Virtual binocular system calibration method based on infinite homography

Abstract

The application provides a virtual binocular system calibration method based on an infinite homography relationship, the virtual binocular system is a double-prism virtual binocular system, a double-prism distortion model based on geometric optics is introduced to image parameters collected by the virtual binocular system, is used for adapting to asymmetric and anisotropic distortion characteristics caused by prism refraction, and describes image point deviation caused by prism refraction and dispersion. By introducing the infinite homography relationship, the application can update the calibration point coordinates in the iterative calculation process of calibration, and constructs a re-projection error target function of mutual projection of left and right eyes according to the relationship, so that the influence of image distortion caused by the double-prism dispersion effect on the system calibration precision is effectively coped with.

Description

Technical Field

[0001] This technical field involves computer vision and stereo vision measurement, and in particular, proposes a virtual binocular system calibration method based on the homography relation at infinity. Background Technology

[0002] Traditional binocular stereo vision system calibration is fundamental to achieving accurate 3D measurement. Classical calibration methods, such as Zhang's calibration method, estimate the camera's intrinsic parameters and distortion coefficients, as well as the external geometric relationships between the binocular systems, by acquiring multiple images of a planar target from different angles. These methods typically rely on high-precision targets and stable hardware configurations and are widely used in industrial inspection, robot navigation, and other fields. However, with technological advancements, some novel stereo imaging systems, such as virtual binocular systems based on biprisms, are emerging. Figure 1 As shown, these systems have attracted attention due to their advantages such as miniaturized structure and simultaneous shooting by left and right cameras. These systems combine a single physical camera with special optics to generate two images with parallax on an image sensor, thus physically achieving monocular imaging and functionally simulating the effect of binocular measurement.

[0003] For virtual binocular systems based on biprisms, the calibration process faces more complex challenges than that of traditional binocular systems. While the biprism, as the core beam-splitting element, achieves light refraction to create left-right parallax, its material, manufacturing precision, and the wavelength characteristics of the transmitted light inevitably introduce chromatic aberration. Chromatic aberration causes different wavelengths of light to refract at different angles, resulting in a relative shift in the projection of the same spatial point onto the left and right virtual camera images—that is, image distortion caused by chromatic aberration. Existing calibration methods, such as the traditional Zhang Zhengyou method and its variants, primarily model and correct lens distortion. However, these models and methods are generally ineffective in addressing the image shift and distortion problems caused by the biprism chromatic aberration effect, severely limiting the application of virtual binocular systems in high-precision measurements.

[0004] A search revealed no patent literature on bi-targeting using the infinite monography relation. Summary of the Invention

[0005] The purpose of this invention is to provide a calibration method based on the homography relation at infinity for a virtual binocular vision system based on a double prism, which aims to address the optical chromatic aberration effect introduced by the double prism and achieve high-precision calibration of the virtual binocular system.

[0006] To solve the above-mentioned technical problems, the technical solution created by this invention is implemented as follows:

[0007] A virtual binocular system calibration method based on infinity homography relation is disclosed. The virtual binocular system is a biprism virtual binocular system. The image parameters acquired by the virtual binocular system adopt a biprism distortion model based on geometric optics to adapt to the asymmetric and anisotropic distortion characteristics caused by prism refraction, and to describe the image point offset caused by prism refraction and dispersion. By introducing the infinity homography relation, this invention can update the calibration point coordinates during the iterative calculation process of calibration, and construct a reprojection error objective function for the mutual projection of the left and right eyes based on this relation, thereby effectively addressing the impact of image distortion caused by biprism dispersion effect on the system calibration accuracy.

[0008] Furthermore, the image parameters acquired by the virtual binocular system are processed using Zhang's calibration method to obtain the initial parameters and distortion parameters of the left and right virtual cameras, which serve as reliable initial values ​​for subsequent nonlinear optimization.

[0009] Moreover, the biprism distortion model:

[0010]

[0011] Among them, (u ud ,v ud f represents the coordinates of the corrected ideal image point in the image coordinate system. u and f v These represent the distortions along the horizontal axis u and the vertical axis v of the image, respectively. and is the biprism distortion coefficient.

[0012] Furthermore, the virtual binocular system calibration method based on infinite homography includes the following steps:

[0013] Step 1: Establish the infinite homography matrix H between the left and right cameras using the planar target image. ∞ Based on the relationship of similar triangles in the polar plane, three-dimensional point reconstruction is achieved, and a reprojection error function of mutual projection is constructed.

[0014] Step 4: Introduce epipolar geometric constraints, describe the epipolar relationship between left and right image points through the fundamental matrix, and use the vertical distance from the image point to the epipolar line as a geometric error term to enhance the geometric consistency and robustness of the system in stereo vision;

[0015] Step 5: By comparing the actual physical distance between adjacent feature points on the calibration board with the reconstructed 3D point distance, a scale consistency error term is constructed to improve the physical scale accuracy of the calibration results and the reliability of actual measurements.

[0016] Step 6: Combine the reprojection error, epipolar constraint error, and interval error according to their weights to construct the final nonlinear optimization objective function. By weighted fusion of multiple types of constraints, the overall optimization of system parameters is achieved.

[0017] Step 7: Use the Levenberg-Marquardt algorithm to perform nonlinear iterative optimization on the joint objective function and comprehensively evaluate the results.

[0018] Moreover, the biprism distortion coefficient is determined by factors such as the biprism refractive index, prism angle, and the relative position of the prism and the camera.

[0019] Furthermore, step 1 utilizes the planar target image to establish an infinitely far homography relationship H between the left and right virtual camera image planes. ∞ The initial relative rotation matrix R of the system is obtained from the initial parameters and distortion parameters of the left and right virtual cameras. The specific steps are as follows:

[0020] (1) Infinite homography matrix H ∞ It describes the projection correspondence between two camera images when the scene plane is at infinity. In this case, the homography matrix depends only on the camera's intrinsic parameters and the relative rotation matrix.

[0021] (2) Three-dimensional reconstruction using the homography relation at infinity: using H ∞ It enables 3D point reconstruction without prior knowledge of the geometric information of the object under test. The reconstruction process is based on the length ratio of similar triangles in the polar plane, so that the reconstructed 3D point coordinates can be directly used to construct reprojection errors and subsequent geometric constraints.

[0022] The polar plane is formed by the optical center C of the left and right virtual cameras. l and C r and spatial point X w Together, we determine that the left camera image point x passes through H. ∞ After transformation, we obtain point x. ∞ =H ∞ x is the image point x of the left camera passing through H. ∞ The transformation corresponds to the point on the right camera image, which lies on the epipolar line of the right camera and represents the spatial point x. w Incident light | C l x w |The projection of a parallel ray at infinity onto the right image plane, and the left camera ray|C l x w It intersects the right polar line at point m. It passes through similar triangles △x. w x'm and △C r x'x ∞ The following length relationships can be established:

[0023]

[0024] At the same time, from similar triangles △c c e'm and △c p e'x ∞ We can obtain:

[0025]

[0026] Combining the two proportional relationships above, we can obtain |x w c p |'s formula:

[0027]

[0028] When the depth |x of the spatial point is obtained w c p |, Space point X w The 3D point coordinates in the right camera coordinate system can then be reconstructed.

[0029]

[0030] Where d is the optical center c of the right camera R The normalized ray direction vector directed toward image point x' can be expressed as d = (K r ) - 1 Therefore, the formula for reconstructing X in the right camera coordinate system is:

[0031]

[0032] Similarly, the formula for reconstructing X in the left camera coordinate system can be obtained as follows:

[0033]

[0034] (3) Constructing the objective function for reprojection error (mutual projection): Reprojection error is the core objective of calibration optimization, measuring the reconstructed coordinates X of spatial points. w Reproject the image points back to the image plane, and calculate the deviation between these points and the actual observed image points, ensuring that the error is minimized while optimizing the parameters:

[0035]

[0036] Where N represents the total number of all feature points of the target, x rl The reconstructed point x in the right camera coordinate system wr The image point projected onto the left camera coordinate system, x lr The reconstructed point x in the left camera coordinate system wl Image points projected onto the right camera coordinate system.

[0037] Furthermore, the specific steps of step 2 are as follows:

[0038] Epipolar constraints are applied, utilizing the epipolar geometry between the virtual binocular cameras to ensure that the optimized parameters strictly satisfy the epipolar geometry of the stereo vision system. The perpendicular distance from the projected image points to the epipolar line is used as a strict geometric error function. In the virtual binocular system, the epipolar geometry between the left and right cameras can be described by a 3x3 fundamental matrix F. For a point x in the left image and its matching point x' in the right image, the epipolar constraints are satisfied:

[0039] x' T Fx=0 (8)

[0040] Where F is determined by the intrinsic parameters K of the left and right cameras. l and K r The relative pose [R|T] is calculated to obtain:

[0041] F = (K r ) -T [T] × R(K l ) -1 (9)

[0042] The objective function expression for the polar constraint is:

[0043]

[0044] Furthermore, the specific steps of step 5 are as follows:

[0045] A target spacing constraint was introduced. By comparing the actual physical spacing of feature points on the calibration board with the 3D coordinate spacing reconstructed based on the current parameters, a physical scale error term was constructed. This enhances the system's scale recovery capability and measurement accuracy during the optimization process. For any pair of adjacent feature points P on the planar target... i and P j Its actual physical distance is D norm The reconstructed coordinates of both in the left camera coordinate system are: and The reconstruction distance between the two is:

[0046]

[0047] Furthermore, the interval error term is defined as the average of the absolute values ​​of the differences between the actual physical distance and the reconstructed distance for all adjacent point pairs, thereby imposing a consistency constraint on the physical scale during the optimization process. The interval error constraint formula is as follows:

[0048]

[0049] Furthermore, the synthetic joint optimization objective function described in step 4 combines the reprojection error function, epipolar constraint function, and interval error constraint from steps 1, 2, and 3 to form a complete nonlinear optimization objective function:

[0050] The advantages and positive effects of this invention are:

[0051] This invention provides a virtual binocular system calibration method based on infinity homography, which has the following significant advantages and positive effects compared with the prior art:

[0052] 1. Effectively addressing chromatic aberration and achieving high-precision calibration: By introducing geometric constraints of the infinite homography relation, this invention can update the coordinates of the calibration point during iterative calculation, effectively addressing image distortion caused by the biprism dispersion effect.

[0053] 2. Simplified calibration process and reduced equipment requirements: This method, like traditional methods, only requires acquiring coplanar target images from different angles to complete the calibration, without the need for precise control of the target's three-dimensional spatial movement or acquisition of the target's absolute position information.

[0054] 3. Design reasonable constraints to enhance system robustness: By introducing the monography relation at infinity, this method can adapt to different biprism manufacturing processes and material properties, and has good robustness to changes in color difference effect. Attached Figure Description

[0055] Figure 1 This is a schematic diagram of the imaging principle of the dual-prism virtual binocular system involved in this invention;

[0056] Figure 2 This is the calibration process for a virtual binocular system based on infinite-distance monography in an example of the present invention;

[0057] Figure 3 This is a schematic diagram of the three-dimensional reconstruction principle based on the homography relation at infinity in an example of the present invention;

[0058] Figure 4 These are schematic diagrams of epipolar constraint error in an example of the present invention. (a) is a schematic diagram of epipolar geometry, and (b) is a diagram defining epipolar distance. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer and easier to understand, the present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit the scope of this application.

[0060] Step 1: Obtain initial values ​​for Zhang's calibration:

[0061] First, the traditional Zhang calibration method is used to process a single image (including images from the left and right virtual cameras) of the virtual stereo system to obtain the initial parameters of the left and right cameras. Due to the chromatic aberration caused by the double prism, the intrinsic and distortion parameters obtained at this time still have errors, but they can be used as reliable initial values ​​for subsequent optimization.

[0062] Step 2: Introduce the biprism distortion model

[0063] Introducing a biprism distortion model: In order to accurately model the effect of optical chromatic aberration on image points, this invention abandons the traditional lens distortion model (radial distortion and tangential distortion) and instead adopts a biprism distortion model, thereby more accurately describing the image point shift caused by prism refraction and dispersion.

[0064] Specifically, this invention employs the following biprism distortion model derived from geometric optics:

[0065]

[0066] Among them, (u ud ,v ud ) represents the coordinates of the corrected ideal image point in the image coordinate system. u and f v These represent the distortion along the horizontal axis u and the vertical axis v of the image, respectively. and The biprism distortion coefficient is determined by system structural parameters (such as biprism refractive index, prism angle, and the relative position of the prism and camera). This model eliminates redundant terms in traditional distortion models that do not conform to the optical characteristics of biprisms, thus better adapting to the asymmetric and anisotropic distortion characteristics caused by prism refraction.

[0067] Step 3: Establish reprojection error constraints based on the homography relation at infinity

[0068] This step aims to establish an infinitely far homography relationship H between the image planes of the left and right virtual cameras using a planar target image. ∞ And obtain the initial relative rotation matrix R of the system from step 1.

[0069] (4) Infinite homography matrix H ∞ This describes the projection correspondence between two camera images when the scene plane is at infinity. In this case, the homography matrix no longer depends on the position of the plane, since it is at infinity, but only on the camera's intrinsic parameters and relative rotation matrix.

[0070] (5) 3D reconstruction using the homography relation at infinity

[0071] Using H ∞It enables 3D point reconstruction without prior knowledge of the geometric information of the object under test. This reconstruction process is based on the length ratio of similar triangles in the polar plane, so that the reconstructed 3D point coordinates can be directly used to construct reprojection errors and subsequent geometric constraints.

[0072] like Figure 3 As shown, the polar plane is formed by the optical centers C of the left and right virtual cameras. l and C r and spatial point X w Determined jointly. The left camera image point x passes through H. ∞ After transformation, we obtain point x. ∞ =H ∞ x is the image point x of the left camera passing through H. ∞ The transformation corresponds to the point on the right camera image, which lies on the epipolar line of the right camera and represents the spatial point x. w Incident light | C l x w The projection of a parallel ray at infinity onto the right image plane. Left camera ray |C l x w It intersects the right polar line at point m. It passes through similar triangles △x. w x'm and △C r x'x ∞ The following length relationships can be established:

[0073]

[0074] At the same time, from similar triangles △c c e'm and △c p e'c ∞ We can obtain:

[0075]

[0076] Combining the two proportional relationships above, we can obtain |x w c p |'s formula:

[0077]

[0078] When the depth |x of the spatial point is obtained w c p |, Space point X w The 3D point coordinates in the right camera coordinate system can then be reconstructed.

[0079]

[0080] Where d is the optical center c of the right camera R The normalized ray direction vector directed toward image point x' can be expressed as d = (K r )- 1 c'. Therefore, the formula for reconstructing X in the right camera coordinate system is:

[0081]

[0082] Similarly, the formula for reconstructing X in the left camera coordinate system can be obtained as follows:

[0083]

[0084] (6) Constructing the objective function for reprojection error (mutual projection): Reprojection error is the core objective of calibration optimization, measuring the reconstructed coordinates X of spatial points. w The image points are reprojected back to the image plane, and the deviation between these points and the actual observed image points is calculated. This invention uses mutual projection to construct the objective function, ensuring that the error is minimized while optimizing the parameters.

[0085]

[0086] Where N represents the total number of all feature points of the target, x rl The reconstructed point x in the right camera coordinate system wr The image point projected onto the left camera coordinate system, x lr The reconstructed point x in the left camera coordinate system wl Image points projected onto the right camera coordinate system.

[0087] Step 4: Establish polar constraints

[0088] To improve the robustness and accuracy of the calibration results, this invention applies epipolar constraints during the reprojection error minimization process. This constraint utilizes the epipolar geometry between virtual binocular cameras to ensure that the optimized parameters strictly satisfy the epipolar geometry of the stereo vision system. This invention uses the perpendicular distance from the projected image points to the epipolar line as a strict geometric error function.

[0089] In a virtual stereo system, the epipolar geometry between the left and right cameras can be described by a 3×3 fundamental matrix F. For a point x in the left image and its matching point x' in the right image, they satisfy the epipolar constraint:

[0090] x' T Fx=0 (9)

[0091] Wherein, F can be determined by the intrinsic parameters K of the left and right cameras. l and K r The relative pose [R|T] is calculated to obtain:

[0092] F = (K r ) -T [T] × R(Kl ) -1 (10)

[0093] like Figure 4 As shown, the objective function expression for the epipolar constraint is:

[0094]

[0095] Step 5: Establish the error constraint for the reconstructed 3D point interval

[0096] To improve the physical scale consistency and actual measurement accuracy of the calibration results, this invention introduces a target interval constraint into the optimization objective. This constraint constructs a physical scale error term by comparing the actual physical spacing of feature points on the calibration board with the 3D coordinate spacing reconstructed based on the current parameters, thereby enhancing the system's scale recovery capability and measurement accuracy during the optimization process.

[0097] For any pair of adjacent feature points P on the planar target i and P j Its actual physical distance is D norm The reconstructed coordinates of both in the left camera coordinate system are: and The reconstruction distance between the two is:

[0098]

[0099] Therefore, the interval error term is defined as the average of the absolute values ​​of the differences between the true physical distance and the reconstructed distance for all adjacent point pairs, thus imposing a consistency constraint on the physical scale during the optimization process. The formula for the interval error constraint is:

[0100]

[0101] Step 6: Synthesize the joint optimization objective function

[0102] Finally, the reprojection error function, epipolar constraint function, and margin error constraint from steps 3, 4, and 5 are combined to form a complete nonlinear optimization objective function:

[0103]

[0104] Step 7: Nonlinear Iterative Optimization of Virtual Binocular Parameters

[0105] To comprehensively evaluate the method proposed in this invention, a comparative experiment was conducted with four existing calibration methods. The experiment used the original image of the calibration board and employed nonlinear optimization based on the Levenberg-Marquardt (LM) algorithm to iteratively optimize the virtual stereo parameters. Based on this, the coplanarity error, collinearity error, and rectangular error of each method were quantitatively evaluated.

[0106] Table 1. Accuracy comparison of different methods based on the original image of the calibration board.

[0107]

[0108] As can be seen from the data in Table 1, the method of the present invention performs best in terms of coplanarity and right angle error, and is second only to the Nie calibration method in terms of collinearity error, demonstrating good geometric accuracy and physical scale consistency, which can meet the application requirements of high-precision stereo vision measurement systems.

Claims

1. A calibration method for a virtual binocular system based on infinity homography, characterized in that: The virtual binocular system is a biprism virtual binocular system. By introducing a biprism distortion model based on geometric optics into the image parameters acquired by the virtual binocular system, it is used to adapt to the asymmetric and anisotropic distortion characteristics caused by prism refraction, describe the image point offset caused by prism refraction and dispersion, and by introducing the infinity homography relation, the calibration point coordinates can be updated during the calibration iterative calculation process. Based on this relation, a reprojection error objective function for the mutual projection of the left and right eyes is constructed, thereby effectively addressing the impact of image distortion caused by biprism dispersion effect on the system calibration accuracy.

2. The virtual binocular system calibration method based on infinite homography as described in claim 1, characterized in that: The image parameters acquired by the virtual binocular system are processed using Zhang's calibration method to obtain the initial parameters and distortion parameters of the left and right virtual cameras, which serve as reliable initial values ​​for subsequent nonlinear optimization.

3. The virtual binocular system calibration method based on infinite homography as described in claim 1, characterized in that: The biprism distortion model: Among them, (v ud ,v ud f represents the coordinates of the corrected ideal image point in the image coordinate system. u and f v These represent the distortions along the horizontal axis u and the vertical axis v of the image, respectively. and is the biprism distortion coefficient.

4. The virtual binocular system calibration method based on infinite homography as described in claim 1, characterized in that: The virtual binocular system calibration method based on infinite-distance homography includes the following steps: Step 1: Establish the infinite homography matrix H between the left and right cameras using the planar target image. ∞ Based on the relationship of similar triangles in the polar plane, three-dimensional point reconstruction is achieved, and a reprojection error function of mutual projection is constructed. Step 4: Introduce epipolar geometric constraints, describe the epipolar relationship between left and right image points through the fundamental matrix, and use the vertical distance from the image point to the epipolar line as a geometric error term to enhance the geometric consistency and robustness of the system in stereo vision; Step 5: By comparing the actual physical distance between adjacent feature points on the calibration board with the reconstructed 3D point distance, a scale consistency error term is constructed to improve the physical scale accuracy of the calibration results and the reliability of actual measurements. Step 6: Combine the reprojection error, epipolar constraint error, and interval error according to their weights to construct the final nonlinear optimization objective function. By weighted fusion of multiple types of constraints, the overall optimization of system parameters is achieved. Step 7: Use the Levenberg-Marquardt algorithm to perform nonlinear iterative optimization on the joint objective function and comprehensively evaluate the results.

5. The virtual binocular system calibration method based on infinite homography as described in claim 3, characterized in that: The biprism distortion coefficient is determined by factors such as the biprism refractive index, prism angle, and the relative position of the prism and the camera.

6. The virtual binocular system calibration method based on infinite homography as described in claim 4, characterized in that: Step 1 uses the planar target image to establish an infinitely far homography relationship H between the image planes of the left and right virtual cameras. ∞ The initial relative rotation matrix R of the system is obtained from the initial parameters and distortion parameters of the left and right virtual cameras. The specific steps are as follows: (1) Infinite homography matrix H ∞ It describes the projection correspondence between two camera images when the scene plane is at infinity. In this case, the homography matrix depends only on the camera's intrinsic parameters and the relative rotation matrix. (2) Three-dimensional reconstruction using the homography relation at infinity: using H ∞ It enables 3D point reconstruction without prior knowledge of the geometric information of the object under test. The reconstruction process is based on the length ratio of similar triangles in the polar plane, so that the reconstructed 3D point coordinates can be directly used to construct reprojection errors and subsequent geometric constraints. The polar plane is formed by the optical center C of the left and right virtual cameras. l and C r and spatial point X w Together, we determine that the left camera image point x passes through H. ∞ After transformation, we obtain point x. ∞ =H ∞ x is the image point x of the left camera passing through H. ∞ The transformation corresponds to the point on the right camera image, which lies on the epipolar line of the right camera and represents the spatial point x. w Incident light | C l x w |The projection of a parallel ray at infinity onto the right image plane, and the left camera ray|C l x w | Intersects the right polar line at point m, and passes through similar triangle △x w x'm and △C r x'x ∞ Establish the following length relationships: At the same time, from similar triangles △c c e'm and △c p e'x ∞ get: Combining the two proportional relationships above, we get |x w c p |'s formula: When the depth |x of the spatial point is obtained w c p |, Space point X w The 3D point coordinates in the right camera coordinate system can then be reconstructed. Where d is the optical center c of the right camera R The normalized ray direction vector directed toward image point x' can be expressed as d = (K r ) -1 Therefore, the formula for reconstructing X in the right camera coordinate system is: Similarly, the formula for reconstructing X in the left camera coordinate system is: (3) Constructing the objective function for reprojection error (mutual projection): Reprojection error is the core objective of calibration optimization, measuring the reconstructed coordinates X of spatial points. w Reproject the image points back to the image plane, and calculate the deviation between these points and the actual observed image points, ensuring that the error is minimized while optimizing the parameters: Where N represents the total number of all feature points of the target, x rl The reconstructed point x in the right camera coordinate system wr The image point projected onto the left camera coordinate system, x lr The reconstructed point x in the left camera coordinate system wl Image points projected onto the right camera coordinate system.

7. The virtual binocular system calibration method based on infinite homography as described in claim 6, characterized in that: The specific steps of step 2 are as follows: Epipolar constraints are applied, utilizing the epipolar geometry between the virtual binocular cameras to ensure that the optimized parameters strictly satisfy the epipolar geometry of the stereo vision system. The perpendicular distance from the projected image points to the epipolar line is used as a strict geometric error function. In the virtual binocular system, the epipolar geometry between the left and right cameras is described by a 3x3 fundamental matrix F. For a point x in the left image and its matching point x' in the right image, the epipolar constraints are satisfied. x' T Fx=0 (8) Where F is determined by the intrinsic parameters K of the left and right cameras. l and K r The relative pose [R|T] is calculated to obtain: F=(K r ) -T [T] × R(K l ) -1 (9) The objective function expression for the polar constraint is:

8. The virtual binocular system calibration method based on infinite homography as described in claim 7, characterized in that: The specific steps of step 5 are as follows: A target spacing constraint was introduced. By comparing the actual physical spacing of feature points on the calibration board with the 3D coordinate spacing reconstructed based on the current parameters, a physical scale error term was constructed. This enhances the system's scale recovery capability and measurement accuracy during the optimization process. For any pair of adjacent feature points P on the planar target... i and P j Its actual physical distance is D norm The reconstructed coordinates of both in the left camera coordinate system are: and The reconstruction distance between the two is:

9. The virtual binocular system calibration method based on infinite homography as described in claim 8, characterized in that: The interval error term is defined as the average of the absolute values ​​of the differences between the true physical distance and the reconstructed distance for all adjacent point pairs, thereby imposing a consistency constraint on the physical scale during the optimization process. The interval error constraint formula is as follows:

10. The virtual binocular system calibration method based on infinite homography as described in claim 9, characterized in that: The synthetic joint optimization objective function described in step 4 combines the reprojection error function, epipolar constraint function, and margin error constraint from steps 1, 2, and 3 to form a complete nonlinear optimization objective function:

Images