Monocular camera pose estimation method based on feature point uncertainty

Abstract

The application discloses a monocular camera pose estimation method based on feature point uncertainty, comprising the following steps: constructing the covariance inverse matrix of the measurement error of the feature point according to the pixel gray distribution information of the feature point data space; performing singular value decomposition on the covariance inverse matrix of the feature point to obtain the affine transformation matrix of the feature point; performing uncertainty weighting to construct a weighted algebraic error function, obtaining a linear equation model according to the weighted algebraic error function, and rewriting the linear equation model into a nonlinear equation model; obtaining a consistent noise variance, solving the nonlinear equation model based on the consistent noise variance to obtain a closed-form solution with consistent bias, obtaining the estimated value of the rotation matrix and the estimated value of the translation vector according to the closed-form solution, and outputting the camera pose estimation result. The feature point uncertainty is considered and the asymptotic bias is eliminated, so that the accuracy and efficiency of the pose estimation are improved.

Description

Technical Field

[0001] This invention relates to a method for estimating the pose of a monocular camera based on the uncertainty of feature points, belonging to the field of visual measurement technology. Background Technology

[0002] Camera pose estimation has wide applications in 3D reconstruction, autonomous driving, camera calibration, augmented reality, and photogrammetry. The camera's pose can be described by a rotation matrix and a translation vector, and is also known as the camera's extrinsic parameters. The extrinsic parameters of the camera are solved by establishing a world coordinate system and a planar pixel coordinate system to describe the 3D-2D matching relationship of feature points; this process is called the PnP problem.

[0003] Most existing PnP algorithms obtain the optimal pose solution by establishing an algebraic error function and using iterative algorithms. However, these algorithms ignore the observation error of feature points and do not consider the uncertainty of feature points. In PnP algorithms that consider the uncertainty of feature points, it is generally assumed that the observation error is isotropic and independently and identically distributed. In actual feature point extraction, due to the large differences in the gray-level distribution patterns around the feature point, the error is often anisotropic and not independently and identically distributed. Summary of the Invention

[0004] Objective: To overcome the shortcomings of existing technologies, this invention provides a monocular camera pose estimation method based on feature point uncertainty. This method considers feature point uncertainty and eliminates asymptotic bias, thereby improving the accuracy and efficiency of pose estimation.

[0005] Technical solution: To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present invention provides a camera pose estimation method based on feature point uncertainty, comprising:

[0007] Based on the pixel grayscale distribution information in the feature point data space, the uncertainty information of the feature points is extracted, and the inverse covariance matrix Q of the measurement error of the feature points is constructed. -1 ;

[0008] The covariance inverse matrix Q of the feature points -1 Perform singular value decomposition to obtain the affine transformation matrix F of the feature points;

[0009] Based on the affine transformation matrix F of the feature points, an uncertainty weighting is applied to the feature points to construct a weighted algebraic error function;

[0010] The linear equation model is obtained based on the weighted algebraic error function;

[0011] Rewrite the linear equation model into a nonlinear equation model;

[0012] The uniform noise variance is obtained based on the nonlinear equation model, and the closed-form solution with consistent deviation is obtained by solving the nonlinear equation model based on the uniform noise variance.

[0013] According to the closed solution Obtain the estimated value of the rotation matrix. Estimates of translation vectors Output the camera pose estimation results.

[0014] In some embodiments, the uncertainty information of feature points is extracted based on the pixel grayscale distribution information in the feature point data space, and the inverse covariance matrix of the measurement error of the feature points is constructed, including:

[0015] The inverse covariance matrix Q of the measurement error at the feature points -1 Q describes the uncertainty of feature points. -1 The modeling is as follows:

[0016]

[0017] Where Q is the covariance matrix of the measurement error, W is the elliptical region centered at the feature point, and w(u,v) is the sum of the pixel gray levels in the elliptical region. and , respectively, are the gradients of the image in the u and v directions.

[0018] In some embodiments, the covariance inverse matrix Q of the feature points -1 Perform singular value decomposition to obtain the affine transformation matrix F of the feature points, including:

[0019] Q -1 Perform singular value decomposition to obtain Q. -1 =U∑ -1 U T ,in,

[0020] Affine transformation matrix

[0021] Where σ1 and σ2 represent the uncertainty of the feature points, ∑ -1 / 2 Transform the uncertain ellipse into a unit circle, U T It is a rotation matrix that rotates a tilted, uncertain ellipse into an ellipse with the same u and v directions as the image plane;

[0022] By projecting the feature point data space onto the weighted covariance space through the affine transformation matrix F, the measurement error of the feature points is transformed from anisotropic distribution to isotropic distribution.

[0023] In some embodiments, an uncertainty weighting is applied to the feature points based on the affine transformation matrix F to construct a weighted algebraic error function, including:

[0024] The camera pinhole imaging model is as follows:

[0025]

[0026] in, Here is the camera intrinsic parameter matrix, in pixels, where K is a known quantity; q i =[u i v i ] T Let i be the i-th pixel planar feature point. d is the reference point in the world coordinate system. i Represents depth information; R and t are the extrinsic parameters to be solved, representing the rotation matrix and translation vector, respectively;

[0027] Based on the affine transformation matrix F of the feature points, and considering the measurement error of the feature points, the camera pinhole imaging model is rewritten as follows:

[0028]

[0029] [u0 v0] T Given that simplifying the above equation and weighting the feature points, we obtain the observation equation:

[0030]

[0031] The weighted algebraic error function is:

[0032]

[0033] Among them, f x f is the pixel focal length in the x-direction. y Let ε be the pixel focal length in the y-direction, u0 be the center of the pixel plane in the u-direction, v0 be the center of the pixel plane in the v-direction, and ε be the focal length of the pixel plane in the y-direction. i Measurement error, W-index matrix E refers to the matrix The measurement error of the i-th pixel planar feature point is represented by... F represents the i-th weighted pixel planar feature point. i Let represent the affine transformation matrix of the i-th pixel planar feature point.

[0034] In some embodiments, obtaining a linear equation model based on a weighted algebraic error function includes:

[0035] Multiply both sides of the weighted algebraic error function by d i According to the camera pinhole imaging model, Intermediate parameter e = [0 0 1] T The measurement equation is obtained as follows:

[0036]

[0037] Rewriting the measurement equation in matrix form yields the linear equation model:

[0038]

[0039] Where M is the coefficient matrix, and x is composed of extrinsic parameters R and t. This is the error term; in I3 represents the tensor product of matrices, and I3 represents a 3×3 identity matrix.

[0040] In some embodiments, rewriting a linear equation model into a nonlinear equation model includes:

[0041] Introducing constraints and To eliminate scale ambiguity, the rotation matrix R is denoted as [r1r2 r3]. T The translation vector t = [t1 t2 t3] T ,have to Substituting into the linear equation model, we have:

[0042]

[0043] Rewritten in matrix form, we obtain the nonlinear equation model:

[0044] Aθ+η=b

[0045] in, It is a vector composed of weighted feature points. η is the error term, and θ is the unknown quantity to be determined. A is the coefficient matrix:

[0046]

[0047] Where α is a scaling factor, and n is the number of feature points. Represents the centroid coordinates of the feature point in the world coordinate system. Represents the coordinates of the i-th weighted pixel planar feature point in the u direction. This represents the coordinates of the i-th weighted pixel planar feature point in the v direction.

[0048] In some embodiments, a consistent noise variance is obtained based on a nonlinear equation model, and a closed-form solution with consistent deviation is obtained by solving the nonlinear equation model based on the consistent noise variance. include:

[0049] Constructing a matrix Define matrix and

[0050] Let A represent the centroid coordinates of the feature points in the world coordinate system, where A is the coefficient matrix and b is a vector composed of weighted feature points.

[0051] Construct a function H(λ) = Φ - λΔ, where λ is a generalized eigenvalue, and solve for the uniform noise variance.

[0052]

[0053] Based on the uniform noise variance, a closed-form solution with uniform deviation is obtained by solving the nonlinear equation model:

[0054]

[0055] In some embodiments, according to the closed-form solution Obtain the estimated value of the rotation matrix. Estimates of translation vectors include:

[0056]

[0057]

[0058]

[0059] Furthermore, it also includes: using singular value decomposition to... Projecting onto the Lie group space, the formula is as follows:

[0060]

[0061] in Indicates will The projection formula onto the Lie group space, where diag(*) denotes the diagonal matrix of (*) and det(*) denotes the determinant of (*), assuming... The singular value decomposition result is in and These are all parameters from the singular value decomposition results.

[0062] In a second aspect, the present invention provides a camera pose estimation device based on feature point uncertainty, including a processor and a storage medium;

[0063] The storage medium is used to store instructions;

[0064] The processor is configured to operate according to the instructions to execute the method according to the first aspect.

[0065] Thirdly, the present invention provides an apparatus comprising,

[0066] Memory;

[0067] processor;

[0068] as well as

[0069] Computer programs;

[0070] The computer program is stored in the memory and configured to be executed by the processor to implement the method described in the first aspect above.

[0071] Fourthly, the present invention provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0072] Beneficial Effects: The monocular camera pose estimation method based on feature point uncertainty provided by this invention has the following advantages: This invention considers the anisotropic noise of feature points, uses an algebraic error function based on the feature point uncertainty weighting coefficient, and establishes a monocular camera pose estimation model based on the feature point uncertainty weighting coefficient; it also derives a solution algorithm for consistent deviation; it establishes an uncertainty model through the image grayscale distribution information around the feature points, integrates it into the objective function of pose estimation, and introduces a least-squares solution to eliminate deviation, thereby eliminating asymptotic deviation of feature points, improving the accuracy of the algorithm, and accurately and reliably obtaining the camera pose. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of a method flow according to an embodiment of the present invention.

[0074] Figure 2 This is a schematic diagram illustrating the uncertainty of feature points according to an embodiment of the present invention. Detailed Implementation

[0075] The present invention will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be used to limit the scope of protection of the present invention.

[0076] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0077] In the description of this invention, the terms "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0078] Example 1

[0079] Firstly, such as Figure 1 As shown, this embodiment provides a camera pose estimation method based on feature point uncertainty, including:

[0080] S1 extracts the uncertainty information of feature points based on the pixel gray-level distribution information in the feature point data space, and constructs the inverse covariance matrix Q of the measurement error of the feature points. -1 ;

[0081] S2 represents the inverse covariance matrix Q of the feature points. -1 Perform singular value decomposition to obtain the affine transformation matrix F of the feature points;

[0082] S3 applies uncertainty weighting to the feature points based on the affine transformation matrix F of the feature points, and constructs a weighted algebraic error function.

[0083] S4 derives the linear equation model based on the weighted algebraic error function;

[0084] S5 rewrites the linear equation model into a nonlinear equation model;

[0085] S6 obtains the uniform noise variance based on the nonlinear equation model, and solves the nonlinear equation model based on the uniform noise variance to obtain a closed-form solution with consistent deviation.

[0086] S7 based on the closed-form solution Obtain the estimated value of the rotation matrix. Estimates of translation vectors Output the camera pose estimation results.

[0087] In some embodiments, step S1 extracts the uncertainty information of feature points based on the pixel grayscale distribution information in the feature point data space, and constructs the inverse covariance matrix of the measurement error of the feature points, including:

[0088] The feature point uncertainty described in this embodiment is [see...]. Figure 2 In practical visual measurements, different feature points have different grayscale pattern distributions on the imaging plane. Extracting image points introduces the directionality of this grayscale distribution, reflected in the u and v directions. Therefore, the uncertainty of image point errors can be used to characterize the anisotropy and non-independent identically distributed nature of the extraction error. Different image point errors have different amplitudes, anisotropic distributions, and uncertainties. Q -1 Describes the feature point q i =[u i v i ] T An uncertain ellipse centered at a circle, where the major and minor axes a and b represent the feature point q. i The uncertainty is represented by the angles of the major axis a and the minor axis b relative to directions u and v. Figure 2 The text primarily illustrates three different types of image point measurement error uncertainty. If the image point error is within... Figure 2 In case (II), it means that the uncertainty of the feature points is isotropic and there is no directional uncertainty, only scale uncertainty. If all image point errors are... Figure 2 (I) and Figure 2 In case (III), the uncertainty of the feature point is directional, and the uncertainty of the image point error direction needs to be considered in actual situations.

[0089] Q -1 The modeling is as follows:

[0090]

[0091] Where Q is the covariance matrix of the measurement error, W is the elliptical region centered at the feature point, and w(u,v) is the sum of the pixel gray levels in the elliptical region. and , respectively, are the gradients of the image in the u and v directions.

[0092] In some embodiments, step S2 involves the covariance inverse matrix Q of the feature points. -1 Perform singular value decomposition to obtain the affine transformation matrix F of the feature points, including:

[0093] Q -1 Perform singular value decomposition to obtain Q. -1 =U∑ -1 U T ,in, Define the transformation matrix

[0094] Where σ1 and σ2 represent the uncertainty of the feature points, ∑ -1 / 2 Transform the uncertain ellipse into a unit circle, U T It is a rotation matrix that rotates a tilted, uncertain ellipse into an ellipse with the same u and v directions as the image plane.

[0095] F is a 2×2 affine transformation matrix, which projects the coordinates of feature points into the weighted covariance space through the transformation F. At this time, the error can be considered to be isotropic.

[0096] In some embodiments, step S3 applies uncertainty weighting to the feature points based on the affine transformation matrix F of the feature points to construct a weighted algebraic error function, including:

[0097] An algebraic measurement error function is established based on the camera pinhole imaging model; the camera pinhole imaging model is shown below:

[0098]

[0099] in, Let K be the camera intrinsic parameter matrix, expressed in pixels. In this invention, K is a known quantity. i =[u i v i ] T For pixel plane feature points, d is the reference point in the world coordinate system. i Represents depth information. R and t are the extrinsic parameters for solving the problem, representing the rotation matrix and translation vector, respectively.

[0100] Based on the affine transformation matrix F of the feature points, and considering the measurement error of the feature points, the camera pinhole imaging model is rewritten as follows:

[0101]

[0102] Because of [u0 v0] T It is known that by simplifying the above equation and weighting the feature points, we can obtain the observation equation:

[0103]

[0104] The weighted algebraic error function is:

[0105]

[0106] Among them, f x f is the pixel focal length in the x-direction. yLet ε be the pixel focal length in the y-direction, u0 be the center of the pixel plane in the u-direction, v0 be the center of the pixel plane in the v-direction, and ε be the focal length of the pixel plane in the y-direction. i Measurement error, W-index matrix E refers to the matrix The measurement error of the i-th pixel planar feature point is represented by... F represents the i-th weighted pixel planar feature point. i Let represent the affine transformation matrix of the i-th pixel planar feature point.

[0107] Step S4, which obtains the linear equation model based on the weighted algebraic error function, includes:

[0108] Multiply both sides of the weighted algebraic error function by d i According to the camera pinhole imaging model, Intermediate parameter e = [0 0 1] T The measurement equation is obtained as follows:

[0109]

[0110] Rewriting the measurement equation in matrix form yields the linear equation model:

[0111]

[0112] Where M is the coefficient matrix, and x is composed of extrinsic parameters R and t. This is the error term; in I3 represents the tensor product of matrices, and I3 represents a 3×3 identity matrix.

[0113] In some embodiments, step S5 rewrites the linear equation model into a nonlinear equation model, including:

[0114] Introducing constraints and To eliminate scale ambiguity, the rotation matrix R is denoted as [r1 r2 r3]. T The translation vector t = [t1 t2 t3] T , can be obtained Substituting into the linear equation model, we have:

[0115]

[0116] Rewritten in matrix form, we obtain the nonlinear equation model:

[0117] Aθ+η=b

[0118] in, It is a vector composed of weighted feature points. η is the error term, and θ is the unknown quantity to be determined. A is the coefficient matrix:

[0119]

[0120] Where α is a scaling factor, and n is the number of feature points. Represents the centroid coordinates of the feature point in the world coordinate system. Represents the coordinates of the i-th weighted pixel planar feature point in the u direction. This represents the coordinates of the i-th weighted pixel planar feature point in the v direction.

[0121] In some embodiments, step S6 obtains the uniform noise variance based on the nonlinear equation model, and solves the nonlinear equation model based on the uniform noise variance to obtain a closed-form solution with consistent deviation. include:

[0122] Constructing a matrix Define matrix and Let A represent the centroid coordinates of the feature points in the world coordinate system, where A is the coefficient matrix and b is a vector composed of weighted feature points.

[0123] Construct a function H(λ) = Φ - λΔ, where λ is a generalized eigenvalue, and solve for its generalized eigenvalues ​​to obtain the consistent noise variance.

[0124] The closed-form solution of the nonlinear system with eliminated bias is as follows:

[0125] In some embodiments, step S7 is based on the closed-form solution. Obtain the estimated value of the rotation matrix. Estimates of translation vectors include:

[0126]

[0127]

[0128]

[0129] Due to the rotation matrix Not necessarily satisfied Therefore, it is necessary to use singular value decomposition to obtain the required conditions. The projection onto the Lie group space is as follows:

[0130]

[0131] in Indicates will The projection formula onto the Lie group space, where diag(*) denotes the diagonal matrix of (*) and det(*) denotes the determinant of (*), assuming... The singular value decomposition result is in and These are all parameters from the singular value decomposition results.

[0132] Example 2

[0133] Secondly, based on Embodiment 1, this embodiment provides a camera pose estimation device based on feature point uncertainty, including a processor and a storage medium;

[0134] The storage medium is used to store instructions;

[0135] The processor is configured to operate according to the instructions to execute the method according to Embodiment 1.

[0136] Example 3

[0137] Thirdly, based on Embodiment 1, this embodiment provides a device, including,

[0138] Memory;

[0139] processor;

[0140] as well as

[0141] Computer programs;

[0142] The computer program is stored in the memory and configured to be executed by the processor to implement the method described in Embodiment 1.

[0143] Example 4

[0144] Fourthly, based on Embodiment 1, this embodiment provides a storage medium on which a computer program is stored, and when the computer program is executed by a processor, it implements the method described in Embodiment 1.

[0145] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0146] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0147] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0148] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0149] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for monocular camera pose estimation based on feature point uncertainty, characterized in that, The method includes: Based on the pixel grayscale distribution information in the feature point data space, the uncertainty information of the feature points is extracted, and the inverse covariance matrix Q of the measurement error of the feature points is constructed. -1 This includes: the inverse covariance matrix Q of the measurement error through the feature points. -1 Q describes the uncertainty of feature points. -1 The modeling is as follows: Where Q is the covariance matrix of the measurement error, W is the elliptical region centered at the feature point, and w(u,v) is the sum of the pixel gray levels of the elliptical region. and These are the gradients of the image in the u and v directions, respectively; The covariance inverse matrix Q of the feature points -1 Perform singular value decomposition to obtain the affine transformation matrix F of the feature points; Based on the affine transformation matrix F of the feature points, an uncertainty weighting is applied to the feature points to construct a weighted algebraic error function; The linear equation model is obtained based on the weighted algebraic error function; Rewrite the linear equation model into a nonlinear equation model; The uniform noise variance is obtained based on the nonlinear equation model, and the closed-form solution with consistent deviation is obtained by solving the nonlinear equation model based on the uniform noise variance. According to the closed solution Obtain the estimated value of the rotation matrix. Estimates of translation vectors Output the camera pose estimation results.

2. The monocular camera pose estimation method based on feature point uncertainty according to claim 1, characterized in that, The covariance inverse matrix Q of the feature points -1 Perform singular value decomposition to obtain the affine transformation matrix F of the feature points, including: Q -1 Perform singular value decomposition to obtain Q. -1 =U∑ -1 U T ,in, Affine transformation matrix Where σ1 and σ2 represent the uncertainty of the feature points, ∑ -1 / 2 Transform the uncertain ellipse into a unit circle, U T It is a rotation matrix that rotates a tilted, uncertain ellipse into an ellipse with the same u and v directions as the image plane; By projecting the feature point data space onto the weighted covariance space through the affine transformation matrix F, the measurement error of the feature points is transformed from anisotropic distribution to isotropic distribution.

3. The monocular camera pose estimation method based on feature point uncertainty according to claim 1, characterized in that, The affine transformation matrix F of the feature points is used to apply uncertainty weighting to the feature points, constructing a weighted algebraic error function, including: The camera pinhole imaging model is as follows: in, Here is the camera intrinsic parameter matrix, in pixels, where K is a known quantity; q i =[u i v i ] T Let i be the i-th pixel planar feature point. d is the reference point in the world coordinate system. i Represents depth information; R and t are the extrinsic parameters to be solved, representing the rotation matrix and translation vector, respectively; Based on the affine transformation matrix F of the feature points, and considering the measurement error of the feature points, the camera pinhole imaging model is rewritten as follows: [u0 v0] T Given that simplifying the above equation and weighting the feature points, we obtain the observation equation: The weighted algebraic error function is: Among them, f x The pixel focal length in the x-direction, f y Let ε be the pixel focal length in the y-direction, u0 be the center of the pixel plane in the u-direction, v0 be the center of the pixel plane in the v-direction, and ε be the focal length of the pixel plane in the y-direction. i Measurement error, W-index matrix E refers to the matrix The measurement error of the i-th pixel planar feature point is represented by... F represents the i-th weighted pixel planar feature point. i Let represent the affine transformation matrix of the i-th pixel planar feature point.

4. The monocular camera pose estimation method based on feature point uncertainty according to claim 3, characterized in that, The linear equation model is obtained based on the weighted algebraic error function, including: Multiply both sides of the weighted algebraic error function by d i According to the camera pinhole imaging model, Intermediate parameter e = [001] T The measurement equation is obtained as follows: Rewriting the measurement equation in matrix form yields the linear equation model: Where M is the coefficient matrix, and x is composed of extrinsic parameters R and t. This is the error term; in I3 represents the tensor product of matrices, and I3 represents a 3×3 identity matrix.

5. The monocular camera pose estimation method based on feature point uncertainty according to claim 4, characterized in that, Rewriting linear equation models into nonlinear equation models includes: Introducing constraints and To eliminate scale ambiguity, the rotation matrix R is denoted as [r1 r2 r3]. T The translation vector t = [t1 t2 t3] T ,have to Substituting into the linear equation model, we have: Rewritten in matrix form, we obtain the nonlinear equation model: Aθ+η=b in, It is a vector composed of weighted feature points. η is the error term, and θ is the unknown quantity to be determined. A is the coefficient matrix: Where α is a scaling factor, and n is the number of feature points. Represents the centroid coordinates of the feature point in the world coordinate system. Represents the coordinates of the i-th weighted pixel planar feature point in the u direction. This represents the coordinates of the i-th weighted pixel planar feature point in the v direction.

6. The monocular camera pose estimation method based on feature point uncertainty according to claim 1, characterized in that, The uniform noise variance is obtained based on the nonlinear equation model, and the closed-form solution with consistent deviation is obtained by solving the nonlinear equation model based on the uniform noise variance. include: Constructing a matrix Define matrix and Let A represent the centroid coordinates of the feature points in the world coordinate system, where A is the coefficient matrix and b is a vector composed of weighted feature points. Construct a function H(λ) = Φ - λΔ, where λ is a generalized eigenvalue, and solve for the uniform noise variance. Solving the nonlinear equation model based on uniform noise variance yields a closed-form solution with consistent bias:

7. The monocular camera pose estimation method based on feature point uncertainty according to claim 1, characterized in that, According to the closed solution Obtain the estimated value of the rotation matrix. Estimates of translation vectors include:

8. The monocular camera pose estimation method based on feature point uncertainty according to claim 7, characterized in that, Also includes: Using singular value decomposition to... Projecting onto the Lie group space, the formula is as follows: in Indicates will The projection formula onto the Lie group space, where diag(*) represents the diagonal matrix of (*) and det(*) represents the determinant of (*). The singular value decomposition result is in and These are all parameters from the singular value decomposition results.

9. A storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the method described in any one of claims 1 to 8.

Images