基于单摄像机和神经网络的三维测量研究  

Study on 3D measurement based on single camera and neural network

作　　者：陈薇伊 隋国荣[1] CHEN Weiyi;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,上海200093

出　　处：《光学技术》2022年第2期214-222,共9页Optical Technique

摘　　要：为了实现更为简单的三维测量同时解决传统立体视觉方法中基线的限制问题,提出一种单摄像机微小角度旋转结合神经网络进行三维测量的方法。方法通过单摄像机小角度旋转采集二维图像数据;利用电动平台移动标定板构建三维空间坐标系;基于摄像机的线性成像模型思想,将代表图像坐标和三维坐标映射关系的投影矩阵替换为BP神经网络,得到二维坐标到三维坐标的直接映射,实现微小基线下的三维测量。实验结果表明,相比于传统方法测量失真的结果,提出的方法对标定板的尺寸测量绝对误差为0.0864mm。方法将神经网络融入单摄视觉领域,可以克服微小基线场景下传统视觉测量方法的缺陷。对移动设备、监控设备及狭窄场景下的三维信息获取具有潜在的应用价值。In order to simplify the three-dimensional measurement and solve the baseline limitation problem in traditional stereo vision methods,a method based on single-camera micro-angle rotation combined with neural network for three-dimensional measurement is proposed.The method gathers two-dimensional image data by rotating a single camera at a small angle.Additionally,the three-dimensional spatial coordinates are established by moving the chessboard calibration target with an electric sliding table.Based on the idea of camera linear imaging model,where the projection matrix represents the mapping relationship between image coordinates and 3 D coordinates is replaced by BP neural network.It obtains the direct mapping from 2 D coordinates to 3 D coordinates and realizes 3 D measurement under micro baseline.The experimental results show that the absolute error of the proposed method is 0.0864 mm compared with the distortion measured by the traditional method.This method integrates neural network into the single-shot vision field,which can make up for the defects of traditional vision measurement methods in micro baseline scene.It has potential application value for mobile equipment,monitoring equipment and 3 D information acquisition in narrow scenes.

关 键 词：光学测量 微小基线 摄像机标定 BP神经网络 

分 类 号：TP394.1[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]