单摄像机立体视觉测量系统的高精度变焦标定技术  

High-precision zoom camera calibration of stereo vision measurement system with single camera

作　　者：殷晨晖 褚鑫磊 杨珊 李丽莹 隋国荣[1] YIN Chenhui;CHU Xinlei;YANG Shan;LI Liying;SUI Guorong(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院

出　　处：《光学技术》2019年第6期668-676,共9页Optical Technique

基　　金：国家973计划(2015CB352001);上海市重点学科项目(S30502)

摘　　要：针对现有用于光学测量的双目变焦系统标定方法难度大、测量精度受限于两摄像机内参一致性等问题。提出一种基于单摄像机的平行双目立体视觉系统实现及其高精度变焦标定方法。方法基于三角测量原理采集图像,利用高精度位移台驱动单摄像机进行平移以保证基线精度;求解离散焦距下的预标定结果并利用BP神经网络模型对其进行拟合,以实现任意焦距下系统内外参数动态估计。实验结果表明,系统预标定的重投影误差小于0.1664 pixel,变焦后图像畸变校正平均误差为0.0982 pixel,立体视觉测量尺寸绝对误差小于0.05mm。方法能弥补传统变焦标定方法的不足,消除双目内参不一致引入的误差,提高视觉系统的测量精度。There are many problems in the calibration process of zooming binocular system for optical measurement.For example,the calibration methods are difficult,and the accuracy is limited by consistency of the internal parameters of two cameras.To solve these problems,a zoom calibration method is proposed to realize parallel binocular stereo vision based on a monocular camera and to get high-precision zoom calibration.It is based on the Triangulation Principle to gather images.The camera is driven to translate by a precision stage to ensure the accuracy of the baseline.Then,the pre-calibration results of the discrete focal lengths are solved and fitted by the BP neural network model.This method can dynamically estimate the internal and external parameters of the system at any focal length.The results show that pre-calibrated reprojection error is less than 0.1664 pixel,the average error of image distortion correction after zooming is 0.0982 pixel,and the absolute error of stereo vision measurement is less than 0.05 mm.This method can make up for the deficiencies of the traditional zoom calibration method,eliminate the error introduced by the inconsistency of the binocular internal parameters,and improve the calibration accuracy of the system.

关 键 词：光学测量 摄像机标定 变焦镜头 立体视觉 BP神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TB811[自动化与计算机技术—计算机科学与技术]