Unmanned aerial vehicle image camera self-calibration method, system, medium and equipment

Abstract

The invention discloses an unmanned aerial vehicle image camera self-calibration method, system, medium and device, and relates to the technical field of camera self-calibration, and the method comprises the steps: carrying out the feature extraction and image matching of an unmanned aerial vehicle image, and constructing an undirected graph # imgabs0 #; according to the undirected graph # imgabs 1 #, generating an undirected graph # imgabs 2 # based on a minimum spanning tree; constructing an epipolar line geometric cost function based on three images of a triangular structure of an undirected graph # imgabs3 # and visible object space points in the three images; based on an undirected graph # imgabs4 #, seed three views and an optimal image of the incremental SFM are selected, an epipolar geometric cost function is used to carry out incremental unstructured bundle adjustment, and in the process, weighted iteration self-calibration is carried out on camera parameters to be calibrated; gNSS-assisted absolute orientation is utilized, GNSS constrained beam adjustment is carried out, the self-calibration precision is further improved, and optimal camera parameters after self-calibration are selected. The method can improve the precision of model reconstruction.

Description

Technical Field

[0001] The present invention relates to the technical field of camera self-calibration, and in particular to a self-calibration method, system, medium and equipment for an unmanned aerial vehicle imaging camera. Background Art

[0002] As one of the important means of collecting data for aerial photogrammetry, drones are widely used in emergency rescue, digital cities, power inspection and other fields. However, when drones collect images of strip scenes, in order to pursue efficiency, they usually collect data in a back-and-forth rectangular or S-shaped flight mode, and there is a lack of effective geometric constraints between flight strips. At the same time, due to the existence of flat areas in the scene and the fact that drones collect data at fixed heights and angles, it is a typical degenerate configuration problem. The bundle adjustment self-calibration of drone images collected by this flight mode and scene is prone to "bowl-shaped" effects, and the reconstructed model is deformed and distorted, making it impossible to perform high-precision three-dimensional reconstruction. The current mainstream method mainly relies on control points for constrained adjustment, but collecting control points in the field is time-consuming and laborious. In addition, the traditional bundle adjustment uses the information of the matching tie points between images to jointly solve the camera parameters and the three-dimensional coordinates of the tie points. However, there are many tie point data matched by drone images, resulting in a huge number of unknown parameters to be jointly solved, and it is difficult to calculate the covariance matrix, so it is impossible to effectively perform weighted adjustment on them. Summary of the invention

[0003] The purpose of the present invention is to solve the problem that the number of unknown parameters in UAV image reconstruction is large and the reconstruction model of bundle adjustment self-calibration is deformed and distorted, and to propose a UAV image camera self-calibration method, comprising the following steps: S1. Obtain drone images, perform feature extraction and image matching on drone images, and construct an undirected graph , represents the vertices of an undirected graph, Represents the edge of an undirected graph, where the vertex of the undirected graph is each image and the edge is the matching relationship between two images; S2. According to the undirected graph , generating an undirected graph based on the minimum spanning tree ; S3, based on undirected graph The three images of the three vertices of the triangle structure in the image and the object point P visible in the three images are used to construct the epipolar geometry cost function; S4, based on undirected graph The seed three views and the optimal image of the incremental SFM are selected, and the incremental unstructured bundle adjustment is performed using the epipolar geometric cost function. The camera parameters to be calibrated are subjected to weighted iterative self-calibration. GNSS-assisted absolute orientation is used to perform GNSS constrained bundle adjustment to further improve the self-calibration accuracy, and the optimal camera parameters after self-calibration are selected.

[0004] Furthermore, S2 is specifically: Traversing an undirected graph All edges, where the edge weight is set to the number of common connection points between the images represented by adjacent vertices, and a minimum spanning tree is generated based on the edge weight, converting the minimum spanning tree structure into an undirected graph , traverse the undirected graph For each edge in Find and The two vertices of the middle edge At the same time, for other connected vertices V, select the optimal candidate image from the vertices V that meets the constraints of the number of connection points and the intersection angle. ,Will With two vertices Connected edges are added without duplication Among them and Constitute the basic unit of epipolar geometric constraints; iterate through the newly added in the last cycle Each edge of , until no new edges are added So far, we get an undirected graph .

[0005] Furthermore, the following cost function is constructed based on the geometric relationship of the epipolar lines:

[0006]

[0007]

[0008] in, represents the geometric cost of the epipolar lines of the two views of the i-th image and the j-th image, represents the normalized three-dimensional coordinates of the feature point projected by the object point P in the i-th image, represents the relative translation vector between the i-th image and the j-th image, represents the normalized three-dimensional coordinates of the feature point projected by the object point P in the jth image, represents the geometric cost of the two-view epipolar line of the j-th image and the k-th image, represents the relative translation vector between the j-th image and the k-th image, represents the normalized three-dimensional coordinates of the feature point projected by the object point P in the kth image, Represents the epipolar geometric cost of the three views of the i-th, j-th, and k-th images.

[0009] Further, Using undirected graphs The connection points of the images corresponding to the vertices of the two longest sides of each triangle in the image are used to establish the epipolar geometric cost function of the two views, and the common connection points of the images corresponding to the three vertices of the triangle are used to construct the three-view geometric cost function.

[0010] Furthermore, based on the undirected graph Select the seed three views and optimal image of incremental SFM, specifically: Undirected Graph The triangle basic unit with the largest number of common visible connection points inside the triangle is selected as the seed three-view image of the incremental SFM for image reconstruction; Traverse along the boundary edges of the reconstructed image triangulation structure, in the undirected graph The vertices A of the triangles that share the same edge with the boundary edge and have not been reconstructed are searched respectively, and the number of common connection points between the vertex A and the image corresponding to the boundary edge is calculated, and the vertex in the vertex A with the largest number of connection points is taken as the optimal image.

[0011] Further, The calculation formula of GNSS constrained bundle adjustment is:

[0012] in, represents the error of GNSS constrained bundle adjustment, represents the jth weight function, Indicates that the object point Projection function projected to the image coordinate system, X represents the camera parameters, represents the two-dimensional coordinates of the jth connection point in the image, w represents the residual weight between the GNSS position and the projection center, represents the projection center, Indicates the GNSS position.

[0013] Further, The inequality constrained GNSS-aided adjustment is expressed as:

[0014] in, represents the error of inequality constrained GNSS-aided adjustment, , represents the projection center of all images, Indicates other parameters except the image projection center parameters. represents a custom weight, and , represents the residual threshold, represents the current accumulated reprojection error, , I represents the identity matrix, Indicates the GNSS position.

[0015] The present invention also provides a self-calibration system for an unmanned aerial vehicle imaging camera, comprising: The first undirected graph construction module is used to obtain drone images, perform feature extraction and image matching on drone images, and construct an undirected graph , represents the vertices of an undirected graph, Represents the edge of an undirected graph, where the vertex of the undirected graph is each image and the edge is the matching relationship between two images; The second undirected graph construction module is used to construct an undirected graph , generating an undirected graph based on the minimum spanning tree ; Geometric cost function building blocks for undirected graphs The three images of the three vertices of the triangle structure in the image and the object point P visible in the three images are used to construct the epipolar geometry cost function; Camera parameter self-calibration module for undirected graph-based The seed three views and the optimal image of the incremental SFM are selected, and the incremental unstructured bundle adjustment is performed using the epipolar geometric cost function. The camera parameters to be calibrated are subjected to weighted iterative self-calibration. GNSS-assisted absolute orientation is used to perform GNSS constrained bundle adjustment to further improve the self-calibration accuracy, and the optimal camera parameters after self-calibration are selected.

[0016] The present invention also proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned self-calibration method for the drone imaging camera is implemented.

[0017] The present invention also proposes an electronic device, including a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the above-mentioned drone imaging camera self-calibration method.

[0018] The beneficial effects brought by the technical solution provided by the present invention are: The present invention utilizes the kernel line geometry cost function to effectively eliminate the unknown parameters of the connection points and only retains the camera parameters as unknowns, which greatly reduces the number of unknown parameters and facilitates weighted adjustment solution. In view of the deformation and distortion problem of the reconstructed model, the high-precision GNSS external constraint adjustment solution is used to alleviate the "bowl-shaped" effect of the self-calibration of the long-range UAV imaging camera and improve the accuracy of model reconstruction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of a self-calibration method for a drone imaging camera according to an embodiment of the present invention; Figure 2 is a schematic diagram of the geometric relationship of the three-view cost function of an embodiment of the present invention; Figure 3 is a block diagram of an electronic device in an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0021] Explanation of terms involved in the embodiments of the present invention: SFM (Structure from motion): Given a sparse corresponding set of multiple images and their image features, the position of 3D points is estimated. This solution process usually involves the simultaneous estimation of 3D geometry (structure) and camera pose (motion). Since images may be disordered and the amount of image data is large, the data sources are wide and rich. In view of the above characteristics, three SFM strategies have emerged, including incremental, hierarchical and global.

[0022] The steps of incremental SFM: Start with a pair of images, calculate the epipolar geometry constraints and the essential matrix; decompose the poses of the two cameras from the essential matrix (rotation matrix R and displacement t), and use triangulation to calculate some 3D points; so the initial 3D reconstruction containing two cameras and some 3D points can be calculated through two images, and then the pnp method (method for solving the correspondence between 3D and 2D points) is used to add a third camera (the third picture), and the correspondence between the calculated 3D points and the newly added 2D points is used to calculate the pose of the third camera, and then triangulate. First use pnp to get the pose, and then use triangulation to get the 3D points. Gradually add cameras (pictures) and gradually find 3D points and corresponding image points.

[0023] Bundle Adjustment (BA): The core of the BA method is to minimize the reprojection error, that is, when there is a 2D point m in an image and the initial 3D coordinate P of a three-dimensional point is calculated, the three-dimensional point P is reprojected to the image plane to obtain m', minimizing the error between the original image point m and the reprojected m'.

[0024] Connecting points: Connecting points are points with the same name in the overlapping parts of different images, which are used to stitch and adjust multiple images.

[0025] Intersection angle: The intersection angle refers to the angle formed by the rays connecting each point in the same name (connection point) with its corresponding photographic center.

[0026] Undirected graph: A graph in which each edge has no direction is called an undirected graph. In an undirected graph, the number of edges to which a vertex is attached is called the degree of the vertex.

[0027] Embodiment 1: The flowchart of the self-calibration method of the drone imaging camera of the embodiment of the present invention is as follows Figure 1 , specifically including the following steps: S1. Obtain drone images, perform feature extraction and image matching on drone images, and construct an undirected graph , represents the vertices of an undirected graph, Represents the edge of an undirected graph, where the vertices of the undirected graph are each image and the edges are the matching relationships between two images.

[0028] The SIFT method can be used to extract features from drone images, perform image matching based on feature descriptors, connect matched images, use images as vertices, and connect matching relationships between them as edges to construct an undirected graph. .

[0029] S2. According to the undirected graph , generating an undirected graph based on the minimum spanning tree .

[0030] Specifically: traverse the undirected graph All edges, where the edge weight is set to the number of common connection points between the images represented by adjacent vertices, and a minimum spanning tree is generated based on the edge weight, converting the minimum spanning tree structure into an undirected graph , traverse the undirected graph For each edge in Find and The two vertices of the middle edge At the same time, for other connected vertices V, select the optimal candidate image from the vertices V that meets the constraints of the number of connection points and the intersection angle. , where the intersection angle constraint should meet the requirement of being greater than 16°. With two vertices Connected edges are added without duplication Among them and Constitute the basic unit of epipolar geometric constraints; iterate through the newly added in the last cycle Each edge of , until no new edges are added So far, we get an undirected graph .

[0031] S3, based on undirected graph The three images of the three vertices of the triangle structure in , and the object point P visible in all three images, construct the epipolar geometry cost function.

[0032] According to the geometric relationship of the epipolar line, the following cost function is constructed:

[0033]

[0034]

[0035] in, represents the geometric cost of the epipolar lines of the two views of the i-th image and the j-th image, represents the normalized three-dimensional coordinates of the feature point projected by the object point P in the i-th image, represents the relative translation vector between the i-th image and the j-th image, represents the normalized three-dimensional coordinates of the feature point projected by the object point P in the jth image, represents the geometric cost of the two-view epipolar line of the j-th image and the k-th image, represents the relative translation vector between the j-th image and the k-th image, represents the normalized three-dimensional coordinates of the feature point projected by the object point P in the kth image, Represents the epipolar geometric cost of the three views of the i-th, j-th, and k-th images.

[0036] Using undirected graphs The connection points of the images corresponding to the vertices of the two longest sides of each triangle in the image are used to establish the epipolar geometric cost function of the two views, and the common connection points of the images corresponding to the three vertices of the triangle are used to construct the three-view geometric cost function. The geometric relationship of the three-view cost function of the embodiment of the present invention is as follows: Figure 2 As shown, , , They represent the relative translation vector between the j-th and k-th images, the relative translation vector between the ith and k-th images, and the relative translation vector between the ith and j-th images respectively; , , Respectively represent the rotation matrix between the j-th and k-th images, the rotation matrix between the ith and k-th images, and the rotation matrix between the ith and j-th images; , , Respectively represent the normalized three-dimensional coordinates of the feature points projected by the object point P in the i-th, j-th, and k-th images; , , is the scale factor. Geometric cost of the epipolar lines of two views and Indicates that the light from the object point to the two cameras is coplanar with the translation vectors of the two cameras; three-view geometry cost Like the trifocal tensor, it can maintain scale consistency when three cameras are collinear.

[0037] S4, based on undirected graph The seed three views and the optimal image of the incremental SFM are selected, and the incremental unstructured bundle adjustment is performed using the epipolar geometric cost function. The camera parameters to be calibrated are subjected to weighted iterative self-calibration. GNSS-assisted absolute orientation is used to perform GNSS constrained bundle adjustment to further improve the self-calibration accuracy, and the optimal camera parameters after self-calibration are selected.

[0038] Incremental SFM starts with an initial seed image pair, adds new optimal images one by one, and iterates local and global bundle adjustments to improve the robustness and accuracy of reconstruction. Among them, the selection of the initial seed image pair and the order of adding images are crucial links in incremental bundle adjustment.

[0039] The present invention adopts the following method to select the seed three views and add the optimal image: Seed three-view selection: undirected graph The triangle basic unit with the largest number of common visible connection points inside the triangle is selected as the seed three views of the incremental SFM for image reconstruction.

[0040] Optimal image addition: Traverse along the boundary edges of the reconstructed image triangulation structure. The boundary edges of the triangulation structure refer to the edges of the triangles in the triangulation structure that do not share the same edges with other triangles. Find the vertex A of the triangle that shares the same edge with the boundary edge and has not been reconstructed, and calculate the number of common connection points between vertex A and the image corresponding to the edge. Similarly, take one of the vertices A with the largest number of connection points as the optimal image. Perform local and global adjustments iteratively, and perform weighted iterative self-calibration on the camera parameters to be calibrated.

[0041] High-precision differential GNSS positioning information can provide reliable external constraints for UAV images of long-flight degraded structures. Combined with GNSS-assisted camera self-calibration, it is an effective means to alleviate the "bowl-shaped" effect.

[0042] Estimate the rotation matrix R, translation matrix T and scaling factor s in the similarity transformation, and transform the image and the 3D connection points into the object coordinate system. Assume that the projection matrix of the image is P and the coordinates of the 3D connection points are , similarity transformation matrix , the coordinates of the two-dimensional point projected onto the image are ,Right now . 3D Point The coordinates after similarity transformation are , in order to ensure that the pixel coordinates of the three-dimensional point projection in the image remain unchanged, the projection matrix P needs to be transformed at the same time. The transformed projection matrix can be derived by the following formula .

[0043]

[0044] The absolute orientation constrained by traditional GNSS is combined with the high-precision GNSS weighted bundle adjustment. The calculation formula of the GNSS constrained bundle adjustment is:

[0045] in, represents the error of GNSS constrained bundle adjustment, represents the jth weight function, Indicates that the object point Projection function projected to the image coordinate system, X represents the camera parameters, represents the two-dimensional coordinates of the jth connection point in the image, w represents the residual weight between the GNSS position and the projection center, represents the projection center, Indicates the GNSS position.

[0046] The present invention integrates GNSS based on the self-calibration of GNSS weighted bundle adjustment and inequality constrained bundle adjustment. The inequality constrained GNSS auxiliary adjustment is expressed as:

[0047] in, represents the error of inequality constrained GNSS-aided adjustment, , represents the projection center of all images, Indicates other parameters except the image projection center parameters. represents a custom weight, and , Indicates ratio A slightly larger residual threshold, represents the current accumulated reprojection error, , I represents the identity matrix.

[0048] Select the optimal camera parameters after self-calibration, specifically: The above steps are used to parallelize the camera self-calibration for the sub-scenes. Due to the scene degradation problem, there are differences in the camera internal orientation parameters of different sub-scenes. Therefore, it is necessary to select the global optimal camera internal orientation parameters from the locally stable sub-scene. The reprojection error cannot directly measure the absolute orientation accuracy, but the residual error between the projection center and the high-precision GNSS in the elevation direction can reflect the quality of the sub-model reconstruction. If there is a large error, it means that there is deformation distortion in the reconstructed sub-scene and the error of the camera internal orientation parameters is large. Therefore, the present invention uses the size of the residual error between the reprojection center and the GNSS position information in the elevation direction to select the optimal camera parameters after self-calibration.

[0049] Through the above steps, a GNSS-constrained long-strip UAV imaging camera self-calibration method is realized, which can robustly calibrate the long-strip structure UAV imaging camera parameters, improve the positioning and orientation accuracy of long-strip scene UAV 3D reconstruction, and provide accurate 3D model support for strip scene 3D reconstruction and spatial analysis applications.

[0050] Embodiment 2: The present invention also provides a self-calibration system for a drone imaging camera, comprising: The first undirected graph construction module is used to obtain drone images, perform feature extraction and image matching on drone images, and construct an undirected graph , represents the vertices of an undirected graph, Represents the edge of an undirected graph, where the vertex of the undirected graph is each image and the edge is the matching relationship between two images; The second undirected graph construction module is used to construct an undirected graph , generating an undirected graph based on the minimum spanning tree ; Geometric cost function building blocks for undirected graphs The three images of the three vertices of the triangle structure in the image and the object point P visible in the three images are used to construct the epipolar geometry cost function; Camera parameter self-calibration module for undirected graph-based The seed three views and the optimal image of the incremental SFM are selected, and the incremental unstructured bundle adjustment is performed using the epipolar geometric cost function. The camera parameters to be calibrated are subjected to weighted iterative self-calibration. GNSS-assisted absolute orientation is used to perform GNSS constrained bundle adjustment to further improve the self-calibration accuracy, and the optimal camera parameters after self-calibration are selected.

[0051] Embodiment 3: In an exemplary embodiment, a computer-readable storage medium is included, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned self-calibration method of the drone imaging camera is implemented.

[0052] Example 4: Please refer to Figure 3 In an exemplary embodiment, an electronic device is also included, including at least one processor, at least one memory, and at least one communication bus.

[0053] Among them, a computer program is stored in the memory, and the computer program includes computer-readable instructions. The processor calls the computer-readable instructions stored in the memory through the communication bus to execute the above-mentioned drone imaging camera self-calibration method.

[0054] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A self-calibration method for an unmanned aerial vehicle imaging camera, characterized in that: The following steps are involved: S1. Obtain drone images, perform feature extraction and image matching on drone images, and construct an undirected graph , represents the vertices of an undirected graph, Represents the edge of an undirected graph, where the vertex of the undirected graph is each image and the edge is the matching relationship between two images; S2. According to the undirected graph , generating an undirected graph based on the minimum spanning tree ; S3, based on undirected graph The three images of the three vertices of the triangle structure in the image and the object point P visible in the three images are used to construct the epipolar geometry cost function; S4, based on undirected graph The seed three views and the optimal image of the incremental SFM are selected, and the incremental unstructured bundle adjustment is performed using the epipolar geometric cost function. The camera parameters to be calibrated are subjected to weighted iterative self-calibration. GNSS-assisted absolute orientation is used to perform GNSS constrained bundle adjustment to further improve the self-calibration accuracy, and the optimal camera parameters after self-calibration are selected.

2. The self-calibration method for a drone imaging camera according to claim 1, characterized in that: S2 is specifically: Traversing an undirected graph All edges, where the edge weight is set to the number of common connection points between the images represented by adjacent vertices, and a minimum spanning tree is generated based on the edge weight, converting the minimum spanning tree structure into an undirected graph , traverse the undirected graph For each edge in Find and The two vertices of the middle edge At the same time, for other connected vertices V, select the optimal candidate image from the vertices V that meets the constraints of the number of connection points and the intersection angle. ,Will With two vertices Connected edges are added without duplication Among them and Constitute the basic unit of epipolar geometric constraints; iterate through the newly added in the last cycle Each edge of , until no new edges are added So far, we get an undirected graph .

3. The self-calibration method for a drone imaging camera according to claim 1, characterized in that: According to the geometric relationship of the epipolar line, the following cost function is constructed: in, represents the geometric cost of the epipolar lines of the two views of the i-th image and the j-th image, represents the normalized three-dimensional coordinates of the feature point projected by the object point P in the i-th image, represents the relative translation vector between the i-th image and the j-th image, represents the normalized three-dimensional coordinates of the feature point projected by the object point P in the jth image, represents the geometric cost of the epipolar lines of the two views of the j-th image and the k-th image, represents the relative translation vector between the j-th image and the k-th image, represents the normalized three-dimensional coordinates of the feature point projected by the object point P in the kth image, Represents the epipolar geometric cost of the three views of the i-th, j-th, and k-th images.

4. The self-calibration method for a drone imaging camera according to claim 3, characterized in that: Using undirected graphs The connection points of the images corresponding to the vertices of the two longest sides of each triangle in the image are used to establish the epipolar geometric cost function of the two views, and the common connection points of the images corresponding to the three vertices of the triangle are used to construct the three-view geometric cost function.

5. The self-calibration method for a drone imaging camera according to claim 1, characterized in that: Based on undirected graph Select the seed three views and optimal image of incremental SFM, specifically: Undirected Graph The triangle basic unit with the largest number of common visible connection points inside the triangle is selected as the seed three-view image of the incremental SFM for image reconstruction; Traverse along the boundary edges of the reconstructed image triangulation structure, in the undirected graph The vertices A of the triangles that share the same edge with the boundary edge and have not been reconstructed are searched respectively, and the number of common connection points between the vertex A and the image corresponding to the boundary edge is calculated, and the vertex in the vertex A with the largest number of connection points is taken as the optimal image.

6. The self-calibration method for a drone imaging camera according to claim 1, characterized in that: The calculation formula of GNSS constrained bundle adjustment is: in, represents the error of GNSS constrained bundle adjustment, represents the jth weight function, Indicates that the object point Projection function projected to the image coordinate system, X represents the camera parameters, represents the two-dimensional coordinates of the jth connection point in the image, w represents the residual weight between the GNSS position and the projection center, represents the projection center, Indicates the GNSS position.

7. The self-calibration method for a drone imaging camera according to claim 1, characterized in that: The inequality constrained GNSS-aided adjustment is expressed as: in, represents the error of inequality constrained GNSS-aided adjustment, , represents the projection center of all images, Indicates other parameters except the image projection center parameters. Represents a custom weight, and , represents the residual threshold, represents the current accumulated reprojection error, , I represents the identity matrix, Indicates the GNSS position.

8. A self-calibration system for drone imaging cameras, characterized in that: include: The first undirected graph construction module is used to obtain drone images, perform feature extraction and image matching on drone images, and construct an undirected graph , represents the vertices of an undirected graph, Represents the edge of an undirected graph, where the vertex of the undirected graph is each image and the edge is the matching relationship between two images; The second undirected graph construction module is used to construct an undirected graph , generating an undirected graph based on the minimum spanning tree ; Geometric cost function building blocks for undirected graphs The three images of the three vertices of the triangle structure in the image and the object point P visible in the three images are used to construct the epipolar geometry cost function; Camera parameter self-calibration module for undirected graph-based The seed three views and the optimal image of the incremental SFM are selected, and the incremental unstructured bundle adjustment is performed using the epipolar geometric cost function. The camera parameters to be calibrated are subjected to weighted iterative self-calibration. GNSS-assisted absolute orientation is used to perform GNSS constrained bundle adjustment to further improve the self-calibration accuracy, and the optimal camera parameters after self-calibration are selected.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises computer-readable instructions, and the processor is configured to call the computer-readable instructions to execute the method according to any one of claims 1 to 7.

Images