一种基于BP神经网络测量物体高度的快速方法.pdf

理论与算法 2017.10 一种基于BP神经网络测量物体高度的快速方法 欧阳美龙 （辽东学院，辽宁丹东，118003 ） 摘要 ：本文提出的方法是在相位测量法的基础上，利用神经网络建立折叠相位与高度的映射关系，不需要严格搭建系统，也不 需要展开相位及标定系统，不必考虑由系统的非线性所带来的误差。测量结果的精度在十个微米左右，标准方差在一个微米以 下，是一种快速有效且准确稳定的机器视觉高度测量方法。 关键词 ：机器视觉检测；高度测量；光栅投影；相位测量法；人工神经网络；相位-高度映射 A Fast Method for Measuring Objects’ Height Based on BP Neural Network OuYang Meilong (Eastern Liaoning University,Dandong Liaoning,118003) Abstract: In this paper, a new effective and simple machine vision measuring method is proposed. This method aims at solving a problem that the central position of laser can hardly be extracted accurately in the methods for measuring objects’ height based on line structured light. This method can slove objects’ phase by means of phase measuring profilometry, and build a nonlinear map between phase of structured light and height based on self-recall function of artificial neural network. Then the information of objects’ height can be obtained so long as the phases of objects are known. Its accuracy is about ten microns. Key words : Machine vision detection;Height measuring;Grating projection;Phase measuring profilometry;Map of phase-height;Artificial neural network 1 系统原理及组成 到物体的高度值。 2 实验设计 投影仪和摄像机采取斜投正采的方式并保持一定夹角不变， 一维平移台沿z轴在竖直方向上平移，采集回的条纹图片保存到 计算机中。 2.1 实验方法及过程 2.1.1 建立网络映射关系 由计算机控制投影仪向一维手动平移台上投射单条的正弦 条纹，平移台精度为0.01mm。以0.05mm的高度为间距移动平移