A robot positioning method based on gradient weighting correction under multi-view

Abstract

The present application belongs to the technical field of visual guidance robot positioning, and particularly relates to a robot positioning method based on gradient weighting correction under multi-view. The present application firstly obtains the camera internal parameter matrix of each vision unit and the three-dimensional space coordinates of target features through calibrating the image of a calibration board in a common field of view; then establishes the matrix conversion relationship and inverse solution equation between the three-dimensional space and the two-dimensional imaging plane, and gradually obtains the external parameter matrix of each auxiliary camera relative to the main camera under the same world coordinate system; then constructs a gradient weighting correction algorithm using the gradient amplitude and phase correlation between the image pairs, filters out the error caused by the too large camera offset angle, and is used for correcting the conversion matrix between the multi-view vision units; finally, the data is converted to the robot base coordinate system through a hand-eye calibration algorithm, and the robot precise positioning based on gradient weighting correction under multi-view is realized. The present application increases the stability and robustness of the multi-view visual guidance robot positioning.

Description

Technical Field

[0001] This invention belongs to the field of vision-guided robot localization technology, specifically relating to a robot localization method based on gradient weighted correction under multiple perspectives. Background Technology

[0002] In the field of vision-guided robot localization technology, the accuracy of visual calibration is a significant factor limiting robot performance. Calibration using multi-view systems composed of multiple vision units further restricts the accuracy of robot localization due to excessive angular variations between cameras.

[0003] Existing methods for improving the positioning accuracy of vision-guided robots mostly employ multi-sensor dynamic fusion to improve calibration accuracy by estimating pose uncertainty. However, these methods suffer from difficulties in data transmission and communication control between different sensors. Summary of the Invention

[0004] To address the multi-view calibration problem among identical sensors, this invention proposes a robot localization method based on gradient weighted correction under multiple viewpoints. By constructing the phase correlation between image pairs in the frequency domain and using the weighted image gradient, the method aims to correct the pose transformation matrix between cameras, thereby increasing the stability and robustness of multi-view guided robot localization and solving the matrix error problem caused by excessive camera angle offset.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A robot localization method based on gradient weighted correction under multiple views, characterized by the following steps:

[0007] Step 1: Three cameras are arranged in a convergent manner to form a multi-view vision unit. All cameras are arranged in a straight line, with each camera sharing a common field of view. All cameras have the same parameters. The camera intrinsic parameter matrix and the three-dimensional spatial coordinates of the target features are obtained by calibrating a checkerboard pattern.

[0008] Step 2: Establish the matrix transformation relationship between the three-dimensional space and the two-dimensional imaging plane and solve the equations in reverse to obtain the pose relationship matrix of all auxiliary cameras in the world coordinate system with the main camera C1 as the main camera.

[0009] Step 3: Process the target image pairs acquired by the multi-view vision unit and calculate the directional gradient G between each image pair. x G y and gradient magnitude |G(x,y)|;

[0010] Step 4: Use Fourier transform to convert the target image from the spatial domain to the frequency domain, and calculate the normalized cross power spectrum Cp(u,v) based on the phase correlation between image pairs;

[0011] Step 5: Perform inverse Fourier transform on the normalized cross power spectrum to obtain the rotation angle and translation parameters between each pair of images acquired by the auxiliary camera C2 relative to the main camera C1 and the auxiliary camera C3 relative to the main camera C1, respectively.

[0012] Step 6: Based on the directional gradient and gradient magnitude obtained in Step 3, weight all W sets of acquired target image pairs to construct a weighted average gradient magnitude |G|. W (x,y)|;

[0013] Step 7: Based on the gradient magnitude and phase correlation obtained in Steps 3 to 5, and combined with the weighted gradient magnitude in Step 6, construct a gradient weighted correction function to obtain the rotation angle and translation parameters between W sets of image pairs. This is used to filter out the error caused by excessive camera offset and correct the transformation matrix.

[0014] Step 8: Combine the rotation and translation parameters obtained from the gradient weighted correction in Step 7;

[0015] Step 9: Transform the 3D coordinates of the target feature points obtained after the correction matrix to the robot's base coordinate system to realize the vision-guided robot localization function. The specific process is as follows:

[0016] a. The calibration calculation process for "eye outside hand" is shown in formula (13):

[0017]

[0018] In formula (13), (x o ,y o ,z o (x, y) represents the robot's base coordinates, and (x, y) represents the main camera's pixel coordinates. The rotation matrix from the main camera coordinates to the robot coordinates. The translation matrix from the main camera coordinates to the robot coordinates. express Parameters obtained after matrix operations;

[0019] b. Establish the relationship matrix between the coordinate system of the main camera C1 and the robot base coordinate system, as shown in formula (14):

[0020]

[0021] The beneficial effects of this invention are as follows: First, by calibrating the calibration board image within the common field of view, the intrinsic parameter matrices of each camera in the visual unit and the three-dimensional spatial coordinates of the target features are obtained. Then, the matrix transformation relationship and inverse equation between the three-dimensional space and the two-dimensional imaging plane are established, gradually obtaining the extrinsic parameter matrix of each auxiliary camera relative to the main camera in the same world coordinate system. Next, a gradient-weighted correction algorithm is constructed using the gradient magnitude and phase correlation between image pairs to filter out errors caused by excessive camera offset angles, and is used to correct the transformation matrix between multi-view visual units. Finally, the data is transformed to the robot's base coordinate system using a hand-eye calibration algorithm, achieving precise robot localization based on gradient-weighted correction under multi-view conditions. The method proposed in this invention fully considers the matrix error problem caused by excessive offset angles during multi-view imaging, and effectively improves the stability and robustness of the algorithm through the gradient-weighted correction function, possessing strong practical value. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the robot localization method based on gradient weighted correction under multiple perspectives in an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram of the multi-view vision unit described in an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of the reverse calibration extrinsic parameter matrix as described in an embodiment of the present invention;

[0025] Figure 4 This is a schematic diagram of the directional gradient and gradient magnitude described in the embodiments of the present invention;

[0026] Figure 5 This is a schematic diagram of the rotation and translation transformation results described in the embodiments of the present invention;

[0027] Figure 6 This is a schematic diagram of the "eye outside the hand" calibration as described in an embodiment of the present invention. Detailed Implementation

[0028] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0029] Please see Figure 1 The present invention provides a flowchart of a robot localization method based on gradient weighted correction under multiple views, comprising the following steps:

[0030] Step 1: Select three CMOS cameras to form a multi-view vision unit. Obtain the camera intrinsic parameter matrix and the three-dimensional spatial coordinates of the target features by calibrating a checkerboard pattern. This invention can operate with three or more cameras, named as follows: the first camera is the main camera C1, and subsequent cameras are auxiliary cameras Ci, where i takes values ​​from 2 to N. All cameras are arranged sequentially along a straight line, ensuring that each subsequent camera shares a common field of view with the preceding camera. This invention can also use other types of cameras, such as CCD cameras, as long as all cameras are of the same model. This invention uses three CMOS cameras as an example.

[0031] Specifically, in this embodiment, the three CMOS cameras are arranged sequentially in a converging configuration (the cameras are cross-mounted, and their optical axes converge at a common point) to form a multi-view vision unit, as follows: Figure 2 As shown, the cameras in the multi-view vision unit have a common field of view, and the field of view covers the entire surface of the target workpiece.

[0032] a. Select the main camera C1 and the auxiliary camera C2 to form a convergent binocular camera. Based on the coordinate system transformation relationship during camera imaging, the transformation relationship between the binocular camera coordinate system and the image coordinate system is obtained as shown in formula (1):

[0033]

[0034] In formula (1), Let f be the camera intrinsic parameter matrix, f be the camera focal length, (u, v) be the pixel coordinates of the image center point, and Z be the ordinate of the spatial point in the camera coordinate system. A spatial point is a point in space, which can be represented in the form (X, Y, Z) when transformed into the camera coordinate system. The spatial point in the camera coordinate system of the main camera C1 is represented as [X...]. l ,Y l Z l ], represented in the camera coordinate system of auxiliary camera C2 as [X r ,Y r Z r ]. Z l The Z-coordinate represents the ordinate of the main camera C1. r Z represents the ordinate of the adjacent auxiliary camera C2. rr This represents the ordinate of auxiliary camera C3. If more auxiliary cameras are added, the order can be sequentially updated. In other words, the subscript in this invention indicates the parameter of the corresponding camera, such as f. rr This indicates the focal length of the auxiliary camera C3. This invention uses the same model and has consistent internal camera parameters. In this invention, the topmost camera is selected as the main camera. Figure 2 As shown.

[0035] b. Transform the pose relationship between the camera coordinate systems using the transformation matrix. It means that, among them The external parameter rotation matrix of the auxiliary camera C2 relative to the main camera C1 is represented by r1 to r9, which are rotation matrix parameters obtained by checkerboard calibration. Let C2 represent the translation vector of the extrinsic parameters of the auxiliary camera C2 relative to the main camera C1. Solve equation (1) simultaneously and combine it with the two-dimensional coordinates (x, y) of the image coordinate system. l ,y l ) and (x r ,y r The relationship matrix between the binocular cameras is obtained as shown in formula (2):

[0036]

[0037] In formula (2), f is the camera focal length, r is the rotation matrix parameter, t is the translation vector parameter, and Z is the ordinate of the spatial point in the camera coordinate system.

[0038] c. Using the camera relationship matrix formula (2), obtain the three-dimensional coordinates of the four non-coplanar target feature points. Spatial feature points (The subscript n in d indicates the quantity, i.e., the number of feature points. The superscript w indicates the world coordinate system.) and camera mapping points. There is a transformation relationship between them as shown in formula (3), and the reference point P m From feature point d n The weighted average is obtained.

[0039]

[0040] In formula (3), and β represents the coordinates of the reference point in the world coordinate system and the camera coordinate system of the main camera C1, respectively. mn Represents the homogeneous barycenter coordinates, derived from the feature point d. n The coordinates are uniquely determined, n = 4. m represents the number of reference points; any four feature points can determine one reference point. Since there can be an infinite number of feature points in space, the range of values ​​for the reference point m can be expressed as m = 1, 2, 3, ..., m. Therefore, by calibrating the checkerboard pattern, the intrinsic parameter matrices of the main camera C1 and the auxiliary camera C2, the extrinsic parameter matrix between the two cameras, and the spatial three-dimensional coordinates of the target feature points are obtained.

[0041] Step 2: Establish the matrix transformation relationship between the three-dimensional space and the two-dimensional imaging plane and solve the equations in reverse, as follows: Figure 3 As shown, the pose relationship matrix of all auxiliary cameras in the world coordinate system with the main camera C1 as the reference is obtained step by step.

[0042] a. Given the camera intrinsic parameter matrix, establish the homogeneous transformation matrix relationship between corresponding points of two-dimensional pixel coordinates and three-dimensional camera coordinates as shown in formula (4):

[0043]

[0044] In formula (4), f is the camera focal length, (u0, v0) represents the pixel coordinates of the image center point, (u m ,v m ) represents the pixel coordinates of the corresponding point, β mn Let s represent the coordinates of the homogeneous barycenter. m This represents the weighted value of the ordinate of a point in the camera coordinate system, where m represents the number of reference points.

[0045] b. Establish the reference point with the centroid removed in the world coordinate system and camera coordinate system matrix representation, and use singular value decomposition to inversely obtain the rotation matrix and translation vector [R3|T3] of the auxiliary camera C3 relative to the main camera C1, as shown in formula (5):

[0046]

[0047] In formula (5), and These represent the coordinates of the reference point in the world coordinate system and the camera coordinate system of the main camera C1, respectively; both are known quantities. The centroid coordinates of the reference point are... and

[0048] Step 3: Process the target image pairs acquired by the multi-view vision unit and calculate the directional gradient G between each image pair. x G y And the gradient magnitude |G(x,y)|, as follows Figure 4 As shown.

[0049] a. Let f(x,y) represent the original image acquired by camera C1. Let the gradient of point A(x,y) in the image be represented by a vector, and the gradient expression is shown in formula (6):

[0050]

[0051] In formula (6), G[f(x,y)] represents the image gradient, which includes the rate of change of gray level at point A(x,y) along the x and y directions, pointing in the direction of the greatest change;

[0052] b. The gradient magnitude of the image containing the function f(x,y) is calculated using the backward difference approximation, as shown in formula (7):

[0053]

[0054] In formula (7), |G(x,y)| represents the gradient magnitude, also known as the gradient modulus, and |f(x+1,y)-f(x,y)| represents the gradient G of the image along the horizontal direction. x |f(x,y+1)-f(x,y)| represents the gradient G of the image along the vertical direction. y .

[0055] Step 4: Use Fourier transform to convert the target image from the spatial domain to the frequency domain, and calculate the normalized cross power spectrum Cp(u,v) based on the phase correlation between image pairs.

[0056] a. Assume that the digital image f(x,y) is an M×N matrix composed of discrete gray-level information, and its spectrum after Fourier transform is shown in formula (8):

[0057]

[0058] In formula (8), u = 0, 1, 2, ..., M-1 and v = 0, 1, 2, ..., N-1 are called frequency variables;

[0059] b. Suppose that the target image f'(x',y') acquired by the auxiliary camera C2 is obtained by translating (x0,y0) and rotating θ0 the image f(x,y) acquired by the main camera C1. Transform the rectangular coordinates during rotation to the polar coordinate system, and express the polar coordinates as x = rcosθ, y = rsinθ. Combining equation (8) yields the spectrum expression before and after the change, as shown in equation (9):

[0060]

[0061] c. According to the properties of Fourier transform, the Fourier transform amplitude is not affected after the image is rotated and translated. The normalized cross power spectrum of the image acquired by the auxiliary camera C2 relative to the image acquired by the main camera C1 is calculated as shown in the following formula (10), and the phase correlation of the two images in polar coordinates and rectangular coordinates is obtained.

[0062]

[0063] In formula (10), F * (u,v) is the complex conjugate of F(u,v). yes The conjugate of complex numbers.

[0064] Specifically, in this embodiment, the normalized cross-power spectrum between the image acquired by the auxiliary camera C3 and the image acquired by the main camera C1 is the same as the process described above for calculating the normalized cross-power spectrum between the image acquired by the auxiliary camera C2 and the image acquired by the main camera C1.

[0065] Step 5: Perform inverse Fourier transform on the normalized cross power spectrum to obtain the rotation angle and translation parameters between each pair of images acquired by the auxiliary camera C2 relative to the main camera C1 and the auxiliary camera C3 relative to the main camera C1.

[0066] Specifically, in this embodiment, an inverse Fourier transform Cp is performed on the cross-power spectrum Cp(u,v). -1 (u,v)=δ([r,θ+θ0]+[x-x0,y-y0]) gives the coordinates of the point where the amplitude is the maximum, which is the rotation angle θ0 and translation parameter (x0,y0) between the two images;

[0067] Step 6: Based on the directional gradient and gradient magnitude obtained in Step 3, weight all W sets of acquired target image pairs to construct a weighted average gradient magnitude |G|. W (x,y)|.

[0068] Specifically, in this embodiment, W sets of calibration board sequence images are used to construct directional gradient weighted images. and As shown in formula (11):

[0069]

[0070] Step 7: Based on the gradient magnitude and phase correlation obtained in Steps 3 to 5, and combined with the weighted gradient magnitude in Step 6, construct a gradient weighted correction function to obtain the rotation angle and translation parameters between W sets of image pairs. This is used to filter out the error caused by excessive camera offset and correct the transformation matrix.

[0071] Step 8: Combine the rotation and translation parameters obtained from the gradient weighted correction in Step 7, and introduce them into the process of inversely solving the extrinsic parameter matrix of the auxiliary camera C3 relative to the main camera C1 to improve the camera calibration accuracy and algorithm robustness.

[0072] Specifically, in this embodiment, by combining formulas (8), (10), and (11), the weighted gradient magnitude is calculated using fast Fourier transform and inverse Fourier transform of the cross power spectrum, resulting in the transformation relationship between the imaging planes of the multi-view camera after gradient weighting correction, as shown in formula (12):

[0073]

[0074] In formula (12), M×N represents an image composed of an M-row, N-column pixel matrix, W represents the number of image groups acquired by the main camera C1, auxiliary cameras C2 and C3, (x,y) represents the pixel coordinates of the imaging plane of auxiliary camera C3, and (x',y') represents the pixel coordinates of the imaging plane of the main camera C1 after transformation. ij ,y ij ) and θ ijThe translation and rotation angle from the i-th image to the j-th image are represented as follows: Figure 5 As shown;

[0075] Steps 1 and 9 calibrated the pose relationship between the main camera C1 and the auxiliary camera C2, obtaining the initial 3D coordinates of the target feature points in the world coordinate system. Step 2 used inverse equation solving, and with the known 3D coordinates of the target feature points in the world coordinate system and the known intrinsic parameter matrix of the auxiliary camera C3, obtained the initial pose relationship of the auxiliary camera C3 relative to the main camera C1. Steps 3 to 8 then corrected the initial pose relationship of the auxiliary camera C3 relative to the main camera C1 obtained in step 2 by constructing a gradient-weighted suppression function, i.e., the rotation matrix and translation matrix, which is equivalent to correcting the 3D coordinates of the target feature points. The 3D coordinates of the target feature points obtained after the correction matrix are transformed into the robot's base coordinate system to realize the localization function of the vision-guided robot.

[0076] a. The transformation relationship between the robot's end-effector coordinate system (TCP-based coordinate system) and the robot's base coordinate system is known, and the poses of all cameras in the multi-view vision unit are transformed to the same world coordinate system, as follows: Figure 6 As shown, the calibration process for "eye outside hand" is as shown in formula (13):

[0077]

[0078] In formula (13), (x o ,y o ,z o (x, y) represents the robot's base coordinates, and (x, y) represents the main camera's pixel coordinates. This is the rotation matrix from the master camera (subscript C is the abbreviation for camera, and superscript O is the abbreviation for robot) coordinates to the robot coordinates. The translation matrix from the main camera coordinates to the robot coordinates. express The parameters obtained after matrix operations. The rotation matrix between the camera coordinate system of the main camera C1 and the robot's base coordinate system. Multiply by the known two-dimensional coordinates of the image coordinate system of the main camera C1 in step 1, and then add the translation matrix from the camera coordinate system of the main camera C1 to the robot's base coordinate system. The resulting 3x3 matrix represents the values ​​of parameters a1 to c2.

[0079] b. Based on the multi-view vision unit calibration principle, establish the relationship matrix between the coordinate system of the main camera C1 and the robot base coordinate system, as shown in formula (14):

[0080]

[0081] The converted 3D coordinate data is input into the robot simulation software to correct the coordinates of the corresponding marker points in the 3D model. After obtaining the workpiece's pose relative to the robot's base coordinates in the real environment, the machining path is programmed offline using the robot simulation software to achieve vision-guided robot positioning and machining operations.

[0082] The main innovations of this invention are: 1) It establishes a robot localization method based on gradient weighted correction under multiple views in the field of vision-guided robot localization; 2) It makes full use of the phase correlation of image pairs in the frequency domain to construct a gradient weighted correction function, which improves the calibration accuracy of multi-view vision units and the robustness of vision-guided localization algorithms; 3) It achieves the goal of correcting the pose transformation matrix between camera pairs through algorithms such as weighted interval gradient magnitude and Fourier transform and inverse transform processing.

[0083] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A robot localization method based on gradient weighted correction under multiple views, characterized in that: Includes the following steps: Step 1: Three cameras are arranged in a convergent manner to form a multi-view vision unit. The cameras are arranged in a straight line, with each camera sharing a common field of view with the previous camera. All cameras have the same parameters. The camera intrinsic parameter matrix and the three-dimensional spatial coordinates of the target features are obtained by calibrating a checkerboard pattern. Step 2: Establish the matrix transformation relationship between the three-dimensional space and the two-dimensional imaging plane and solve the equations in reverse to obtain the pose relationship matrix of all auxiliary cameras in the world coordinate system with the main camera C1 as the main camera. Step 3: Process the target image pairs acquired by the multi-view vision unit and calculate the directional gradient between each image pair. Gx , Gy and gradient magnitude | G ( x , y )|; Step 4: Use Fourier transform to convert the target image from the spatial domain to the frequency domain, and calculate the normalized cross-power spectrum based on the phase correlation between image pairs. Cp ( u , v ); Step 5: Perform inverse Fourier transform on the normalized cross power spectrum to obtain the rotation angle and translation parameters between each pair of images acquired by the auxiliary camera C2 relative to the main camera C1 and the auxiliary camera C3 relative to the main camera C1, respectively. Step 6: Based on the directional gradient and gradient magnitude obtained in Step 3, weight all the acquired gradients... W For each pair of target images, construct a weighted average gradient magnitude. GW ( x , y )|; Step 7: Based on the gradient magnitude and phase correlation obtained in steps 3 to 5, and combined with the weighted gradient magnitude from step 6, construct a gradient weighted correction function to obtain... W The rotation angle and translation parameters between image pairs are used to filter out errors caused by excessive camera offset and to correct the transformation matrix; Step 8: Combine the rotation and translation parameters obtained from the gradient weighted correction in Step 7; Official (12) In formula (12), M × N The image is represented by M OK N Composed of a column of pixel matrices, W This indicates the number of image groups acquired by the main camera C1, auxiliary cameras C2 and C3. x , y ) represents the pixel coordinates of the C3 imaging plane of the auxiliary camera, ( x’ , y’ ) represents the pixel coordinates of the main camera's C1 imaging plane after transformation. xij , yij )and θij Indicates the first i Image to the first j The translation and rotation angle of the image; Step 9: Transform the 3D coordinates of the target feature points obtained after the correction matrix to the robot's base coordinate system to realize the vision-guided robot localization function. The specific process is as follows: a. The calibration calculation process for "eye outside hand" is shown in formula (13): Official (13) In formula (13), For robot base coordinates, ( x , y ) are the main camera pixel coordinates. The rotation matrix from the main camera coordinates to the robot coordinates. The translation matrix from the main camera coordinates to the robot coordinates. express Parameters obtained after matrix operations; b. Establish the relationship matrix between the coordinate system of the main camera C1 and the robot base coordinate system, as shown in formula (14): Official (14).

2. The robot localization method based on gradient weighted correction under multiple views according to claim 1, characterized in that: The camera is a CMOS camera.

3. A robot localization method based on gradient weighted correction under multiple views, as described in claim 1 or 2, characterized in that: Step 1 is as follows: a. Select the main camera C1 and the auxiliary camera C2 to form a converging binocular camera, and obtain the transformation relationship between the binocular camera coordinate system and the image coordinate system: Official (1) In formula (1), Represents the camera intrinsic parameter matrix. f For the camera focal length, ( u , v () represents the pixel coordinates of the image center point. Z [X] represents the ordinate of a point in the camera coordinate system; the point in the camera coordinate system of the main camera C1 is represented as [X]. l ,Y l Z l ], represented in the camera coordinate system of auxiliary camera C2 as [X r ,Y r, Z r ]; b. Transform the pose relationship between the camera coordinate systems using the transformation matrix. It means that, among them This represents the extrinsic rotation matrix of the auxiliary camera C2 relative to the main camera C1. r 1 to r 9 represents the rotation matrix parameter. The translation vector of the extrinsic parameters of the auxiliary camera C2 relative to the main camera C1 is used to obtain the relationship matrix between the two cameras: Official (2) In formula (2), f For camera focal length, r For the rotation matrix parameters, t For translation vector parameters, Z This represents the ordinate of a point in the camera coordinate system. c. Using the camera relationship matrix formula (2), obtain the three-dimensional coordinates of the four non-coplanar target feature points. Spatial feature points and camera mapping point There is a transformation relationship between them as shown in formula (3), with reference point Pm From feature points dn Weighted average yields: Official (3) In formula (3), and These represent the coordinates of the reference point in the world coordinate system and the camera coordinate system of the main camera C1, respectively. βmn Represents the homogeneous barycenter coordinates, derived from the feature points. dn The only certainty, n = 4; m This represents the number of reference points.

4. The robot localization method based on gradient weighted correction under multiple views according to claim 3, characterized in that: Step 2 is as follows: a. Given the camera intrinsic parameter matrix, establish the homogeneous transformation matrix relationship between corresponding points of two-dimensional pixel coordinates and three-dimensional camera coordinates as shown in formula (4): Official (4) In formula (4), f For the camera focal length, ( u 0, v 0) represents the pixel coordinates of the image center point, ( um , vm () represents the pixel coordinates of the corresponding point. βmn Represents the coordinates of the homogeneous barycenter. sm This represents the weighted value of the ordinate of a point in the camera coordinate system. m The number of reference points; b. Establish the reference point with its centroid removed in the world coordinate system and camera coordinate system as a matrix representation, and use singular value decomposition to inversely obtain the rotation matrix and translation vector of the auxiliary camera C3 relative to the main camera C1. R 3 | T 3], as shown in formula (5): Official (5) In formula (5), and These represent the coordinates of the reference point in the world coordinate system and the camera coordinate system of the main camera C1, respectively, both of which are known quantities.

5. The robot localization method based on gradient weighted correction under multiple views according to claim 4, characterized in that: Step 3 is as follows: a. Using functions f ( x , y () represents the original image acquired by camera C1, and let the midpoint of the image be... A ( x , y The gradient of is represented by a vector, and its expression is shown in formula (6): Official (6) In formula (6), G [ f ( x , y [)] represents the image gradient, including points A ( x , y grayscale along ) x direction and y The rate of change of direction, pointing in the direction of the greatest change; b. Use backward difference approximation to calculate the function. f ( x , y The gradient magnitude of the image containing the gradient is shown in formula (7): Official (7) In formula (7), | G ( x , y | represents the magnitude of the gradient, also known as the gradient's modulus. f ( x +1, y )- f ( x , y | represents the gradient of the image along the horizontal direction. Gx , | f ( x , y +1)- f ( x , y | represents the gradient of the image along the vertical direction. Gy .

6. The robot localization method based on gradient weighted correction under multiple views according to claim 5, characterized in that: Step 4 is as follows: a) Assuming a digital image f ( x , y It is composed of discrete grayscale information. M × N The spectrum of the matrix after its Fourier transform is shown in formula (8): Official (8) In formula (8), u =0,1,2,..., M -1, v =0,1,2,..., N -1 is called the frequency variable; b. The target image acquired by auxiliary camera C2 f’ ( x’ , y’ Images acquired by the main camera C1 f ( x , y Translation x 0, y 0) and rotation θ After obtaining 0, the rectangular coordinates during rotation are transformed to polar coordinates, and the polar coordinates are expressed as follows: x = r cos θ , y = r sin θ , u=ωcosφ , v=ωsinφ Combining equation (8), we obtain the spectrum expression before and after the change, as shown in equation (9): Official (9) c. Calculate the normalized cross power spectrum of the image acquired by the auxiliary camera C2 relative to the image acquired by the main camera C1, and obtain the phase correlation of the two images in polar coordinates and rectangular coordinates. Official (10) In formula (10), F * ( u , v )yes F ( u , v The conjugate of complex numbers, F * ( ω , φ )yes F ( ω , φ The conjugate of complex numbers; The normalized cross-power spectrum between the image acquired by the auxiliary camera C3 and the image acquired by the main camera C1 is obtained through the same process as the normalized cross-power spectrum between the image acquired by the auxiliary camera C2 and the image acquired by the main camera C1.

7. The robot localization method based on gradient weighted correction under multiple views according to claim 6, characterized in that: Step 5 is as follows: Cross-power spectrum Cp ( u , v Perform inverse Fourier transform Cp -1 ( u , v ) = δ ([ r , θ + θ 0]+[ x - x 0, y - y The coordinates of the point where the amplitude is maximum are the rotation angle between the two images. θ 0 and translation parameters ( x 0, y 0).

8. The robot localization method based on gradient weighted correction under multiple views according to claim 7, characterized in that: Step 6 is as follows: Assume the calibration plate sequence images total... W Groups, constructing directional gradient weighted groups and As shown in formula (11): Official (11).

Images