Rapid binocular camera calibration method

Abstract

The invention relates to the technical field of machine vision, in particular to a rapid binocular camera calibration method. A rapid binocular camera calibration method comprises the following steps: using a one-dimensional calibration rod as a calibration object, installing a binocular camera on a cradle head with an automatic vertical calibration function, and ensuring that the optical axis of the binocular camera is always parallel to the ground and is vertical to the calibration rod; meanwhile, the calibration rod is also fixed on the holder with an automatic vertical calibration function, so that the calibration rod is ensured to be vertical to the ground; based on an RTK positioning technology, an RTK positioning antenna is installed on a binocular camera and a calibration rod, position information of the binocular camera and the calibration rod is accurately obtained by using an RTK positioning system, and internal and external parameters of the binocular camera are determined by combining calibration rod images shot at multiple angles, so that rapid calibration of the binocular camera is realized. The method is easy and convenient to operate and high in calibration speed, forging parts do not need to be moved in the calibration process, and the method is particularly suitable for application requirements in the complex industrial environment.

Description

Technical Field

[0001] The present invention relates to the technical field of machine vision, and specifically relates to a fast binocular camera calibration method to further improve the camera calibration efficiency. Background Art

[0002] In the dimensional inspection of forgings, accurate measurement is of utmost importance. Camera calibration, as the core technology of optical measurement systems, is a crucial link in improving measurement accuracy and reliability. Through efficient and non-contact optical measurement techniques, the calibrated camera can quickly and accurately obtain the dimensions of forgings, especially suitable for large forgings with complex shapes, providing reliable data support for quality control and subsequent processing. Current camera calibration techniques can be roughly divided into three categories: camera self-calibration methods, active vision calibration methods, and reference-based camera calibration methods. Chinese Patent 202311786941.8 calibrates a binocular camera by collecting multiple image frames of a calibration board, calculating the pixel displacement distance between adjacent frames, screening out target frames with pixel displacement not less than a preset threshold, then extracting sub-pixel-level corner coordinates in the target frames, and finally calibrating the binocular camera based on these corner coordinates. This method requires obtaining multiple image frames of the calibration board, performing pixel displacement calculations and corner extraction on multiple frames of images, increasing the computational complexity of the calibration process. In addition, the pixel displacement is sensitive to the preset threshold. Too high a threshold may result in the omission of valid frames, while too low a threshold may introduce noise interference. Chinese Patent 202311832150.4 proposes a multi-view camera calibration method based on deep learning. By generating a three-dimensional point data set to calculate the internal and external parameter matrices, and using reprojection error and mean square error for model training, the internal parameter information is output. This method requires a large amount of training, and the effect of the model highly depends on the design of the loss function and the adjustment of hyperparameters during the training process. The training process is complex and usually requires a large amount of computing resources and time. In addition, in practical applications, complex environments such as light changes, noise interference, and occlusion may cause unstable model predictions, affecting the calibration accuracy. Chinese Patent 202411021475.9 uses checkerboard image data at different angles for sampling point extraction, parameter solution, distortion parameter estimation, and inverse distortion iterative optimization, and outputs the camera internal parameters and distortion parameters to improve the robustness of calibration. This method requires the camera to collect multiple images, perform feature vector calculations in multiple poses, and repeatedly perform parameter estimation and distortion correction through multiple rounds of iterative optimization. Under the conditions of high-resolution images and large-scale sampling points, the computational complexity increases significantly, resulting in a cumbersome and time-consuming calibration process and low calibration efficiency. Chinese Patent 202410996333.8 obtains images of the calibration board at different object distances, performs elliptical fitting on concentric circle markings, calculates the center of the circle and averages it to obtain the final coordinates of the marked points. Matching these coordinates with the reference coordinate system of the calibration board to obtain the calibration result of the camera. However, when the tilt angle of the calibration board is large, the elliptical fitting will become inaccurate. If the camera shooting angle is large or the camera is close to the calibration board, it will also cause the fitted elliptical contour and the center of the circle to be inaccurate, thus affecting the final calibration result. Based on the above technical status and limitations, the present invention proposes a fast camera calibration method, aiming to improve the existing technology and provide a calibration method with convenient operation and efficient calibration process. Summary of the invention

[0003] In view of the problem that the existing camera calibration method is cumbersome and time-consuming, resulting in low calibration efficiency, the present invention proposes a fast binocular camera calibration method. The method is based on RTK (Real-time kinematic) positioning technology. By installing RTK positioning antennas on the binocular camera and the calibration rod, the binocular camera and the calibration rod are fixed on a gimbal with automatic vertical calibration, and the RTK positioning system is used to accurately obtain the three-dimensional coordinates of the two. Combined with the calibration rod images taken from multiple angles, the internal and external parameters of the binocular camera are determined, thereby realizing the fast calibration of the binocular camera, improving the efficiency of the calibration process, simplifying the operation steps, and reducing the influence of human factors.

[0004] The present invention is implemented by the following technical scheme: a fast binocular camera calibration method, using a one-dimensional calibration rod as a calibration object, installing the binocular camera on a pan-tilt with automatic vertical calibration, ensuring that the binocular camera optical axis is always parallel to the ground and perpendicular to the calibration rod no matter at what angle the binocular camera is placed; at the same time, the calibration rod is also fixed on the pan-tilt with automatic vertical calibration to ensure that it is perpendicular to the ground; the calibration rod and the binocular camera are equipped with an RTK positioning system for accurately obtaining the three-dimensional coordinates of the two, by moving the binocular camera to different positions, capturing the calibration rod image from multiple angles, and combining the position information of the binocular camera and the calibration rod to obtain the internal and external parameters of the binocular camera, thereby achieving fast calibration.

[0005] The above-mentioned fast binocular camera calibration method selects three calibration rods and places them evenly around the forging. The calibration rods are red and white calibration rods with diffuse reflection and no reflection. The calibration rods are evenly divided into fixed lengths, and each section is equal in length. The red and white junction of the calibration rod is set as the mark point. During calibration, the optical axis of the binocular camera is aligned with the mark point of the calibration rod. By setting the resolution, it is ensured that all the mark points of the three calibration rods can be simultaneously covered and clearly captured within the field of view of the binocular camera.

[0006] The above-mentioned fast binocular camera calibration method, the RTK positioning system includes an RTK base station and an antenna, the antenna is installed on the binocular camera and the calibration pole, the antennas installed on the binocular camera and the calibration pole are based on the same RTK base station, which ensures that both can be positioned in the same coordinate system; through this configuration, no matter how the position of the binocular camera or the calibration pole changes, the RTK positioning system can update the positional relationship between the binocular camera and the calibration pole in real time, thereby ensuring the accuracy of the calibration process.

[0007] The above-mentioned fast binocular camera calibration method. There is radial distortion in the images captured by the binocular camera. In this case, first, feature points in the images are extracted and matched with the physical coordinates of the landmark points on the calibration rod. During the matching process, the radial distortion coefficients are calculated using the radial distortion model. After obtaining the radial distortion coefficients, the images are corrected for distortion. The coordinates of each pixel in the image are adjusted through the correction formula to eliminate the radial distortion.

[0008] The above-mentioned fast binocular camera calibration method. The images after radial distortion correction are filtered and denoised, and feature points in the images are extracted.

[0009] After extracting the feature points, the internal and external parameters of the binocular camera are calculated using these feature points. The external parameters refer to the position and orientation of the binocular camera in space, including the translation vector and the rotation matrix. Since the orientation of the binocular camera remains unchanged throughout the shooting process, the rotation matrix is the identity matrix. The three-dimensional coordinates of the binocular camera and the calibration rod are obtained through the RTK positioning system, and the translation vector of the binocular camera is calculated by computing the coordinate difference between the two. The internal parameters of the binocular camera include the focal length, the principal point coordinates, and the internal parameter matrix. The focal length and the principal point coordinates of the binocular camera are deduced from the physical distance between the landmark points on the calibration rod and the pixel distance between the feature points in the image. Based on the focal length and the principal point coordinates, the internal parameter matrix of the binocular camera is further determined.

[0010] The above-mentioned fast binocular camera calibration method. After the image is corrected for radial distortion, the image coordinates are re-projected into the physical coordinate system, and the reprojection error between the corrected image coordinates and the physical coordinates is calculated. And the least squares method is used to iteratively optimize the radial distortion coefficients to minimize the reprojection error in all images.

[0011] The above-mentioned fast binocular camera calibration method. In the case of partial occlusion of the image or changes in the lighting conditions, since each segment of the calibration rod has the same length and the boundary between the red and white sections is regular, the positions of the feature points are deduced based on the length ratio relationship between the red and white sections.

[0012] A fast binocular camera calibration method proposed by the present invention introduces a pan-tilt head with automatic vertical calibration and an RTK positioning system into the binocular camera calibration process, significantly improving the calibration efficiency. Compared with the existing calibration methods that rely on calculating feature vectors with multiple postures, this solution has the advantages of simple operation and fast calibration speed, and there is no need to move the forging during the calibration process, which is particularly suitable for the application requirements in complex industrial environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flowchart of the binocular camera calibration of the present invention.

[0014] Figure 2This is a schematic diagram of binocular camera calibration for the present invention.

[0015] In the figure: 1 - antenna, 2 - binocular camera, 3 - pan-tilt head, 4 - forging, 5 - calibration rod, 6 - RTK base station. Specific implementation manner

[0016] A fast binocular camera calibration method is to place the binocular camera 2 and the calibration rod 5 on the pan-tilt head 3 with automatic vertical calibration function to ensure that the calibration rod 5 is perpendicular to the ground, and the optical axis of the binocular camera 2 is always parallel to the ground and perpendicular to the calibration rod 5. The calibration rod 5 and the binocular camera 2 are equipped with an RTK positioning system for accurately calculating the three-dimensional coordinates of the two. The RTK positioning system includes an RTK base station 6 and an antenna 1. The antenna 1 is installed on the binocular camera 2 and the calibration rod 5, and the antennas 1 installed on the binocular camera 2 and the calibration rod 5 are based on the same RTK base station 6. Place three calibration rods 5 evenly around the forging 4. The calibration rod 5 should be placed in a flat and evenly lit environment to avoid shadows and direct strong light to ensure the quality of the captured images. To improve the calibration accuracy, select a red and white calibrated rod 5 with diffuse reflection and non-reflection, and evenly segment the calibration rod 5 at a fixed length, with each segment having the same length; set the red and white boundary of the calibration rod 5 as the landmark point, and align the optical axis of the binocular camera 2 with the landmark point of the calibration rod 5. By setting an appropriate resolution, ensure that all landmark points of the three calibration rods 5 can be covered and clearly captured within the field of view of the binocular camera 2. After fixing the baseline of the binocular camera 2, take pictures of the calibration rod 5. Subsequently, move the three calibration rods 5 to new positions respectively to ensure that they maintain an appropriate distance from the forging 4 and provide perspective coverage of different angles on the surface of the forging 4. Then, keep the attitude of the binocular camera 2 unchanged, move the position of the binocular camera 2, and take pictures again.

[0017] During the shooting process of the binocular camera 2, due to factors such as the optical characteristics of the lens, there is generally lens distortion, including radial distortion and tangential distortion. Tangential distortion is mainly caused by the non - parallelism between the lens and the image plane. Since the optical axis of the binocular camera 2 is parallel to the ground, the calibration rod 5 is perpendicular to the ground, and the attitude of the binocular camera 2 remains unchanged during the shooting process, its lens optical axis is always perpendicular to the image plane. Therefore, the influence of tangential distortion can be ignored, and only radial distortion needs to be considered. This reduces the computational complexity in the distortion correction process and improves the calibration efficiency. In this case, first, feature points in the image are extracted and matched with the physical coordinates of the fiducial points. During the matching process, the radial distortion coefficient is calculated using the radial distortion model. After obtaining the radial distortion coefficient, the image is corrected for distortion. Specifically, the coordinates of each pixel point in the image are adjusted through the correction formula to eliminate the radial distortion effect. After correction, the image coordinates are re - projected into the physical coordinate system, and the reprojection error between the corrected image coordinates and the physical coordinates is calculated. To further improve the accuracy, optimization algorithms such as the least - squares method are used to iteratively optimize the radial distortion coefficient to minimize the reprojection error in all images. The corrected image can be checked through further experiments and verification to ensure that the geometric shape in the image is restored to the correct state.

[0018] The image after radial distortion correction is filtered and denoised, and feature points in the image are extracted. Even in the case of partial occlusion of the image or changes in lighting conditions, since each section of the calibration rod 5 has the same length and the boundary between the red and white sections has strong regularity, the position of the feature points can be directly deduced based on the length ratio relationship between the red and white sections. This method greatly simplifies the process of feature point extraction.

[0019] After extracting the feature points, the internal and external parameters of the binocular camera 2 are calculated using these feature points. The external parameters refer to the position and attitude of the binocular camera 2 in space, usually including the translation vector and the rotation matrix. Since the attitude of the binocular camera 2 remains unchanged throughout the shooting process, the rotation matrix can be assumed to be the identity matrix. The three - dimensional coordinates of the binocular camera 2 and the calibration rod 5 are obtained through the RTK positioning system, and the translation vector of the binocular camera 2 is calculated by computing the coordinate difference between the two. This method avoids the complex process of shooting with multiple attitude changes, simplifies the calculation steps, and improves the calibration speed of the binocular camera 2.

[0020] The internal parameters of the binocular camera 2 include the focal length, principal point coordinates, internal parameter matrix, etc., which describe the imaging characteristics of the binocular camera 2. Since both the binocular camera 2 and the calibration rod 5 are perpendicular to the ground and the optical axis is always aligned with the fiducial points of the calibration rod 5, combined with the geometric imaging principle, the focal length and principal point coordinates of the binocular camera 2 can be deduced from the physical distance between the fiducial points of the calibration rod and the pixel distance between the feature points in the image. Based on the focal length and principal point coordinates, the internal parameter matrix of the binocular camera 2 is further determined.

[0021] After the binocular camera 2 is calibrated, the image coordinates are converted into actual physical coordinates by using the obtained internal and external parameters, so as to realize the dimensional measurement of the forging 4.

Claims

1. A fast binocular camera calibration method, characterized in that: Use a one-dimensional calibration rod (5) as the calibration object. Install the binocular camera (2) on a pan-tilt head (3) with automatic vertical calibration to ensure that the optical axis of the binocular camera (2) is always parallel to the ground and perpendicular to the calibration rod (5). At the same time, the calibration rod (5) is also fixed on the pan-tilt head (3) with automatic vertical calibration to ensure its perpendicularity to the ground. The calibration rod (5) and the binocular camera (2) are equipped with an RTK positioning system for accurately obtaining the three-dimensional coordinates of both. By moving the binocular camera (2) to different positions, capturing calibration rod images from multiple angles, and combining the position information of the binocular camera (2) and the calibration rod (5), the internal and external parameters of the binocular camera (2) are obtained, thus achieving rapid calibration.

2. A fast binocular camera calibration method according to claim 1, characterized in that: Select three calibration rods (5) and place them evenly around the forging (4). The calibration rod (5) is a red-white alternating calibration rod with diffuse reflection and non-reflective properties, and the calibration rod (5) is evenly segmented according to a fixed length, with each segment having the same length. Set the red-white junction of the calibration rod (5) as the landmark point. During calibration, the optical axis of the binocular camera (2) is aligned with the landmark point of the calibration rod (5). By setting the resolution, ensure that all landmark points of the three calibration rods (5) can be simultaneously covered and clearly captured within the field of view of the binocular camera (2).

3. A fast binocular camera calibration method according to claim 2, characterized in that: The RTK positioning system includes an RTK base station (6) and an antenna (1). The antenna (1) is installed on the binocular camera (2) and the calibration rod (5). The antennas (1) installed on the binocular camera (2) and the calibration rod (5) are based on the same RTK base station (6), which ensures that both can be positioned in the same coordinate system.

4. A fast binocular camera calibration method according to claim 3, characterized in that: There is radial distortion in the images captured by the binocular camera (2). In this case, first extract the feature points in the images and match them with the physical coordinates of the landmark points of the calibration rod (5). During the matching process, use the radial distortion model to calculate the radial distortion coefficient. After obtaining the radial distortion coefficient, perform distortion correction on the images. Adjust the coordinates of each pixel point in the images through the correction formula to eliminate the radial distortion.

5. A rapid binocular camera calibration method according to claim 4, characterized in that: Perform filtering and noise reduction processing on the images after radial distortion correction, and extract the feature points in the images. After extracting the feature points, use these feature points to calculate the internal and external parameters of the binocular camera (2). The external parameters refer to the position and attitude of the binocular camera (2) in space, including the translation vector and the rotation matrix. Since the attitude of the binocular camera (2) remains unchanged throughout the shooting process, the rotation matrix is the identity matrix. Obtain the three-dimensional coordinates of the binocular camera (2) and the calibration rod (5) through the RTK positioning system, and calculate the coordinate difference between the two to obtain the translation vector of the binocular camera (2). The internal parameters of the binocular camera (2) include the focal length, the principal point coordinates, and the internal parameter matrix. By calculating the physical distance between the landmark points of the calibration rod (5) and the pixel distance between the feature points in the image, the focal length and the principal point coordinates of the binocular camera (2) are deduced. Using the obtained focal length and principal point coordinates, determine the internal parameter matrix of the binocular camera (2).

6. A fast binocular camera calibration method according to claim 4 or 5, characterized in that: After the image radial distortion is corrected, the image coordinates are re-projected into the physical coordinate system, the reprojection error between the corrected image coordinates and the physical coordinates is calculated, and the radial distortion coefficients are iteratively optimized using the least squares method to minimize the reprojection error in all images.

7. A fast binocular camera calibration method according to claim 4 or 5, characterized in that: In the case of partial occlusion of the image or changes in lighting conditions, since each section of the calibration rod (5) has the same length and the boundary between the red and white sections is regular, the positions of the feature points are deduced based on the length ratio relationship between the red and white sections.

Images