A small calibration field calibration plate and a point high-precision external parameter calibration method

Abstract

This application belongs to the field of measurement technology, and particularly relates to a high-precision extrinsic parameter calibration method for calibration plates and points in a small calibration field. This invention can be easily and quickly deployed and implemented, and in a frequently changing small calibration field, it can solve for the extrinsic parameters between calibration plates or coded calibration points at low cost, high efficiency, and accuracy. Furthermore, it requires minimal technical expertise from operators, does not require highly specialized and expensive hardware, has a simple configuration, and the entire process is automated with minimal human intervention.

Description

Technical Field

[0001] This application belongs to the field of measurement technology, and in particular relates to a small calibration field calibration plate and a high-precision external parameter calibration method. Background Technology

[0002] Currently, whether in autonomous driving, or in the construction of calibration fields for real-time localization and mapping (SLAM) and photogrammetry, most applications employ specialized surveying equipment with millimeter-level or higher precision (such as 3D laser / vision scanners, surveying robots, and total stations) to precisely measure the 3D coordinates of coded points. After data processing and calculation, the extrinsic parameters of the calibration board and points relative to a specific reference frame within the calibration field are ultimately determined. Then, based on the coded control points on these calibration boards, or individual coded control points, calibration of autonomous vehicle battery electric vehicles (BEVs), the intrinsic and extrinsic parameters of SLAM hardware sensor combinations, and high-precision intrinsic and extrinsic parameter calibration between photogrammetric cameras can be achieved. Finally, based on these intrinsic and extrinsic parameters, various visual intelligence algorithms for spatial perception can be developed.

[0003] However, the above calibration method involves a large amount of measurement work, and the operation of such high-precision measuring equipment is complex and expensive, ultimately resulting in high costs for constructing calibration fields. Furthermore, the calibration plates and points within the calibration field require external parameter verification and recalibration as time progresses and new layout requirements change, leading to extremely high overall costs.

[0004] Therefore, how to ensure the accuracy of the extrinsic parameters of the calibration board (containing coded points) and points (with independent coded values) in the calibration field, while achieving high efficiency, low cost, and convenient and rapid deployment, has become the primary problem to be solved in the research and development of visual algorithm applications. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a high-precision extrinsic parameter calibration method for calibration boards and points in a small calibration field. This invention can be easily and quickly deployed and implemented, and can solve for the extrinsic parameters between calibration boards or coded calibration points in a small calibration field with frequent changes, at low cost, high efficiency, and high accuracy. Moreover, it requires low technical expertise from operators, does not require highly specialized and expensive hardware equipment, has simple configuration, and the entire process is automated with minimal human intervention.

[0006] To achieve its purpose, the present invention employs the following solution: a small calibration field calibration board and a high-precision external parameter calibration method, comprising the following steps: 1) Using a single calibration board, the visual inertial intrinsic and extrinsic parameters of the visual-inertial binocular SLAM hardware system were calibrated; 2) Use truth-based tools to measure the approximate initial values ​​of the extrinsic parameters between the calibration plates, as well as other constraint information, as priors for the optimization algorithm to solve for the extrinsic parameters; 3) Within a local area of ​​the calibration site, use the above-mentioned VIO hardware to execute a visual-inertial SLAM algorithm with prior information constraints; 4) By repeatedly moving the SLAM VIO device, the corner points on the coding calibration board and the corner points on the independent coding points are repeatedly observed. With the specially configured SLAM VIO hardware, the mapping and optimization accuracy of the corner points is ensured. Through continuous optimization and calculation of the SLAM algorithm, the accurate 3D coordinate values ​​of the coding points in the calibration field and the extrinsic parameters of the calibration board and points are finally calculated.

[0007] Technical limitations: The current methods for constructing calibration plates and coding points in calibration fields require extensive manual measurement and calculation, and have limited accuracy (total stations and 3D laser scanners typically have millimeter-level accuracy, and station-changing measurements introduce uncontrollable errors; 3D laser scanners also cannot accurately determine the range or location of calibration plates and points; furthermore, these specialized devices are complex to operate, and high-precision measurement modes require advanced and complex techniques, which is not conducive to the construction of low-density calibration fields or the need for frequent measurements; ordinary measuring rulers, even at short distances of 5m, can have measurement errors at the centimeter level, and cannot measure the angles between calibration plates). These methods cannot take into account the prior conditions of the calibration field for joint adjustment calculations, thus failing to further improve the external parameter measurement accuracy of calibration plates and points. The invention takes into account the prior scale and rotation parameters between the coding calibration boards, and can calculate the three-dimensional coordinates of a large number of coding corner points in the calibration field. For cameras used in binocular or multi-view SLAM, the larger the target surface size of the image sensor and the focal length of the lens, the higher the IMU accuracy. Ultimately, the extrinsic parameter accuracy of the calibration boards and points in the calibration field calculated by the invention is also higher. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the layout of the calibration plate and points in this invention; Figure 2 This refers to the design of independent coding points (including calibration check points) in this invention; Figure 3 This is the style of the calibration plate (containing coded calibration points) in this invention; Figure 4 This is a flowchart illustrating the main operation of the present invention. Detailed Implementation

[0009] Example 1: A method for high-precision extrinsic parameter calibration of a small calibration field calibration board and point, comprising the following steps: 1) Using a single calibration board, the visual inertial intrinsic and extrinsic parameters of the visual-inertial binocular SLAM hardware system were calibrated; 2) Use truth-based tools to measure the approximate initial values ​​of the extrinsic parameters between the calibration plates, as well as other constraint information (such as the true values ​​of GPS with cm-level accuracy outdoors and the true values ​​of motion capture with sub-millimeter accuracy indoors), as priors for the optimization algorithm to solve for the extrinsic parameters; 3) Within a local area of ​​the calibration site, use the above-mentioned VIO hardware to execute a visual-inertial SLAM algorithm with prior information constraints; 4) By repeatedly moving the SLAM VIO device, corner points on the coding calibration board and independent coding points are repeatedly observed. With specially configured SLAM VIO hardware, the accuracy of corner point mapping optimization is ensured. Through continuous optimization and calculation using the SLAM algorithm, the precise 3D coordinates of the coding points within the calibration field, as well as the extrinsic parameters of the calibration board and points, are finally calculated. This step, involving extensive repeated observations, is one of the key points in ensuring the stability and accuracy of the calculation system. Extensive repeated observations are also a key aspect of this invention in ensuring the stability of the calculation system and improving the accuracy of the final extrinsic parameter calculation.

[0010] The following describes a more detailed method for calibrating a small calibration field calibration board and a high-precision extrinsic parameter calibration method, comprising the following steps: 1. Arrange calibration plates and independent coding points in the calibration field, and at the same time, increase the texture richness of the calibration field; 2. Calibrate the visual inertial SLAM hardware using a single calibration board to calibrate the internal and external parameters of the visual inertial SLAM hardware.

[0011] 3. Using measuring tools (such as ordinary measuring rulers, total stations, etc.), measure the approximate extrinsic parameters between calibration plates, the initial position values ​​of independent coded calibration points, or the approximate distance between coded points in the calibration plates and other calibration coded points or independent coded points, and integrate other prior information into the configuration of the SLAM algorithm.

[0012] 4. Within the calibration field, run the visual-inertial SLAM algorithm modified according to the optimization scheme. The visual-inertial sensor device needs to move sufficiently within the calibration field so that the coded corner points on the calibration board, as well as independent coded corner points, can be repeatedly observed and triangulated multiple times (no less than 30 times). Use the coded corner points on the calibration board within the calibration field, or the independently set coded corner points within the calibration field, as landmark points for preliminary optimization calculations of SLAM front-end tracking.

[0013] 5. In the backend BA optimization of SLAM, prior information serves as a priori constraint on the value to be optimized. During the optimization solution, the prior information is incorporated into the optimization algorithm. This prior information includes the initial values ​​of the calibration board extrinsic parameters measured previously, the estimated extrinsic coordinates of the independent coded calibration points, the baseline length and orientation constraints of the marker points [optional process], and the scale parameters of the calibration board. Finally, the LM algorithm is used to determine the final calibration field, calibration board extrinsic parameters, and independent coded points.

[0014] 6. SLAM will output the optimized state variables to obtain the extrinsic parameters between calibration boards and the extrinsic parameters of the independent coded control points.

[0015] Multiple calibration boards and independent coded calibration points can be deployed within the scene. The approximate lengths between these calibration boards and points can be measured and input into the optimization model as prior constraints with a certain confidence level. In the figure, the lines connecting the coded points in the calibration boards illustrate the approximate measurement method. The device shown at the bottom of the figure is a VIO device combining a high-precision camera and an IMU.

[0016] In summary, this invention addresses the following: 1. By utilizing a SLAM VIO device with a large image sensor target surface, lens focal length, and baseline length, and running a SLAM algorithm with prior constraint information, it identifies and solves for calibration boards and independently coded calibration points, ultimately solving for the extrinsic parameter schemes of the calibration boards and points, thus solving the problem of constructing small calibration fields. 2. Regarding the SLAM VIO device hardware, a key point is that the image sensor target surface is no less than 1 inch, and the baseline length of the dual cameras is no less than 40 cm, distinguishing it from ordinary SLAM VIO devices. 3. The SLAM algorithm with prior constraint information typically includes: approximate initial values ​​of extrinsic parameters, prior values ​​of the calibration board and independently coded points (including length and angle information), and other constraint information (such as GPS ground truth values ​​with cm-level accuracy outdoors and motion capture ground truth values ​​with sub-millimeter accuracy indoors). 4. The SLAM VIO device is moved multiple times within the calibration field to enable repeated observation of corner points on the coding calibration board and corner points on independent coding points, achieving more than 10 repeated observations.

Claims

1. A method for high-precision extrinsic parameter calibration of a small calibration field calibration plate and point, characterized in that, Includes the following steps: 1) Using a single calibration board, the visual inertial intrinsic and extrinsic parameters of the visual-inertial binocular SLAM hardware system were calibrated; 2) Use truth-based tools to measure the approximate initial values ​​of the extrinsic parameters between the calibration plates, as well as other constraint information, as priors for the optimization algorithm to solve for the extrinsic parameters; 3) Within a local area of ​​the calibration site, use the above-mentioned VIO hardware to execute a visual-inertial SLAM algorithm with prior information constraints; 4) By repeatedly moving the SLAM VIO device, the corner points on the coding calibration board and the corner points on the independent coding points are repeatedly observed. With the specially configured SLAM VIO hardware, the mapping and optimization accuracy of the corner points is ensured. Through continuous optimization and calculation of the SLAM algorithm, the accurate 3D coordinate values ​​of the coding points in the calibration field and the extrinsic parameters of the calibration board and points are finally calculated.

Images