基于SVR回归拟合的摄像机标定算法  

Camera Calibration Algorithm Based on SVR Regression Fitting

作　　者：郑百强 王泽民[1] 王炜博 ZHENG Baiqiang;WANG Zemin;WANG Weibo(College of Telecommunications,Xi’an Technological University,Xi’an 710021)

机构地区：[1]西安工业大学电信学院,西安710021

出　　处：《现代制造技术与装备》2023年第2期30-34,共5页Modern Manufacturing Technology and Equipment

摘　　要：在摄像机标定过程中,为得到像素坐标中不同像素坐标值对应的唯一世界坐标值,需要求解相机坐标系和世界坐标系之间的旋转矩阵参数、平移矩阵参数及物理器件特性参数,操作步骤烦琐。因此,提出了一种支持向量回归(Support Vector Regression,SVR)拟合一体化标定算法。首先,为实现图像平面坐标系到世界坐标系的坐标转换,选择大小为26 mm×26 mm的棋盘格作为标定块。每次采集棋盘格,统计图像中的棋盘角点。多次重复采集后,整理图像坐标与世界坐标的数据。其次,训练SVR回归模型,得到SVR的核函数和惩罚因子。最后,定量分析实验结果。相较于传统的标定方法,SVR的方法省略了求解相机畸变的步骤,解决了相机的非线性问题,具有良好的精确度和鲁棒性。In the camera calibration process, in order to obtain the unique world coordinate values corresponding to the different pixel coordinate values in the pixel coordinates, the rotation matrix parameter, translation matrix parameter and physical device characteristic parameter between camera coordinate system and world coordinate system need to be solved. Therefore,an integrated calibration algorithm of Support Vector Regression(SVR) fitting is proposed. First of all, in order to realize the coordinate transformation from plane coordinate system to world coordinate system, a 26 mm×26 mm checkerboard grid is selected as the calibration block. Every time the collection of checkerboard grid, statistical image of the checkerboard corners. After repeated collection, collate image coordinates and world coordinates data. Secondly, the SVR regression model is trained, and the kernel function and penalty factor of SVR are obtained. Finally, the quantitative analysis results. Compared with the traditional calibration method, the SVR method omits the steps to solve the camera distortion, and solves the nonlinear problem of the camera,has good accuracy and robustness.

关 键 词：摄像机标定 成像模型 支持向量回归(SVR) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP181[自动化与计算机技术—计算机科学与技术]