Binocular camera synchronous automatic calibration system and calibration method

Abstract

The invention relates to the technical field of computer vision and image measurement, and discloses a binocular camera synchronous automatic calibration system and a calibration method. The inner ring flange plate is fixed on the base and is used for fixing M calibration plates; the outer ring flange plate is arranged on the base, rotates around the radius tangential direction of the inner ring flange plate and is used for arranging and driving N groups of binocular cameras; the cross-shaped sliding table assembly is arranged on the inner ring flange plate and is used for driving the calibration plate to carry out position transformation; the camera bracket is fixed on the outer ring flange plate and is used for fixing a binocular camera; the control and data processing unit is connected with the outer ring flange plate, the cross-shaped sliding table assembly and the binocular camera and used for controlling the outer ring flange plate to rotate and controlling the calibration plate to move, and the binocular camera collects images and then completes calibration. According to the invention, automatic acquisition of multiple groups of images with different poses can be completed in one rotation period, full-automatic calibration of multiple groups of binocular cameras is realized, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer vision and image measurement, and in particular to a system and method for synchronous automatic calibration of a binocular camera. Background Art

[0002] Under the wave of industrialization, binocular positioning is rapidly penetrating into the capillaries of discrete manufacturing and smart terminals along the three main lines of automation, onlineization, and miniaturization. For example, in industrial robots, binocular positioning provides millimeter-level three-dimensional grasping coordinates for collaborative arms, and the production line rhythm is synchronized with the calibration rhythm; in autonomous driving, the on-board binocular system compensates for temperature drift and vibration through online self-calibration; in security monitoring, the RDID drone detection system uses binocular positioning to achieve real-time three-dimensional positioning of the drone and the operator; in consumer electronics, AR / VR headsets use miniaturized binocular modules, and complete 0.1 mm-level spatial tracking calibration through a fully automatic calibration station before leaving the factory.

[0003] Binocular calibration is a prerequisite for obtaining three-dimensional geometric information using stereo vision. Its accuracy directly determines the reliability of subsequent ranging, reconstruction, or positioning. Traditional methods rely on manually or robot-assisted placement of a planar checkerboard target. After acquiring images in 10-20 different poses, the Zhang Zhengyou calibration method and stereo calibration process are used to determine internal and external parameters. However, this paradigm presents three significant issues when applied to mass production: First, manual intervention leads to uncontrollable timing, and the acquisition process is time-consuming and labor-intensive, making online calibration difficult on mass production lines. Second, factors such as illumination variations, target contamination, and motion blur lead to variable image quality and fluctuating reprojection errors. This makes it difficult to simultaneously capture the full field of view of the binocular camera at varying depths and angles within a confined space, resulting in large errors in the extrapolation of calibration parameters. Third, offline calibration results lack traceability. If the camera undergoes micro-displacement due to temperature drift or mechanical vibration, rework and re-calibration are necessary, resulting in high after-sales costs. In addition, existing methods are designed for visible light cameras and lack support for multimodal systems such as infrared, fisheye, and RGB-D. Distortion models and baseline compensation strategies need to be customized for each model, further limiting production flexibility.

[0004] Therefore, researching a fully automatic, high-precision and scalable dual-target positioning technology has become an urgent need to promote the mass implementation of stereo vision products. Summary of the Invention

[0005] This invention provides a system and method for the simultaneous automated calibration of binocular cameras. These methods can automatically capture multiple sets of images in different poses within a single rotation cycle, enabling fully automated calibration of multiple binocular cameras. This eliminates the efficiency bottlenecks and human errors associated with manually placing calibration plates and triggering capture. This method fundamentally suppresses reprojection bias caused by inconsistent image quality, meeting the efficiency requirements of mass production lines while significantly improving calibration accuracy and efficiency.

[0006] The present invention provides a system for synchronous automatic calibration of a binocular camera, comprising: base; an inner ring flange, the inner ring flange being fixed on the base and being used to fix M calibration plates; An outer ring flange, which is arranged on the base and rotates around the inner ring flange, and is used to arrange and drive N groups of binocular cameras to rotate around the calibration plate; A cross slide assembly, the cross slide assembly being arranged on the inner ring flange, the calibration plate being fixed thereon, and being used to drive the calibration plate to change its position in the front-back and left-right directions so as to adjust the optical axis distance between the binocular camera and the calibration plate; A camera bracket, fixed on the outer ring flange, for placing and fixing the binocular camera; A control and data processing unit is fixed inside the base and connected to the outer ring flange, the cross slide assembly, and the binocular camera. The control and data processing unit is used to control the rotation of the outer ring flange, the movement of the calibration plate, and the binocular camera to capture images, and calculate the internal and external parameters of the binocular camera based on the collected image set to achieve automatic calibration of the binocular camera.

[0007] Furthermore, the inner ring flange is circular in shape, the outer ring flange is annular in shape, and the inner ring of the outer ring flange surrounds the outer circumference of the inner ring flange; The calibration plate is a checkerboard calibration plate or a center-shaped calibration plate, and the number of the calibration plates and the number of the binocular cameras are both greater than or equal to 2.

[0008] Furthermore, the calibration plate and the cross-slide assembly are connected through a calibration plate bracket. A plurality of cross-slide assemblies are provided on the inner ring flange. Each cross-slide assembly corresponds to a calibration plate bracket and a calibration plate. The heights of the plurality of calibration plate brackets are different, and the height difference of the plurality of calibration plate brackets is Δh, and Δh satisfies Δh=tanθ*D, wherein D is the closest distance from the calibration plate to the optical center of the binocular camera, θ is the vertical field of view angle of the binocular camera, and the pitch angle can be adjusted independently to simulate different field of view inclination angles.

[0009] Furthermore, the process of controlling the binocular camera to collect images in the control and data processing unit is as follows: after the binocular camera detects the calibration plate, the binocular camera is controlled to synchronously collect the image of the calibration plate and record the current calibration plate area S n With the center point coordinate O n At the same time, the cross slide assembly starts to move backward slowly and uniformly. When the new calibration plate area S is detected n+1 With S n The difference is greater than the corresponding area threshold T s , or the new center point coordinates O n+1 With O n The difference is greater than the corresponding coordinate point deviation threshold O s When the calibration plate changes its spatial position, the image of the current calibration plate is acquired; otherwise, the image acquisition signal is not triggered.

[0010] Furthermore, in the control and data processing unit, after the binocular camera calibration image acquisition starts, the outer ring flange is controlled to rotate at a constant speed within one rotation cycle, and the binocular camera performs real-time detection of the calibration plate; when the outer ring flange completes one rotation cycle, the binocular camera is controlled to stop real-time detection and acquisition of the image.

[0011] Furthermore, in the control and data processing unit, when the calibration plate is a checkerboard calibration plate, Harris corner detection or Shi-Tomasi algorithm is used to detect potential corners in the image, valid areas are extracted through morphological operations and contour analysis, and valid corners are screened using quadrilateral fitting and geometric constraints; When the calibration plate is a circular center-type calibration plate, the bright / dark circular spots in the LoG or DoG detection image are used as the results of blob extraction. Then, the neighborhood of each blob is binarized and the contour is extracted. The least squares method is used to fit an ellipse, and the coordinates and radius of the circle center are calculated. Valid circle points are screened through geometric symmetry and grayscale distribution, and the extracted circle center is optimized at the sub-pixel level using the grayscale centroid method.

[0012] Furthermore, when calculating the internal reference in the dual-target calibration, When the calibration plate is a checkerboard calibration plate, call the Zhang Zhengyou calibration algorithm to perform corner detection → homography matrix estimation → closed-form solution of the extrinsic parameter matrix → reprojection error minimization on each image, thereby outputting the extrinsic parameter matrix K and distortion coefficient D of the left and right cameras; When the calibration plate is a circular center-type calibration plate, ellipse detection → eccentricity compensation → circle center positioning are performed on the circular center-type calibration plate image to obtain the image center coordinates. The correspondence between the ellipse detection and the world coordinates of the known planar circular array is used to initially estimate the intrinsic and extrinsic parameters through the homography matrix. The camera intrinsic parameter matrix K, distortion coefficient D, and extrinsic parameters of each image are globally optimized with the goal of minimizing the circle center reprojection error.

[0013] Furthermore, ellipse matching feature point positioning is performed on each left and right image respectively, and the world coordinate system coordinate is set to plane Z=0 to construct a 3D-2D coordinate correspondence. Taking the left camera coordinate system as the reference, the right camera extrinsic parameter [R|T] and the two camera intrinsic parameter matrices K1 and K2, and the distortion coefficients D1 and D2 are set as joint optimization variables. The Levenberg-Marquardt method is used to minimize the reprojection error of all circle centers. After iterative convergence, the final binocular extrinsic parameter matrix R and translation vector T are output to complete the calibration.

[0014] The present invention also provides a calibration method for binocular camera synchronous automatic calibration, based on the binocular camera synchronous automatic calibration system as described above, the calibration method specifically includes: S1. Fix the binocular camera on the camera bracket, move the calibration plate back and forth to observe the image quality of the calibration plate, and make sure that the calibration plate does not exceed the camera field of view of the binocular camera. At the same time, adjust the depth of field of the binocular camera lens to ensure that the image is always clear within the target range. Then reset the cross slide assembly that drives the calibration plate to prepare for image acquisition. S2. During one rotation cycle, the control and data processing unit controls the outer ring flange to rotate at a set speed, while the binocular camera performs real-time detection of the calibration plate. When the binocular camera detects the calibration plate, the control and data processing unit controls the binocular camera to capture an image of the calibration plate. S3, the control and data processing unit reads the preset area threshold T s Deviation threshold O from coordinate point s , record the calibration plate area S0 and center coordinate point O0 in the current calibration plate image, and control the cross slide assembly to drive the calibration plate to move back and forth, and calculate the calibration plate area S of the subsequent calibration plate image n and the center point coordinates O n , the calibration plate area S of the calibration plate image will be n and the center point coordinates O n Compared with the calibration plate area S in the previous image n-1 and the center point coordinates O n-1 Compare them, if the absolute value deviation between the two is greater than the preset area threshold T s Deviation threshold O from coordinate point s , then the current calibration plate image is collected and put into the image dataset, otherwise the current calibration plate image is not collected; S4. After obtaining the area S0 and center coordinate point O0 of the calibration plate in the first calibration plate image, the control and data processing unit starts to control the cross slide assembly to drive the calibration plate to move back and forth, which is used to simulate the process of collecting calibration images under different distances from the camera module to the calibration plate; when the calibration plate image is no longer detected, the cross slide assembly is controlled to stop driving the calibration plate to move until the binocular camera detects the calibration plate again; S5. After completing the acquisition of the binocular image set, the control and data processing unit uses a binocular calibration algorithm for calibration.

[0015] Furthermore, the S5 specifically includes: S501, calibration plate key point detection: If the calibration plate is a checkerboard calibration plate, Harris corner detection or Shi-Tomasi algorithm is used to detect potential corners in the calibration plate image, valid areas are extracted through morphological operations and contour analysis, and valid corners are screened using quadrilateral fitting and geometric constraints; If the calibration plate is a circular center calibration plate, the bright / dark circular spots in the LoG or DoG detection image are used as the blob extraction results, and then the neighborhood of each blob is binarized and contour extracted. The ellipse is fitted using the least squares method to calculate the center coordinates and radius. Valid circular points are screened based on geometric symmetry and grayscale distribution, and the extracted center is optimized at the sub-pixel level using methods such as the grayscale centroid method. S502. Calculate internal reference in binocular calibration: If the calibration plate is a checkerboard calibration plate, call the Zhang Zhengyou calibration algorithm to perform corner detection → homography matrix estimation → closed-form solution of the extrinsic parameter matrix → reprojection error minimization on each image, thereby outputting the extrinsic parameter matrix K and distortion coefficient D of the left and right cameras; If the calibration plate is a circular center calibration plate, perform ellipse detection → eccentricity compensation → circle center positioning on the circular center calibration plate image to obtain the image center coordinates; use the correspondence between the coordinates and the world coordinates of the known planar circle array to initially estimate the intrinsic and extrinsic parameters through the homography matrix, and globally optimize the camera intrinsic parameter matrix K, distortion coefficient D, and extrinsic parameters of each image with the goal of minimizing the circle center reprojection error; S503, perform stereo correction and calibrate binocular external parameters: Ellipse matching feature point positioning is performed on each left and right image respectively. The world coordinate system coordinate is set to plane Z=0, and a 3D-2D coordinate correspondence is constructed. Taking the left camera coordinate system as the reference, the right camera extrinsic parameter [R|T] and the two camera intrinsic parameter matrices K1 and K2, and the distortion coefficients D1 and D2 are set as joint optimization variables. The Levenberg-Marquardt method is used to minimize the reprojection error of all circle centers. After iterative convergence, the final binocular extrinsic parameter matrix R and translation vector T are output to complete the calibration. S504. Calculate the calibration accuracy error for evaluation: The reprojection error is calculated based on the calibration results. If the reprojection error is less than 0.4 pixel, the calibration error is considered acceptable and the calibration result of the binocular camera is obtained. Otherwise, the binocular calibration should be performed again.

[0016] The beneficial effects of the present invention are: 1. Fully automate the calibration process: Through automatic shooting control, online image quality assessment, and adaptive screening, the efficiency bottlenecks and human errors caused by manual placement of calibration plates and manual triggering of shooting are completely eliminated, meeting the rhythm requirements of mass production lines.

[0017] 2. Unify and improve calibration accuracy: During the image acquisition stage, lighting, blur, and occlusion are scored in real time, and only high-scoring images are retained for solution, fundamentally suppressing reprojection bias caused by inconsistent image quality.

[0018] 3. Build a closed-loop quality traceability system: Generate a unique calibration parameter file and corresponding image fingerprint for each factory device, support subsequent traceability and rapid online recalibration, and reduce after-sales maintenance costs.

[0019] 4. Compatible with multiple camera models and lenses: The algorithm is universal for multimodal imaging systems such as visible light, infrared, and fisheye imaging systems. Through adaptive distortion models and baseline compensation mechanisms, it ensures consistent accuracy across different batches of hardware. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of the structure of the system for synchronous automatic calibration of binocular cameras in the present invention.

[0021] Figure 2 Schematic diagram of the process of the automatic calibration method of the binocular camera in the present invention.

[0022] Figure 3 Schematic diagram of the flow of the binocular camera automatic control algorithm in the present invention.

[0023] In the accompanying drawings, there are a camera bracket 1, an inner ring flange 2, an outer ring flange 3, a cross slide assembly 4, a calibration plate bracket 5, and a control and data processing unit 6.

[0024] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0025] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0026] Binocular calibration originates from the theory of epipolar constraints in stereo vision geometry. Its core task is to solve the extrinsic parameter matrix (rotation R and translation T) between the two cameras, thereby establishing a mapping relationship between 3D spatial points and the coordinates of the left and right images. The traditional process follows a two-step approach: monocular calibration first, then binocular calibration. Zhang's calibration method is first used to estimate the intrinsic parameters of the left and right cameras (focal length, principal point, and distortion coefficients). Extrinsic parameters are then calculated using feature points with the same name. A nonlinear optimization is performed using the reprojection error as the objective function to further reduce the calibration error.

[0027] Existing binocular camera calibration usually uses a handheld or static bracket method, placing a flat calibration plate in different positions and manually collecting images. Therefore, it is difficult to cover the full field of view of the binocular camera and different depths and inclination angles at one time in a limited space. This easily leads to poor repeatability of the calibration plate's position, resulting in the calibration accuracy being greatly affected by human factors. In addition, the acquisition process is time-consuming and labor-intensive, making it difficult to achieve online calibration on mass production lines.

[0028] The present invention provides a compact, highly automated binocular calibration device capable of automatically capturing multiple sets of images in different poses within a single rotation cycle, enabling the calibration of multiple binocular cameras. This significantly improves calibration accuracy and efficiency, while ensuring that the calibration parameters of binocular cameras produced on a single production line are substantially similar. The present invention is used to accurately obtain the relative pose parameters between cameras and is further applied in scenarios such as three-dimensional reconstruction, robot navigation, autonomous driving, or industrial inspection. Specifically, the calibration method is applicable to systems combining visible light cameras, infrared cameras, fisheye cameras, or multispectral cameras.

[0029] like Figure 1 As shown, the present invention provides a system for synchronous automatic calibration of a binocular camera, comprising: base; An inner ring flange 2, fixed on the base, having a circular shape and used to fix M (M ≥ 2) calibration plates, wherein the calibration plates are checkerboard calibration plates or center-shaped calibration plates; An outer ring flange 3 is annular in shape and is disposed on the base. The inner ring of the outer ring flange 3 surrounds the outer circumference of the inner ring flange 2 and rotates around the outer circumference of the inner ring flange 2. The outer ring flange 3 is arranged along the radial direction of the inner ring and drives N groups (N ≥ 2) of binocular cameras to rotate around the calibration plate. A cross slide assembly 4 is provided on the inner ring flange 2, on which the calibration plate is fixed, and is used to drive the calibration plate to change its position in the front-back and left-right directions to adjust the optical axis distance between the binocular camera and the calibration plate; A camera bracket 1 is fixed on the outer ring flange 3 and is used to place and fix the binocular camera; The control and data processing unit 6 is fixed inside the base and connected to the outer ring flange 3, the cross slide assembly 4, and the binocular camera. It is used to control the rotation of the outer ring flange 3, the movement of the calibration plate, and the acquisition of images by the binocular camera, that is, the synchronous control of the stepper motor control and the image acquisition; then, the internal and external parameters of the binocular camera are calculated based on the acquired image set to realize the automatic calibration of the binocular camera.

[0030] In one embodiment, the calibration plate is connected to the cross-slide assembly 4 via a calibration plate bracket 5. Multiple cross-slide assemblies 4 are provided on the inner ring flange 2, each cross-slide assembly 4 corresponding to a calibration plate bracket 5 and a calibration plate. The height of the calibration plate bracket 5 is adjusted so that the heights of the multiple calibration plate brackets 5 are different, thereby making the calibration plates unequal in height and having a certain height difference between them. The height difference between the multiple calibration plate brackets 5 and the calibration plates is Δh, and Δh satisfies Δh=tanθ*D, where D is the closest distance from the calibration plate to the optical center of the binocular camera, θ is the vertical field of view angle of the binocular camera, and the pitch angle can be adjusted independently to simulate different field of view inclination angles.

[0031] In one embodiment, the process of controlling the binocular camera to collect images in the control and data processing unit 6 is as follows: after the binocular camera detects the calibration plate, the binocular camera is controlled to synchronously collect the calibration plate image and record the current calibration plate area S n With the center point coordinate O n At the same time, the cross slide assembly 4 starts to move backward slowly and uniformly. When the new calibration plate area S is detected n+1 With S n The difference is greater than the corresponding area threshold T s , or the new center point coordinates O n+1 With O n The difference is greater than the corresponding coordinate point deviation threshold O s When the calibration plate changes its spatial position, the image of the current calibration plate is acquired; otherwise, the image acquisition signal is not triggered.

[0032] After the binocular camera calibration image acquisition starts, the outer ring flange 3 is controlled to rotate at a constant speed within one rotation cycle, and the binocular camera performs real-time detection of the calibration plate; when the outer ring flange 3 completes one rotation cycle, the binocular camera is controlled to stop real-time detection and acquisition of images.

[0033] That is, the process of controlling the binocular camera to collect images in the control and data processing unit is as follows: after the system equipment is started, within one rotation cycle, the outer ring flange is controlled to rotate at a constant speed, and at the same time, the binocular camera performs real-time detection of the calibration plate; when the binocular camera detects the calibration plate, the binocular camera is controlled to synchronously collect the image of the calibration plate and record the current calibration plate area S n With the center point coordinate O n At the same time, the cross slide assembly starts to move backward slowly and uniformly. When the new calibration plate area S is detected n+1 With S n The difference is greater than the corresponding area threshold T s , or the new center point coordinates O n+1 With O n The difference is greater than the corresponding coordinate point deviation threshold O s When the calibration plate changes its spatial position, the current calibration plate image is collected, otherwise the image collection signal is not triggered; when the outer ring flange completes a rotation cycle, the binocular camera is controlled to stop real-time detection and collection of images.

[0034] In one embodiment, the detection algorithms used by the control and data processing unit include but are not limited to Harris corner detection or Shi-Tomasi algorithm and related optimization algorithms, LoG or DoG detection and related circle center fitting algorithms used for circle center detection. The process of performing dual-target calibration in the control and data processing unit 6 is as follows: when the calibration plate is a checkerboard calibration plate, Harris corner detection or Shi-Tomasi algorithm is used to detect potential corners in the image, effective areas are extracted through morphological operations and contour analysis, and effective corners are screened using quadrilateral fitting and geometric constraints; when calculating intrinsic parameters in dual-target calibration, Zhang Zhengyou calibration algorithm is called to perform corner detection → homography matrix estimation → closed-form solution of internal and external parameter matrices → reprojection error minimization on each image, thereby outputting the internal and external parameter matrices K and distortion coefficients D of the left and right cameras.

[0035] When the calibration plate is a circular-centered calibration plate, the bright / dark circular spots in the LoG or DoG detection image are used as the blob extraction result. The neighborhood of each blob is then binarized and contour extracted. Ellipses are fitted using the least squares method to calculate the center coordinates and radius. Valid circular points are selected based on geometric symmetry and grayscale distribution, and the extracted center is optimized at the sub-pixel level using the grayscale centroid method. When calculating intrinsic parameters in dual-target calibration, the image of the circular-centered calibration plate undergoes ellipse detection, eccentricity compensation, and center positioning to obtain the image center coordinates. Using this correspondence with the world coordinates of a known planar circular array, the intrinsic and extrinsic parameters are initially estimated using a homography matrix. The camera intrinsic parameter matrix K, distortion coefficient D, and extrinsic parameters for each image are then globally optimized with the goal of minimizing the center reprojection error.

[0036] Ellipse matching feature point positioning is performed on each left and right image respectively. The world coordinate system coordinate is set to plane Z=0, and a 3D-2D coordinate correspondence is constructed. Taking the left camera coordinate system as the reference, the right camera extrinsic parameter [R|T] and the two camera intrinsic parameter matrices K1 and K2, and the distortion coefficients D1 and D2 are set as joint optimization variables. The Levenberg-Marquardt method is used to minimize the reprojection error of all circle centers. After iterative convergence, the final binocular extrinsic parameter matrix R and translation vector T are output to complete the calibration.

[0037] like Figure 2 、 Figure 3 As shown, the present invention also provides a calibration method for binocular camera synchronous automatic calibration, based on the binocular camera synchronous automatic calibration system as described above, the calibration method specifically includes: S1. Equipment debugging and initialization First, fix the binocular camera on the camera bracket 1, move the calibration plate back and forth to observe the image quality of the calibration plate, and make sure that the calibration plate does not exceed the camera field of view of the binocular camera. At the same time, adjust the depth of field of the binocular camera lens to ensure that the image is always clear within the target range. Then reset the cross slide assembly 4 that drives the calibration plate to prepare to start collecting multiple sets of binocular images.

[0038] S2, the device controls the outer ring to rotate After binocular calibration image acquisition begins, the control and data processing unit 6 controls the outer ring flange 3 to rotate at a set constant speed within one rotation cycle, while the binocular camera performs real-time detection of the calibration plate. Once the binocular camera detects the calibration plate, the control and data processing unit 6 controls the binocular camera to capture an image of the calibration plate. Once the outer ring completes one rotation cycle, the binocular camera stops real-time detection and image acquisition, and the device modules return to their original positions.

[0039] S3. Calibration plate image detection and acquisition The control and data processing unit 6 reads the preset area threshold T s Deviation threshold O from coordinate point s , record the calibration plate area S0 and center coordinate point O0 in the current calibration plate image, and at the same time control the cross slide assembly 4 to drive the calibration plate to start moving back and forth, and calculate the calibration plate area S of the subsequent calibration plate image n and the center point coordinates O n , the calibration plate area S of the calibration plate image will be n and the center point coordinates O n Compared with the calibration plate area S in the previous image n-1 and the center point coordinates O n-1 Compare them, if the absolute value deviation between the two is greater than the preset area threshold T s Deviation threshold O from coordinate points , it is considered that the calibration plate has changed its spatial posture, and this calibration plate image should be collected as the calibration image dataset, otherwise the current calibration plate image will not be collected.

[0040] S4. Motion control of calibration board After obtaining the calibration plate area S0 and center coordinate point O0 in the first calibration plate image, the control and data processing unit 6 starts to control the cross-slide assembly 4 to drive the calibration plate to move back and forth, which is used to simulate the process of collecting calibration images under different distances from the camera module to the calibration plate; when the calibration plate image cannot be detected, the cross-slide assembly 4 is controlled to stop driving the calibration plate to move until the binocular camera re-detects the calibration plate.

[0041] S5. After the outer ring rotates one circle, the outer ring stops rotating, completing the binocular image set acquisition. At the same time, the calibration plate bracket 5 and the cross slide assembly 4 are reset. The control and data processing unit 6 uses the binocular positioning algorithm to perform binocular positioning. The generated binocular positioning results are burned into the binocular camera module. The binocular camera module can be removed and used. The binocular positioning algorithm specifically includes: S501, calibration plate key point detection: Detect key points in the calibration plate. If the calibration plate is a checkerboard calibration plate, use Harris corner detection or Shi-Tomasi algorithm to detect potential corners in the calibration plate image. Extract valid areas through morphological operations (such as dilation and erosion) and contour analysis. Use quadrilateral fitting and geometric constraints (such as corner collinearity and spacing rules) to screen valid corners.

[0042] If the calibration target is a circular-centered target, the LoG (Laplacian of Gaussian) or DoG (Difference of Gaussians) method is used to detect bright and dark circular spots in the image as blob extraction results. Each blob's neighborhood is then binarized and contour extracted. An ellipse is fitted using the least squares method to calculate the center coordinates and radius. Valid dots are screened based on geometric symmetry (such as dot spacing and row and column alignment) and grayscale distribution (the contrast between the dot center and the background). The extracted center is then optimized at the subpixel level using methods such as the grayscale centroid method.

[0043] S502. Calculate internal reference in binocular calibration: If the calibration plate is a checkerboard calibration plate, call Zhang Zhengyou calibration algorithm to perform corner detection → homography matrix estimation → closed-form solution of extrinsic parameter matrix → reprojection error minimization for each image, thereby outputting the extrinsic parameter matrix K and distortion coefficient D of the left and right cameras.

[0044] If the calibration plate is a circular center calibration plate, perform ellipse detection → eccentricity compensation → circle center positioning on the circular center calibration plate image to obtain the image center coordinates; use its correspondence with the world coordinates of the known planar circle array to initially estimate the intrinsic parameters / extrinsic parameters through the homography matrix, and globally optimize the camera intrinsic parameter matrix K, distortion coefficient D and extrinsic parameters of each image with the goal of minimizing the circle center reprojection error.

[0045] S503, perform stereo correction and calibrate binocular external parameters: Ellipse matching feature point positioning is performed on each left and right image respectively. The world coordinate system coordinate is set to plane Z=0, and a 3D-2D coordinate correspondence is constructed. Taking the left camera coordinate system as the reference, the right camera extrinsic parameter [R|T] and the two camera intrinsic parameter matrices K1 and K2, and the distortion coefficients D1 and D2 are set as joint optimization variables. The Levenberg-Marquardt method is used to minimize the reprojection error of all circle centers. After iterative convergence, the final binocular extrinsic parameter matrix R and translation vector T are output to complete the calibration.

[0046] S504. Calculate the calibration accuracy error for evaluation: The reprojection error is calculated based on the calibration results. If the reprojection error is less than 0.4 pixel, the calibration error is considered acceptable and the calibration result of the binocular camera is obtained. Otherwise, the binocular calibration should be performed again.

[0047] In the present invention, the control and data processing unit 6 outputs a synchronous trigger signal, causing each group of binocular cameras to capture images at the same time. At the same time, the outer ring of the concentric flange is driven to rotate about its central axis, and the calibration plate located on the inner ring is driven to produce a two-dimensional translation along the cross guide rail, so that the calibration plate forms a multi-pose, full-coverage motion trajectory within the common field of view of each binocular camera. In each pose of the trajectory, each group of binocular cameras synchronously captures an image of the calibration plate to obtain a multi-binocular, multi-view calibration image dataset. Finally, center detection, decentering compensation, and joint optimization are performed on the dataset to calculate the intrinsic parameters, distortion coefficients, and extrinsic parameter matrices of all binocular cameras at one time.

[0048] This invention directly addresses the four major pain points of traditional dual-target calibration: "manual oscillation, uneven frame quality, beat bottlenecks, and consistency loss." It establishes a "zero-manual, zero-defective, zero-replacement" calibration system. With adaptive visual quality control, millisecond-level closed-loop solution, and multi-model co-linear adaptation, it achieves automated, standardized, and scalable output of 3D visual benchmarks in mass production scenarios. Its purpose is to: 1. Fully automate the calibration process: Through automatic shooting control, online image quality assessment, and adaptive screening, the efficiency bottlenecks and human errors caused by manual placement of calibration plates and manual triggering of shooting are completely eliminated, meeting the rhythm requirements of mass production lines.

[0049] 2. Unify and improve calibration accuracy: During the image acquisition stage, lighting, blur, and occlusion are scored in real time, and only high-scoring images are retained for solution, fundamentally suppressing reprojection bias caused by inconsistent image quality.

[0050] 3. Build a closed-loop quality traceability system: Generate a unique calibration parameter file and corresponding image fingerprint for each factory device, support subsequent traceability and rapid online recalibration, and reduce after-sales maintenance costs.

[0051] 4. Compatible with multiple camera models and lenses: The algorithm is universal for multimodal imaging systems such as visible light, infrared, and fisheye imaging systems. Through adaptive distortion models and baseline compensation mechanisms, it ensures consistent accuracy across different batches of hardware.

[0052] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0053] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A system for synchronous automatic calibration of binocular cameras, characterized in that: include: base; an inner ring flange, the inner ring flange being fixed on the base and being used to fix M calibration plates; An outer ring flange, which is arranged on the base and rotates around the inner ring flange, and is used to arrange and drive N groups of binocular cameras to rotate around the calibration plate; A cross slide assembly, the cross slide assembly being arranged on the inner ring flange, the calibration plate being fixed thereon, and being used to drive the calibration plate to change its position in the front-back and left-right directions so as to adjust the optical axis distance between the binocular camera and the calibration plate; A camera bracket, fixed on the outer ring flange, for placing and fixing the binocular camera; A control and data processing unit is fixed inside the base and connected to the outer ring flange, the cross slide assembly, and the binocular camera. The control and data processing unit is used to control the rotation of the outer ring flange, the movement of the calibration plate, and the binocular camera to capture images, and calculate the internal and external parameters of the binocular camera based on the collected image set to achieve automatic calibration of the binocular camera.

2. The system for synchronous automatic calibration of binocular cameras according to claim 1, characterized in that: The inner ring flange is circular in shape, the outer ring flange is annular in shape, and the inner ring of the outer ring flange surrounds the outer circumference of the inner ring flange; The calibration plate is a checkerboard calibration plate or a center-shaped calibration plate, and the number of the calibration plates and the number of the binocular cameras are both greater than or equal to 2.

3. The system for synchronous automatic calibration of binocular cameras according to claim 2, characterized in that: The calibration plate and the cross slide assembly are connected through a calibration plate bracket. A plurality of cross slide assemblies are provided on the inner ring flange. Each cross slide assembly corresponds to a calibration plate bracket and a calibration plate. The heights of the plurality of calibration plate brackets are different, and the height difference of the plurality of calibration plate brackets is Δh, and Δh satisfies Δh=tanθ*D, where D is the closest distance from the calibration plate to the optical center of the binocular camera, θ is the vertical field of view angle of the binocular camera, and the pitch angle can be adjusted independently to simulate different field of view inclination angles.

4. The system for synchronous automatic calibration of binocular cameras according to claim 2, characterized in that: The process of controlling the binocular camera to collect images in the control and data processing unit is as follows: after the system equipment is started, within a rotation cycle, the outer ring flange is controlled to rotate at a constant speed, and at the same time, the binocular camera performs real-time detection of the calibration plate; when the binocular camera detects the calibration plate, the binocular camera is controlled to synchronously collect the calibration plate image and record the current calibration plate area S n With the center point coordinate O n At the same time, the cross slide assembly starts to move backward slowly and uniformly. When the new calibration plate area S is detected n+1 With S n The difference is greater than the corresponding area threshold T s , or the new center point coordinates O n+1 With O n The difference is greater than the corresponding coordinate point deviation threshold O s When the calibration plate changes its spatial position, the current calibration plate image is collected, otherwise the image collection signal is not triggered; when the outer ring flange completes a rotation cycle, the binocular camera is controlled to stop real-time detection and collection of images.

5. The system for synchronous automatic calibration of binocular cameras according to claim 4, characterized in that: When the binocular camera to be calibrated detects the calibration plate, the feature points of the calibration plate are identified through corner detection and related optimization methods, and the internal and external parameters related to the binocular camera are solved through the feature points.

6. A calibration method for synchronous automatic calibration of a binocular camera, characterized in that: A system for synchronous automatic calibration of binocular cameras according to any one of claims 1 to 5, wherein the calibration method specifically comprises: S1. Fix the binocular camera on the camera bracket, move the calibration plate back and forth to observe the image quality of the calibration plate, and make sure that the calibration plate does not exceed the camera field of view of the binocular camera. At the same time, adjust the depth of field of the binocular camera lens to ensure that the image is always clear within the target range. Then reset the cross slide assembly that drives the calibration plate to prepare for image acquisition. S2. During one rotation cycle, the control and data processing unit controls the outer ring flange to rotate at a set speed, while the binocular camera performs real-time detection of the calibration plate. When the binocular camera detects the calibration plate, the control and data processing unit controls the binocular camera to capture an image of the calibration plate. S3, the control and data processing unit reads the preset area threshold T s Deviation threshold O from coordinate point s , record the calibration plate area S0 and center coordinate point O0 in the current calibration plate image, and control the cross slide assembly to drive the calibration plate to move back and forth, and calculate the calibration plate area S of the subsequent calibration plate image n and the center point coordinates O n , the calibration plate area S of the calibration plate image will be n and the center point coordinates O n Compared with the calibration plate area S in the previous image n-1 and the center point coordinates O n-1 Compare them, if the absolute value deviation between the two is greater than the preset area threshold T s Deviation threshold O from coordinate point s , then the current calibration plate image is collected and put into the image dataset, otherwise the current calibration plate image is not collected; S4. After obtaining the area S0 and center coordinate point O0 of the calibration plate in the first calibration plate image, the control and data processing unit starts to control the cross slide assembly to drive the calibration plate to move back and forth, which is used to simulate the process of collecting calibration images under different distances from the camera module to the calibration plate; when the calibration plate image is no longer detected, the cross slide assembly is controlled to stop driving the calibration plate to move until the binocular camera detects the calibration plate again; S5. After completing the acquisition of the binocular image set, the control and data processing unit uses a binocular calibration algorithm for calibration.

7. The calibration method for synchronous automatic calibration of a binocular camera according to claim 6, characterized in that: The S5 specifically includes: S501, calibration plate key point detection: If the calibration plate is a checkerboard calibration plate, Harris corner detection or Shi-Tomasi algorithm is used to detect potential corners in the calibration plate image, valid areas are extracted through morphological operations and contour analysis, and valid corners are screened using quadrilateral fitting and geometric constraints; If the calibration plate is a circular center calibration plate, the bright / dark circular spots in the LoG or DoG detection image are used as the blob extraction results, and then the neighborhood of each blob is binarized and contour extracted. The ellipse is fitted using the least squares method to calculate the center coordinates and radius. Valid points are screened based on geometric symmetry and grayscale distribution, and the extracted center is optimized at the sub-pixel level using the grayscale centroid method. S502. Calculate internal reference in binocular calibration: If the calibration plate is a checkerboard calibration plate, call the Zhang Zhengyou calibration algorithm to perform corner detection → homography matrix estimation → closed-form solution of the extrinsic parameter matrix → reprojection error minimization on each image, thereby outputting the extrinsic parameter matrix K and distortion coefficient D of the left and right cameras; If the calibration plate is a circular center calibration plate, perform ellipse detection → eccentricity compensation → circle center positioning on the circular center calibration plate image to obtain the image center coordinates; use the correspondence between the coordinates and the world coordinates of the known planar circle array to initially estimate the intrinsic and extrinsic parameters through the homography matrix, and globally optimize the camera intrinsic parameter matrix K, distortion coefficient D, and extrinsic parameters of each image with the goal of minimizing the circle center reprojection error; S503, perform stereo correction and calibrate binocular external parameters: Ellipse matching feature point positioning is performed on each left and right image respectively. The world coordinate system coordinate is set to plane Z=0, and a 3D-2D coordinate correspondence is constructed. Taking the left camera coordinate system as the reference, the right camera extrinsic parameter [R|T] and the two camera intrinsic parameter matrices K1 and K2, and the distortion coefficients D1 and D2 are set as joint optimization variables. The Levenberg-Marquardt method is used to minimize the reprojection error of all circle centers. After iterative convergence, the final binocular extrinsic parameter matrix R and translation vector T are output to complete the calibration. S504. Calculate the calibration accuracy error for evaluation: The reprojection error is calculated based on the calibration results. If the reprojection error is less than 0.4 pixel, the calibration error is considered acceptable and the calibration result of the binocular camera is obtained. Otherwise, the binocular calibration should be performed again.

Images