R A2C Represents the rotation matrix of the photoelectric theodolite coordinate system converted to the camera coordinate system. The azimuth angle α and pitch angle β are recorded by the encoder in the photoelectric theodolite to obtain the R A2C The rotation matrix is:

[0075]

[0076] The R T2A Represents the rotation matrix of the target coordinate system to the camera coordinate system, which is composed of the parameters to be determined ω, κ, The direction cosine coordinate axis conversion matrix is composed of the following mathematical model:

[0077]

[0078] The (X T0 , Y T0 , Z T0 ) τ represents the coordinates of the center of the target coordinate system in the photoelectric theodolite coordinate system. According to the azimuth angle α and elevation angle β recorded by the encoder and the distance value D measured by the laser rangefinder, the coordinates of the aircraft tracking point in the photoelectric theodolite coordinate system are as follows:

[0079]

[0080] After linearization according to the analytical formula of the two-dimensional contour point coordinates of the simulated image, we get:

[0081]

[0082] According to the distance between the linearized simulated image contour points and the real image contour points, let the abscissa of the i-th point and the ordinate of the j-th point on the real contour be denoted as and Then the horizontal distance and vertical distance between the m-th point on the simulated image contour and the real image contour points are defined as:

[0083]

[0084] According to the horizontal distance and vertical distance described above, define the maximum value of the two as the distance between the m-th point and the real image contour point, denoted as According to the distance between the m-th point and the real image contour point, define the coincidence deviation distance of the contour as Iterate the coincidence deviation until the iteration convergence condition is satisfied, and output the optimal attitude value at this time Perform coordinate transformation to obtain the optimal attitude value E in the station coordinate system.

[0085] Figure 5 It shows the variation of the coincidence deviation distance of the pan-contour pixel points with the elevation angle deviation when the true elevation angle is 15°. The abscissa represents the deviation value of the elevation angle. According to the data of the track prediction, its value range is limited between ±5°. The smaller the absolute value of the elevation angle deviation, the shorter the coincidence deviation distance of the pan-contour points. When the influence of errors is not considered, when the elevation angle deviation value is 0, the deviation distance of the pan-contour points is equal to 0.

[0086] Although the above description of the illustrative embodiments of the present invention is provided for the understanding of those skilled in the art of the present technology, and it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

Claims

1. A real-time target attitude measurement method based on three-dimensional modeling, characterized in that: It includes the following steps: Step 1: Use trajectory prediction to provide an initial attitude value; Obtain a sequence of real images captured by an optoelectronic theodolite. According to the sequence of real images and based on trajectory prediction, use the velocity direction to replace the central axis direction of the flying target, and obtain the initial attitude value A of the flying target in the camera coordinate system; Step 2: Obtain an initial image of the 3D model with the initial attitude value; According to the initial attitude value, drive the 3D model in the simulation imaging system to obtain the simulated image data captured by the simulation imaging system; Step 3: Establish the feature correlation between the 3D model and the real image; According to the image of the flying target, i.e., the real image, and the image of the 3D model, i.e., the simulated image, determine the contour features of the flying target and the contour features of the 3D model through image segmentation and contour extraction techniques. Further, taking the image of the 3D model as an intermediary, construct the 3D-2D feature correlation of the general contour points in the model body coordinate system and the real image coordinate system; Step 4: Iteratively optimize in the rough matching stage; According to the 3D-2D feature correlation of the general contour points, establish a central projection equation based on 4 feature point pairs, and obtain the tentative attitude value B of the 3D model relative to the camera coordinate system through the EPnP algorithm; According to the initial attitude value A and the tentative attitude value B, obtain the deviation between the real image and the simulated image; Iterate the deviation until the deviation value is less than the set threshold, and output the attitude estimation value C in the rough matching stage; Step 5: Define the coincidence deviation distance in the fine matching stage; According to the attitude estimation value C in the rough matching stage, obtain the simulated image that best matches the real image, and define the distance between the general contour points of the real image and the simulated image as the coincidence deviation distance; Among them, the smaller the distance, the more accurate the attitude value in the fine matching stage; Iterate the distance until the distance value is less than the set threshold, and obtain the optimal attitude value D of the flying target relative to the camera coordinate system; Step 6: Calculate the attitude value in the station coordinate system; According to the optimal attitude value D of the flying target relative to the camera coordinate system and the rotation parameter information of the camera with the optoelectronic theodolite, obtain the optimal attitude information E of the flying target relative to the station coordinate system, and the station coordinate system is the optoelectronic theodolite coordinate system.

2. The method according to claim 1, characterized in that Obtain a sequence of real images captured by an optoelectronic theodolite. According to trajectory prediction, obtain the image and the initial attitude value A of the flying target from the sequence of real images; Obtain the simulated image information output by the simulation imaging system, and obtain the image of the 3D model from the simulated image information, including: Use the trajectory prediction method, based on the Kalman filtering principle, calculate the velocity vector of the flying target through the position information of the tracking points, use the velocity vector to replace the central axis direction of the flying target, obtain the initial attitude value A of the flying target in the camera coordinate system, and according to the initial attitude value A, drive the 3D model in the simulation imaging system to obtain the image of the 3D model at the initial attitude value; The simulation imaging system is built by OpenGL.

3. The method according to claim 2, characterized in that, According to the image of the flying target and the image of the 3D model, establish the feature correlation between the real image and the 3D model, including: Extract the image contour according to the two-dimensional feature extraction algorithm, which is the Sobel operator or the Canny operator. Extract the ORB features of the contour points in the two images and perform feature matching to establish the 2D rel -2D imag conversion relationship. According to the back-projection relationship, establish the 2D imag -3D imag corresponding relationship between the simulated image and the 3D model; the back-projection relationship is the fast hashing algorithm for color indexes and the fast intersection algorithm for light rays and spatial triangles. Using the simulated image as an intermediary, establish the 2D rel -3D imag feature association.

4. The method according to claim 3, characterized in that, Based on the 2D-3D feature association between the real image and the 3D model, establish the central projection equation rel -3D imag and obtain a tentative attitude value B of the 3D model relative to the camera coordinate system through the EPnP algorithm, where f represents the camera focal length, (x, y) represents the two-dimensional image coordinates, (X, Y, Z) represents the three-dimensional coordinates in the camera coordinate system, and τ represents the transpose. ​ 5. The method according to claim 4, characterized in that Based on the initial attitude value A and the tentative attitude value B, the deviation between the real image and the simulated image is obtained as where ω is the roll angle, κ is the pitch angle, and is the yaw angle; represents the initial attitude value A, represents the tentative attitude value B. Iteratively optimize according to the deviation value. When takes the minimum value, the attitude estimation value C in the rough matching stage is obtained, denoted as 6. The method according to claim 5, wherein Based on the pose estimation value in the rough matching stage, a simulated image that best matches the real image is obtained, and the two-dimensional contour point coordinates of the simulated image are: Among them, represents the similarity ratio, f represents the camera focal length, and (x0, y0) τ represents the principal point eccentricity; (x, y) τ represents the observation value of the two-dimensional image; (X C , Y C , Z C ) τ represents the coordinates of any point on the flying target in the camera coordinate system; (X T , Y T , Z T ) τ represents the coordinates of any point on the flying target in the target coordinate system; F represents the non-linear model; the R A2C represents the rotation matrix for converting the photoelectric theodolite coordinate system to the camera coordinate system. By recording the azimuth angle α and the pitch angle β in the photoelectric theodolite, the R A2C rotation matrix is: The said R T2A represents the rotation matrix for converting the target coordinate system to the camera coordinate system, which is composed of the direction cosines of the parameters ω, κ to be determined, and the coordinate axis transformation matrix. The mathematical model is as follows: The (X T0 , Y T0 , Z T0 ) τ represents the coordinates of the center of the target coordinate system in the photoelectric theodolite coordinate system. According to the azimuth angle α and elevation angle β recorded by the encoder and the distance value D measured by the laser rangefinder, the coordinates of the flight target tracking point in the photoelectric theodolite coordinate system are as follows:

7. The method according to claim 6, wherein After linearization according to the analytical formula of the two-dimensional contour point coordinates of the simulated image, it is obtained that: According to the distances between the linearized simulated image contour points and the real image contour points, let the abscissa of the \(i\)-th point and the ordinate of the \(j\)-th point on the real contour be denoted as and Then, the horizontal distance and the vertical distance between the \(m\)-th point on the simulated image contour and the real image contour points are defined as: According to the horizontal distance and the vertical distance, define the maximum value of the two as the distance between the m-th point and the contour point of the real image, denoted as According to the distance between the m-th point and the contour point of the real image, define the coincidence deviation distance of the contour as Iterate the coincidence deviation until the iteration convergence condition is satisfied, and output the optimal attitude value at this time Perform coordinate system transformation to obtain the optimal attitude value E in the station coordinate system.

Images