Camera extrinsic parameter calibration method based on epipolar constraint and terminal

Abstract

The invention discloses a camera external parameter calibration method and terminal based on epipolar constraint, and the method comprises the steps: obtaining a baseline length between two to-be-calibrated cameras, and respectively shooting a target scene through the two to-be-calibrated cameras to obtain a left image and a right image; performing feature extraction and feature matching on the left image and the right image to obtain feature point pairs; performing epipolar constraint calculation according to the internal parameters of the two cameras to be calibrated and the feature point pairs to obtain a target rotation matrix and an original translation vector of the two cameras to be calibrated; and performing scale recovery on the original translation vector through the baseline length to obtain a target translation vector. According to the invention, precise calibration of the external parameters of the binocular camera can be realized in a long-distance scene.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a camera extrinsic parameter calibration method and terminal based on epipolar constraints. Background Art

[0002] In the field of computer vision and photogrammetry, the extrinsic calibration of the binocular camera system is a key step in 3D reconstruction and measurement. Traditional binocular camera extrinsic calibration methods mainly include the following: calibration object-based calibration method, self-calibration method, and motion-based calibration method. The calibration object-based calibration method refers to the use of special calibration objects to complete the extrinsic calibration, and the calibration object is generally a chessboard or dot array. The self-calibration method does not require special calibration objects, but uses natural features in the scene to match multi-frame image sequences. The motion-based calibration method is to achieve calibration by moving the camera system.

[0003] The calibration method based on calibration objects requires special calibration objects. When shooting long-distance scenes with telephoto cameras, the small imaging scale of the long-distance target easily leads to the failure of corner detection, so this method requires the customization of a larger chessboard. However, large-sized chessboards have the disadvantages of being difficult to manufacture (it requires a larger manufacturing platform and more sophisticated processing equipment, and it is necessary to ensure the high accuracy and uniformity of large-area chessboard patterns), inconvenient to move and use (the weight and volume of large chessboards will increase, which is not conducive to moving and shooting them), and difficult material selection (it is necessary to select materials with strong rigidity, high surface flatness and economical and practical materials, and common materials are easy to deform when manufactured in large sizes).

[0004] The self-calibration method has high requirements on scene texture, so it does not work well in environments with single texture, and has high computational complexity and is easily affected by noise. In long-distance scenes, due to small parallax, the accuracy is often insufficient.

[0005] Motion-based calibration methods require precise camera motion, which is difficult to achieve in some application scenarios (such as fixed-mounted surveillance systems). They require high accuracy in feature point tracking and are easily affected by occlusion and lighting changes. In long-distance scenes, small tracking errors may lead to large calibration errors.

[0006] In summary, the current external parameter calibration methods of binocular cameras have certain limitations in long-distance scenes, especially in applications such as power transmission line monitoring. Due to factors such as high installation position, long monitoring distance, and complex environment, traditional methods are difficult to meet the requirements of fast, accurate, and easy operation. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a camera extrinsic parameter calibration method and terminal based on epipolar constraints, which can realize accurate calibration of binocular camera extrinsic parameters in long-distance scenes.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0009] A camera extrinsic calibration method based on epipolar constraints, comprising:

[0010] Obtaining a baseline length between two cameras to be calibrated, and photographing a target scene using the two cameras to be calibrated to obtain a left image and a right image;

[0011] Performing feature extraction and feature matching on the left image and the right image respectively to obtain feature point pairs;

[0012] Obtaining target rotation matrices and original translation vectors of the two cameras to be calibrated by performing epipolar constraint calculations according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs;

[0013] The original translation vector is scaled using the baseline length to obtain a target translation vector.

[0014] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0015] A camera extrinsic calibration terminal based on epipolar constraints comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the following steps when executing the computer program:

[0016] Obtaining a baseline length between two cameras to be calibrated, and photographing a target scene using the two cameras to be calibrated to obtain a left image and a right image;

[0017] Performing feature extraction and feature matching on the left image and the right image respectively to obtain feature point pairs;

[0018] Obtaining target rotation matrices and original translation vectors of the two cameras to be calibrated by performing epipolar constraint calculations according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs;

[0019] The original translation vector is scaled using the baseline length to obtain a target translation vector.

[0020] The beneficial effects of the present invention are as follows: images of the target scene are captured respectively by two cameras to be calibrated, and epipolar constraint calculations are performed based on feature point pairs between the images of the target scene to obtain the target rotation matrix and the original translation vector. Since the original translation vector obtained by the epipolar constraint calculation has scale uncertainty, the baseline length between the two cameras to be calibrated is used to restore the scale of the original translation vector to obtain the true target translation vector. Compared with the traditional calibration method, the present invention does not need to rely on specific calibration objects, which reduces the calibration cost and difficulty, and uses feature point pairs in the target scene image for calibration, reducing dependence on specific scenes. The present invention only performs external parameter calibration through two images, which solves the problem that the traditional calibration method is limited in application in large-scale environments and expands the application range of binocular cameras. In addition, the present invention effectively restores the scale information of the translation vector through the baseline length to ensure the accuracy of the calibration result. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A flowchart of a camera extrinsic calibration method based on epipolar constraints provided in an embodiment of the present invention;

[0022] Figure 2 A schematic diagram of the installation of a binocular camera provided by an embodiment of the present invention;

[0023] Figure 3 A schematic diagram of the structure of a camera extrinsic calibration terminal based on epipolar constraints provided in an embodiment of the present invention;

[0024] Description of labels:

[0025] 100. A camera extrinsic calibration terminal based on epipolar constraints; 101. Memory; 102. Processor. DETAILED DESCRIPTION

[0026] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0027] An embodiment of the present invention provides a camera extrinsic calibration method based on epipolar constraints, comprising:

[0028] Obtaining a baseline length between two cameras to be calibrated, and photographing a target scene using the two cameras to be calibrated to obtain a left image and a right image;

[0029] Performing feature extraction and feature matching on the left image and the right image respectively to obtain feature point pairs;

[0030] Obtaining target rotation matrices and original translation vectors of the two cameras to be calibrated by performing epipolar constraint calculations according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs;

[0031] The original translation vector is scaled using the baseline length to obtain a target translation vector.

[0032] From the above description, it can be seen that the beneficial effects of the present invention are: the images of the target scene are respectively captured by two cameras to be calibrated, and the epipolar constraint calculation is performed based on the feature point pairs between the images of the target scene to obtain the target rotation matrix and the original translation vector. Since the original translation vector obtained by the epipolar constraint calculation has scale uncertainty, the baseline length between the two cameras to be calibrated is used to restore the scale of the original translation vector to obtain the true target translation vector. Compared with the traditional calibration method, the present invention does not need to rely on specific calibration objects, which reduces the calibration cost and difficulty, and uses the feature point pairs in the target scene image for calibration, reducing the dependence on specific scenes. The present invention only performs external parameter calibration through two images, which solves the problem that the traditional calibration method is limited in application in large-scale environments and expands the application range of binocular cameras. In addition, the present invention effectively restores the scale information of the translation vector through the baseline length to ensure the accuracy of the calibration result.

[0033] Further, the extracting and matching features of the left image and the right image to obtain feature point pairs respectively includes:

[0034] Extracting feature points of the left image and the right image respectively by using a SIFT algorithm, and calculating descriptors corresponding to the feature points;

[0035] Matching the feature points of the left image and the right image using a feature matching algorithm according to the descriptor to obtain a candidate feature point pair;

[0036] The candidate feature point pairs are screened by the RANSAC algorithm to obtain feature point pairs.

[0037] From the above description, it can be seen that the SIFT algorithm is used to detect the feature points of the two images to achieve feature point positioning and improve the stability of subsequent matching. The matching relationship between these feature points is determined by the feature matching algorithm, so as to determine the image point pairs obtained by shooting the same point with different cameras, thereby describing the geometric relationship between the two cameras based on the image points and improving the accuracy of subsequent external parameter calibration.

[0038] Furthermore, the step of performing epipolar constraint calculation according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs to obtain the target rotation matrix and the original translation vector of the two cameras to be calibrated includes:

[0039] Normalizing the feature point pairs according to the intrinsic parameters of the two cameras to be calibrated;

[0040] Constructing an essential matrix equation group according to the epipolar constraint condition and the normalized characteristic point pairs, and solving the essential matrix equation group to obtain an essential matrix;

[0041] Decomposing the essential matrix by singular value to obtain a plurality of candidate rotation matrices and candidate translation vectors;

[0042] Detecting depth information of any group of the feature point pairs respectively in the multiple candidate rotation matrices and candidate translation vectors;

[0043] If the depth information is a positive value, it is determined that the candidate rotation matrix and the candidate translation vector corresponding to the depth information are the target rotation matrix and the original translation vector of the two cameras to be calibrated.

[0044] From the above description, we can know that the epipolar constraint refers to the constraint formed by the image point and the camera optical center under the projection model when the same point is projected onto two images with different perspectives. The system of equations about the essential matrix is ​​constructed through the epipolar constraint conditions to solve the essential matrix, and then the rotation matrix and translation vector are obtained by singular value decomposition of the essential matrix. Since there are multiple solutions obtained by decomposition, the depth information of the feature point pairs is detected to verify the correctness of the solution, thereby ensuring the accuracy of the external parameter calibration of the binocular camera.

[0045] Further, the scaling of the original translation vector by using the baseline length to obtain the target translation vector includes:

[0046] After the original translation vector is normalized, the baseline length is multiplied by the normalized original translation vector to obtain a target translation vector.

[0047] From the above description, it can be seen that when constructing the essential matrix, the absolute distance of the feature point pair in the world coordinate system is unknown, and the absolute scale of the translation vector cannot be directly restored from the essential matrix. Therefore, the actual size of the translation vector is directly represented by the baseline length, thereby restoring the absolute scale of the translation vector and improving the accuracy of the external parameter calibration.

[0048] Furthermore, obtaining the baseline length between two cameras to be calibrated includes:

[0049] The baseline length between the two cameras to be calibrated is determined by a monocular camera ranging algorithm or a rangefinder.

[0050] From the above description, it can be seen that the baseline length between the two cameras to be calibrated is determined by the monocular convergence ranging algorithm, thereby reducing equipment costs. The baseline length between the two cameras to be calibrated is directly measured by the rangefinder, thereby reducing computing resource consumption and improving calibration efficiency.

[0051] Another embodiment of the present invention provides a camera extrinsic calibration terminal based on epipolar constraints, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the following steps when executing the computer program:

[0052] Obtaining a baseline length between two cameras to be calibrated, and photographing a target scene using the two cameras to be calibrated to obtain a left image and a right image;

[0053] Performing feature extraction and feature matching on the left image and the right image respectively to obtain feature point pairs;

[0054] Obtaining target rotation matrices and original translation vectors of the two cameras to be calibrated by performing epipolar constraint calculations according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs;

[0055] The original translation vector is scaled using the baseline length to obtain a target translation vector.

[0056] From the above description, it can be seen that the beneficial effects of the present invention are: the images of the target scene are respectively captured by two cameras to be calibrated, and the epipolar constraint calculation is performed based on the feature point pairs between the images of the target scene to obtain the target rotation matrix and the original translation vector. Since the original translation vector obtained by the epipolar constraint calculation has scale uncertainty, the baseline length between the two cameras to be calibrated is used to restore the scale of the original translation vector to obtain the true target translation vector. Compared with the traditional calibration method, the present invention does not need to rely on specific calibration objects, which reduces the calibration cost and difficulty, and uses the feature point pairs in the target scene image for calibration, reducing the dependence on specific scenes. The present invention only performs external parameter calibration through two images, which solves the problem that the traditional calibration method is limited in application in large-scale environments and expands the application range of binocular cameras. In addition, the present invention effectively restores the scale information of the translation vector through the baseline length to ensure the accuracy of the calibration result.

[0057] Further, the extracting and matching features of the left image and the right image to obtain feature point pairs respectively includes:

[0058] Extracting feature points of the left image and the right image respectively by using a SIFT algorithm, and calculating descriptors corresponding to the feature points;

[0059] Matching the feature points of the left image and the right image using a feature matching algorithm according to the descriptor to obtain a candidate feature point pair;

[0060] The candidate feature point pairs are screened by the RANSAC algorithm to obtain feature point pairs.

[0061] From the above description, it can be seen that the SIFT algorithm is used to detect the feature points of the two images to achieve feature point positioning and improve the stability of subsequent matching. The matching relationship between these feature points is determined by the feature matching algorithm, so as to determine the image point pairs obtained by shooting the same point with different cameras, thereby describing the geometric relationship between the two cameras based on the image points and improving the accuracy of subsequent external parameter calibration.

[0062] Furthermore, the step of performing epipolar constraint calculation according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs to obtain the target rotation matrix and the original translation vector of the two cameras to be calibrated includes:

[0063] Normalizing the feature point pairs according to the intrinsic parameters of the two cameras to be calibrated;

[0064] Constructing an essential matrix equation group according to the epipolar constraint condition and the normalized characteristic point pairs, and solving the essential matrix equation group to obtain an essential matrix;

[0065] Decomposing the essential matrix by singular value to obtain a plurality of candidate rotation matrices and candidate translation vectors;

[0066] Detecting depth information of any group of the feature point pairs respectively in the multiple candidate rotation matrices and candidate translation vectors;

[0067] If the depth information is a positive value, it is determined that the candidate rotation matrix and the candidate translation vector corresponding to the depth information are the target rotation matrix and the original translation vector of the two cameras to be calibrated.

[0068] From the above description, we can know that the epipolar constraint refers to the constraint formed by the image point and the camera optical center under the projection model when the same point is projected onto two images with different perspectives. The system of equations about the essential matrix is ​​constructed through the epipolar constraint conditions to solve the essential matrix, and then the rotation matrix and translation vector are obtained by singular value decomposition of the essential matrix. Since there are multiple solutions obtained by decomposition, the depth information of the feature point pairs is detected to verify the correctness of the solution, thereby ensuring the accuracy of the external parameter calibration of the binocular camera.

[0069] Further, the scaling of the original translation vector by using the baseline length to obtain the target translation vector includes:

[0070] After the original translation vector is normalized, the baseline length is multiplied by the normalized original translation vector to obtain a target translation vector.

[0071] From the above description, it can be seen that when constructing the essential matrix, the absolute distance of the feature point pair in the world coordinate system is unknown, and the absolute scale of the translation vector cannot be directly restored from the essential matrix. Therefore, the actual size of the translation vector is directly represented by the baseline length, thereby restoring the absolute scale of the translation vector and improving the accuracy of the external parameter calibration.

[0072] Furthermore, obtaining the baseline length between two cameras to be calibrated includes:

[0073] The baseline length between the two cameras to be calibrated is determined by a monocular camera ranging algorithm or a rangefinder.

[0074] From the above description, it can be seen that the baseline length between the two cameras to be calibrated is determined by the monocular convergence ranging algorithm, thereby reducing equipment costs. The baseline length between the two cameras to be calibrated is directly measured by the rangefinder, thereby reducing computing resource consumption and improving calibration efficiency.

[0075] The embodiments of the present invention provide a camera extrinsic parameter calibration method and terminal based on epipolar constraints, which can be applied to the scenario of power transmission line monitoring, and can realize the accurate calibration of binocular camera extrinsic parameters in long-distance scenarios. The following is an explanation through specific embodiments:

[0076] Please refer to Figure 1 to Figure 2 , Embodiment 1 of the present invention is:

[0077] A camera extrinsic calibration method based on epipolar constraints, comprising:

[0078] S1. Obtain a baseline length between two cameras to be calibrated, and use the two cameras to be calibrated to respectively photograph a target scene to obtain a left image and a right image.

[0079] In some embodiments, before step S1, a static scene with rich textures is selected, and two cameras to be calibrated are respectively used to take a series of overlapping images from different positions and angles, and the number of images is greater than 20, to ensure that there is enough overlapping area between the images. The self-calibration method based on the bundle adjustment method explores the geometric relationship between multiple views, and then derives the intrinsic parameter matrix K and distortion coefficient d of the two cameras to be calibrated.

[0080] In some embodiments, after determining the intrinsic parameter matrix and distortion coefficient of the two cameras to be calibrated, the two cameras to be calibrated are respectively installed on the left and right sides of the transmission line tower to form a binocular camera detection. During the installation process, the shooting positions of the two cameras to be calibrated need to be adjusted so that the target scene to be observed is located in the center of the camera screen.

[0081] Specifically, in step S1, obtaining the baseline length between two cameras to be calibrated includes:

[0082] S11. Determine the baseline length between two cameras to be calibrated by using a monocular camera ranging algorithm or a rangefinder.

[0083] In some embodiments, the rangefinder is a laser rangefinder. Figure 2 As shown in the figure, after the two cameras to be calibrated are installed and fixed on the transmission line tower, a baffle with a marking point is first set above the right camera to receive the laser of the laser rangefinder, and then a laser rangefinder with an adjustable bracket is set above the left camera, and the laser rangefinder is turned on and adjusted so that the laser is aligned with the marking point in the baffle above the right camera. At this time, the distance value obtained by the laser rangefinder is the baseline length between the two cameras to be calibrated.

[0084] S2. Perform feature extraction and feature matching on the left image and the right image respectively to obtain feature point pairs.

[0085] Specifically, the step S2 includes:

[0086] S21. Extract feature points of the left image and the right image respectively by using a SIFT algorithm, and calculate descriptors corresponding to the feature points.

[0087] In some embodiments, the descriptor is a detailed mathematical representation of each detected feature point, including the location, scale, and orientation information of the feature point, as well as the gradient information of the image area around the feature point.

[0088] S22. Match the feature points of the left image and the right image using a feature matching algorithm according to the descriptor to obtain candidate feature point pairs.

[0089] In some embodiments, the feature matching algorithm may include brute force matching, KD tree and FLANN. KD tree is a data structure that partitions K-dimensional data space and is mainly used for searching key data in multidimensional space. FLANN is an efficient numerical or string matching algorithm library.

[0090] S23, screening the candidate feature point pairs by using the RANSAC algorithm to obtain feature point pairs.

[0091] It should be noted that the RANSAC algorithm is an iterative method for estimating model parameters in noisy data. Since there may be a large number of mismatches in the process of matching feature point pairs, the RANSAC algorithm is needed to screen the candidate feature point pairs to ensure that the matched feature point pairs are correct as much as possible.

[0092] S3. Perform epipolar constraint calculation according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs to obtain the target rotation matrix and the original translation vector of the two cameras to be calibrated.

[0093] Specifically, step S3 includes:

[0094] S31. Normalizing the feature point pairs according to the intrinsic parameters of the two cameras to be calibrated.

[0095] In some embodiments, for any pair of feature points p1 (left image) and p2 (right image), the feature points in the images taken by the two cameras to be calibrated are normalized according to the intrinsic parameters of the two cameras to be calibrated. Specifically, in the left image, Among them, x1 represents the normalized coordinates of the feature point p1 in the left image, represents the inverse matrix of the intrinsic parameter matrix K1 of the left camera; in the right image, Among them, x2 represents the normalized coordinates of the feature point p2 in the left image. represents the inverse matrix of the intrinsic parameter matrix K2 of the left camera. The coordinates of the feature point pair p1 (left image) and p2 (right image) after normalization are x1 = [u1, v1, 1] T , x2=[u2,v2,1] T .

[0096] S32, constructing an essential matrix equation group according to the epipolar constraint condition and the normalized characteristic point pairs, and solving the essential matrix equation group to obtain an essential matrix.

[0097] In some embodiments, according to the epipolar constraint, it can be obtained:

[0098]

[0099] Among them, the essential matrix E = [e1, e2, e3, e4, e5, e6, e7, e8, e9] T Therefore, the above epipolar constraint formula (1) can be converted to [u1u2,u1v2,u1,v1u2,v1v2,v1,u2,v2,1]·e=0.

[0100] Similarly, the above method can be used to transform other feature point pairs. If 8 pairs of feature point pairs are used for transformation, the essential matrix equation group is:

[0101]

[0102] Among them, u i and v iRepresents the coordinate value of the ith feature point pair. Thus, each element e1, e2, …, e9 in the essential matrix E can be solved. If the number of feature point pairs exceeds 8, each element e1, e2, …, e9 in the essential matrix E can be solved by the least squares method or nonlinear optimization algorithm.

[0103] S33. Decomposing the essential matrix by singular value decomposition to obtain multiple candidate rotation matrices and candidate translation vectors.

[0104] In some embodiments, if the singular value decomposition of the essential matrix E is E=UΣV T , where U and V represent orthogonal matrices, Σ=diag(σ,0,0), diag represents a diagonal matrix, and σ represents the singular value of the essential matrix E. For any essential matrix E, there are two possible rotation matrices and translation vectors corresponding to it, which are specifically expressed as:

[0105]

[0106] in, represents the rotation matrix obtained by rotating 90 degrees along the Z axis. Since the essential matrix E is equivalent to the essential matrix E when it takes a negative sign, the same result can be obtained for any t taking a negative sign.

[0107] Therefore, two candidate rotation matrices and two candidate translation vectors can be decomposed from the essential matrix E.

[0108] S34, detecting depth information of any group of the feature point pairs in the multiple candidate rotation matrices and candidate translation vectors respectively.

[0109] In some embodiments, any set of feature point pairs is substituted into the calculation formula of the candidate rotation matrix and the candidate translation vector to obtain the depth information of the set of feature point pairs under the two cameras to be calibrated.

[0110] S35: If the depth information is a positive value, determine that the candidate rotation matrix and the candidate translation vector corresponding to the depth information are the target rotation matrix and the original translation vector of the two cameras to be calibrated.

[0111] S4. Rescaling the original translation vector using the baseline length to obtain a target translation vector.

[0112] Specifically, step S4 includes:

[0113] S41. After normalizing the original translation vector, multiply the baseline length by the normalized original translation vector to obtain a target translation vector.

[0114] In some embodiments, according to the epipolar constraint It can be seen that the original translation vector t obtained by decomposing the essential matrix E is ∧ The scale and direction of the original translation vector t ∧ The direction of can be determined by specifying the positive and negative direction of the X-axis component of the translation vector from the left camera to the right camera. ∧ The scale of the original translation vector t ∧ Normalize the original translation vector t ∧ The length is 1, determined by multiplying the baseline length.

[0115] In some embodiments, the triangulation principle can be combined to perform three-dimensional reconstruction, estimate the spatial position of the feature point pair, and then obtain the relative size of the target object. Combined with the actual size of the known target object, the scale factor is calculated to adjust the scale of the translation vector and achieve scale recovery.

[0116] Please refer to Figure 3 , Embodiment 2 of the present invention is:

[0117] A camera extrinsic calibration terminal 100 based on epipolar constraint includes a memory 101, a processor 102, and a computer program stored in the memory 101 and running on the processor 102. When the processor 102 executes the computer program, each step of a camera extrinsic calibration method based on epipolar constraint in embodiment 1 is implemented.

[0118] In summary, the present invention provides a camera extrinsic parameter calibration method and terminal based on epipolar constraints. The two cameras to be calibrated only need to take images of the target scene respectively, and perform epipolar constraint calculations based on feature point pairs between the images of the target scene to obtain the target rotation matrix and the original translation vector. There is no need to rely on specific calibration objects, which reduces the calibration cost and difficulty. In addition, the calibration is performed using feature point pairs in the target scene image, which reduces dependence on specific scenes. Since the original translation vector obtained by epipolar constraint calculation has scale uncertainty, the baseline length between the two cameras to be calibrated is used to restore the scale of the original translation vector to obtain the true target translation vector. There is no need for complex scale recovery calculations, which reduces the consumption of computing resources while ensuring the accuracy of the calibration results. The present invention performs extrinsic parameter calibration using only two images, which solves the problem of limited application of traditional calibration methods in large-scale environments and expands the application range of binocular cameras.

[0119] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A camera extrinsic calibration method based on epipolar constraints, characterized in that: include: Obtaining a baseline length between two cameras to be calibrated, and photographing a target scene using the two cameras to be calibrated to obtain a left image and a right image; Performing feature extraction and feature matching on the left image and the right image respectively to obtain feature point pairs; Obtaining target rotation matrices and original translation vectors of the two cameras to be calibrated by performing epipolar constraint calculations according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs; The original translation vector is scaled using the baseline length to obtain a target translation vector.

2. The camera extrinsic calibration method based on epipolar constraints according to claim 1, characterized in that: The extracting and matching the features of the left image and the right image to obtain the feature point pairs comprises: Extracting feature points of the left image and the right image respectively by using a SIFT algorithm, and calculating descriptors corresponding to the feature points; Matching the feature points of the left image and the right image using a feature matching algorithm according to the descriptor to obtain a candidate feature point pair; The candidate feature point pairs are screened by the RANSAC algorithm to obtain feature point pairs.

3. The camera extrinsic calibration method based on epipolar constraints according to claim 1, characterized in that: The step of performing epipolar constraint calculation according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs to obtain the target rotation matrix and the original translation vector of the two cameras to be calibrated comprises: Normalizing the feature point pairs according to the intrinsic parameters of the two cameras to be calibrated; Constructing an essential matrix equation group according to the epipolar constraint condition and the normalized characteristic point pairs, and solving the essential matrix equation group to obtain an essential matrix; Decomposing the essential matrix by singular value to obtain a plurality of candidate rotation matrices and candidate translation vectors; Detecting depth information of any group of the feature point pairs respectively in the multiple candidate rotation matrices and candidate translation vectors; If the depth information is a positive value, it is determined that the candidate rotation matrix and the candidate translation vector corresponding to the depth information are the target rotation matrix and the original translation vector of the two cameras to be calibrated.

4. The camera extrinsic calibration method based on epipolar constraints according to claim 1, characterized in that: The scaling of the original translation vector by using the baseline length to obtain the target translation vector comprises: After the original translation vector is normalized, the baseline length is multiplied by the normalized original translation vector to obtain a target translation vector.

5. The camera extrinsic calibration method based on epipolar constraints according to claim 1, characterized in that: The method of obtaining the baseline length between two cameras to be calibrated includes: The baseline length between the two cameras to be calibrated is determined by a monocular camera ranging algorithm or a rangefinder.

6. A camera extrinsic calibration terminal based on epipolar constraints, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the following steps are implemented: Obtaining a baseline length between two cameras to be calibrated, and photographing a target scene using the two cameras to be calibrated to obtain a left image and a right image; Performing feature extraction and feature matching on the left image and the right image respectively to obtain feature point pairs; Obtaining target rotation matrices and original translation vectors of the two cameras to be calibrated by performing epipolar constraint calculations according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs; The original translation vector is scaled using the baseline length to obtain a target translation vector.

7. The camera extrinsic calibration terminal based on epipolar constraints according to claim 6, characterized in that: The extracting and matching the features of the left image and the right image to obtain the feature point pairs comprises: Extracting feature points of the left image and the right image respectively by using a SIFT algorithm, and calculating descriptors corresponding to the feature points; Matching the feature points of the left image and the right image using a feature matching algorithm according to the descriptor to obtain a candidate feature point pair; The candidate feature point pairs are screened by the RANSAC algorithm to obtain feature point pairs.

8. The camera extrinsic calibration terminal based on epipolar constraints according to claim 6, characterized in that: The step of performing epipolar constraint calculation according to the intrinsic parameters of the two cameras to be calibrated and the feature point pairs to obtain the target rotation matrix and the original translation vector of the two cameras to be calibrated comprises: Normalizing the feature point pairs according to the intrinsic parameters of the two cameras to be calibrated; Constructing an essential matrix equation group according to the epipolar constraint condition and the normalized characteristic point pairs, and solving the essential matrix equation group to obtain an essential matrix; Decomposing the essential matrix by singular value to obtain a plurality of candidate rotation matrices and candidate translation vectors; Detecting depth information of any group of the feature point pairs respectively in the multiple candidate rotation matrices and candidate translation vectors; If the depth information is a positive value, it is determined that the candidate rotation matrix and the candidate translation vector corresponding to the depth information are the target rotation matrix and the original translation vector of the two cameras to be calibrated.

9. The camera extrinsic calibration terminal based on epipolar constraints according to claim 6, characterized in that: The scaling of the original translation vector by using the baseline length to obtain the target translation vector comprises: After the original translation vector is normalized, the baseline length is multiplied by the normalized original translation vector to obtain a target translation vector.

10. The camera extrinsic calibration terminal based on epipolar constraints according to claim 6, characterized in that: The method of obtaining the baseline length between two cameras to be calibrated includes: The baseline length between the two cameras to be calibrated is determined by a monocular camera ranging algorithm or a rangefinder.

Images