CN-121810816-B - Polynomial fitting-based large-view-field camera light calibration method and system

CN121810816BCN 121810816 BCN121810816 BCN 121810816BCN-121810816-B

Abstract

A method and a system for calibrating light of a large-view-field camera based on polynomial fitting are provided, wherein measuring targets are respectively arranged on a camera and a rigid tool of a calibration plate, a fixed coordinate system is established, pose of the coordinate system is directly measured by using external high-precision equipment such as a laser tracker, and decoupling of pose errors and an image algorithm is realized. The method comprises the steps of acquiring a plurality of calibration plate images, respectively carrying out global polynomial surface fitting on sparse feature angular points of each image, establishing continuous analytic mapping from a pixel to a three-dimensional space, and carrying out space straight line fitting on a plurality of three-dimensional points corresponding to the same pixel to obtain an accurate ray equation of the pixel. The calibration result is stored based on the camera target seat coordinate system, and the conversion relation between the coordinate system and the world coordinate system can be directly transferred and used only by re-measuring the conversion relation during actual measurement. The method is suitable for any complex imaging system and has the advantages of high precision, strong universality, good robustness, outstanding engineering practicability and the like.

Inventors

WU ZHEN
NIU ZHENQI
WAN SONGLIN
ZHAO YANAN
JIANG GUOCHANG
WEI CHAOYANG

Assignees

中国科学院上海光学精密机械研究所

Dates

Publication Date: 20260512
Application Date: 20260309

Claims (10)

1. The large-view-field camera light calibration method based on polynomial fitting is characterized by comprising the following steps of: Establishing a calibration plate pixel coordinate system { P } and a calibration plate target seat coordinate system { S } which are fixedly connected with a calibration plate, establishing a camera target seat coordinate system { C } which is fixedly connected with a camera to be calibrated, and pre-calibrating a static conversion matrix T {P→S} between the calibration plate pixel coordinate system { P } and the calibration plate target seat coordinate system { S }; The data acquisition step comprises the steps of placing a calibration plate in a plurality of different poses in a camera view field, for each pose, directly measuring a dynamic transformation matrix T n {S→C} of a coordinate system { S } of a target seat of the calibration plate relative to a coordinate system { C } of the target seat of the camera by using high-precision external measurement equipment independent of the camera, and simultaneously acquiring an image of the calibration plate to obtain pixel coordinates of characteristic angular points Calculating three-dimensional coordinates of each characteristic corner under a camera target seat coordinate system { C } according to the static transformation matrix T {P→S} and the dynamic transformation matrix T n {S→C} Global map construction step, for each pose n, using pixel coordinates of characteristic corner points Is an independent variable, and takes three-dimensional coordinates of the independent variable under the { C } system As dependent variables, respectively performing independent polynomial surface fitting on three space coordinate components of X, Y and Z, and establishing a global analytic mapping function P (u c , v c ) from any pixel coordinate to a three-dimensional point under a { C } system; Generating a light model, namely, for any pixel coordinate According to the global analytic mapping function P (u c , v c ) under N different poses, N three-dimensional points under the { C } system are obtained, and the light parameters corresponding to the pixels are obtained by carrying out space straight line fitting on the N points, so that a camera light model is generated.
2. The polynomial fitting-based large-field-of-view camera light calibration method according to claim 1, wherein in the reference establishing step, the pre-calibration of the static conversion matrix T {P→S} is obtained by simultaneously measuring physical feature points of at least three known pixel coordinates on a calibration plate and target feature points on a calibration plate target seat coordinate system { S } by using a high-precision external measurement device, and solving by using a point cloud fitting or absolute orientation algorithm.
3. The method for calibrating light rays of a large-field camera based on polynomial fitting according to claim 1, wherein in the global map construction step, a zernike polynomial or a legendre polynomial is adopted as a basis function for polynomial surface fitting, and continuous analytical expression of any complex optical distortion is realized through linear combination of the basis functions.
4. The method for calibrating light of a large-field camera based on polynomial fitting according to claim 1, wherein in the step of generating a light model, the step of uniquely determining the origin O (u c , u c ) of the light is further included in the step of defining a point on the fitted spatial straight line satisfying a preset geometric constraint as the origin of the light.
5. The polynomial fitting-based large-field-of-view camera light calibration method according to claim 4, wherein the preset geometric constraint is a center of gravity of a projection point of the N three-dimensional points on a plane passing through an origin of a coordinate system { C } of a camera target holder and perpendicular to the spatial linear direction and an intersection point of the plane and the linear line, or on a plane passing through an origin of a coordinate system { C } and perpendicular to the spatial linear direction.
6. The polynomial fitting-based large-field-of-view camera light calibration method according to claim 1, wherein in the light model generation step, in order to improve calculation efficiency, sparse sampling is performed on pixels on a camera image surface, spatial straight line fitting is performed on sampling points only to generate a sparse light field, and light parameters of non-sampling points are obtained by interpolating light parameters of adjacent sampling points.
7. The method for calibrating the light of the large-field camera based on polynomial fitting according to claim 1 is characterized by further comprising a coordinate system migration application step of storing a calibrated light model based on a camera target holder coordinate system { C }, and in an actual measurement system, measuring a conversion matrix T {C→W} between a { C } system and a world coordinate system { W }, converting the light model into the world coordinate system for use, and decoupling and multiplexing a calibration result and a measurement scene are realized.
8. The polynomial fitting-based large-field-of-view camera light calibration method according to any one of claims 1 to 7, wherein the high-precision external measurement device comprises one or more of a laser tracker, a laser radar, a total station, a photogrammetry system or a three-coordinate measuring machine, the calibration plate is a physical medium or a programmable display device capable of presenting a preset characteristic corner pattern, and the characteristic corner is a checkerboard corner, a circular mark point or a mark point obtained by stripe decoding.
9. A large field of view camera light calibration system for implementing the method of any one of claims 1-8, comprising: The surface of the calibration plate unit is provided with a preset characteristic angular point pattern, and the rigid tool is fixedly provided with at least three first measurement targets which are not collinear and used for establishing a coordinate system { S } of a target seat of the calibration plate; the camera unit is fixedly provided with at least three second non-collinear measuring targets on the rigid tool and is used for establishing a camera target seat coordinate system { C }; The high-precision external measurement device is independent of the camera unit and is used for directly measuring the space positions of the characteristic points of the first measurement target and the second measurement target so as to solve a static transformation matrix T {P→S} and a dynamic transformation matrix T n {S→C} ; And the data processing unit is respectively connected with the camera unit and the high-precision external measurement equipment and is used for receiving the image data and the measurement data, performing coordinate conversion calculation, polynomial mapping function fitting and space straight line fitting, and generating and storing a camera light model.
10. The large field of view camera light calibration system according to claim 9, wherein the data processing unit communicates with the high precision external measurement device in real time, to synchronize measurement data with image acquisition, and is configured to perform sparse sampling of camera image plane pixels and interpolation calculation of non-sampling point light parameters.

Description

Polynomial fitting-based large-view-field camera light calibration method and system Technical Field The invention belongs to the technical field of precision optical measurement and machine vision, and particularly relates to a large-view-field camera light calibration method and system based on polynomial fitting. The method is particularly suitable for vision measurement systems with extremely high requirements on the accuracy of the direction of the light rays of the camera, such as phase deflection measurement, three-dimensional imaging of structured light, precise assembly and detection of vision guidance and the like, and can provide a general, high-accuracy and specific parameterized model-independent light ray direction calibration solution for any imaging optical system (including systems with obvious aberration or adopting non-central projection models). Background In a precise measurement system such as surface shape deflection measurement, structured light three-dimensional reconstruction and the like, the light calibration precision of a camera directly influences the measurement accuracy of the whole system. The traditional camera calibration model is mainly divided into two main types of parameterized small-hole imaging models and non-parameterized light models. The Zhang Zhengyou calibration method based on the pinhole imaging model performs well under the conventional perspective imaging condition, but for scenes with obvious optical aberration or non-central projection imaging systems such as telecentric lenses and light field cameras, the calibration accuracy tends to be reduced due to model errors. The calibration method based on the light model does not depend on a specific optical parametric model, the camera is regarded as a black box, and the calibration is realized by establishing the mapping relation between pixels and three-dimensional space light, so that the method has better adaptability to a complex imaging system. However, when the existing light calibration method is developed to the high precision and high practicability, the technical challenges are still faced, namely ① calibration precision is limited by an image algorithm, the existing method depends on a plurality of calibration plate images, and the relative pose between the calibration plates is estimated through geometric constraint among feature points. The process tightly couples the final calibration precision with the image feature extraction quality and the optimization algorithm performance, the precision is easily influenced by image noise, feature point distribution and numerical optimization stability, and certain limitation exists in metering-level high-precision application. ② The locality and smoothness of the spatial mapping construction are insufficient, namely, after the corresponding relation of sparse feature points is acquired, the existing method generally adopts local interpolation based on a discrete point set to estimate the spatial position of a non-feature pixel. The method has good effect in the dense areas of the characteristic points, but the problem that interpolation errors are increased and mapping results are not smooth easily occurs in the edges of the view fields or the sparse areas of the characteristic points, and unexpected fluctuation can occur in the fitting of the light ray directions. ③ The calibration process is complex and has no system integration, and in order to obtain high-precision pose estimation, the existing method often needs to strictly control the placement pose of a calibration plate, the image acquisition condition and the consistency of feature extraction. In actual operation, factors causing image quality degradation or unstable feature extraction may directly affect repeatability and accuracy of calibration results, and high requirements are put on experience of operators and experimental environments, and in addition, camera calibration results cannot be simply transferred to a new measurement system. Therefore, there is still a need for a camera light calibration method with better adaptability, stability and operation friendliness while ensuring high precision, so as to support further development and application of various precision vision measurement systems. Disclosure of Invention The invention provides a large-view-field camera light model calibration method and system based on a polynomial, which are used for solving the problems that the existing camera calibration technology has strong dependence on an image algorithm, is not smooth in space mapping construction, is complex in calibration flow and is difficult to migrate. According to the method, targets with known three-dimensional coordinates are shot at different positions, global surface fitting is conducted on pixel-three-dimensional coordinate relations in a camera view field at each position by utilizing a polynomial, and then space straight line fitting is conducted on a plurality of three