论文总字数：19860字

摘 要

随着科技的进步以及人们生活需求的提高，物体的三维位姿估计得到了飞速的发展。更是在机器视觉技术崛起并快速发展的今天，原有的位姿估计技术被颠覆，借助机器视觉构建物体位姿的技术得到完善，并被广泛使用。因其操作简单，且准确性高。当今社会，在工业生产、服务机器人抓取、智慧城市系统等诸多领域都开始尝试使用机器视觉的方式构建物体三维位姿。

本设计主要从单目视觉领域出发，在前人的基础上探索构建物体三维位姿的方法。基于HALCON软件平台，采用基于形状的 3D 匹配的位姿估计。首先需要获取被测物体的清晰照片，然后对图片进行滤波、平滑、锐化等简单预处理。如有需要，也可进行腐蚀膨胀。以减少因拍照因素而带来的噪声干扰，使最终构建的三维位姿模型更加准确。对预处理过的图像进行边界提取，并使用算法提取其特征点。为之后的特征匹配做准备。以提取到的特征信息为基础，在HALCON中利用相应算法进行基于性状进行三维位姿匹配。并完成物体三维位姿的估计。最终完成物体三维位姿估计系统的开发。

关键词：机器视觉 单目视觉 三维位姿 特征提取 三维位姿匹配 滤波 平滑 锐化

Development of 3D pose estimation system based

on monocular vision

Abstract

With the advancement of technology and the improvement of people's living needs, the three-dimensional pose estimation of objects has been rapidly developed. Even today, with the rise of machine vision technology and rapid development, the original pose estimation technology has been subverted, and the technology of constructing object poses with machine vision has been perfected and widely used. Because of its simple operation and high accuracy. In today's society, in the fields of industrial production, service robotic capture, and smart city systems, we have begun to use machine vision to construct three-dimensional poses of objects.

This design mainly starts from the field of monocular vision and explores the method of constructing the three-dimensional pose of objects on the basis of predecessors. Based on the HALCON software platform, pose estimation based on shape-based 3D matching. First, you need to get a clear picture of the measured object, and then filter, smooth, sharpen and other simple preprocessing. Corrosion expansion can also be carried out if necessary. In order to reduce the noise interference caused by the photographing factors, the final constructed three-dimensional pose model is more accurate. Perform boundary extraction on the preprocessed image and extract its feature points using an algorithm. Prepare for subsequent feature matching. Based on the extracted feature information, the corresponding algorithm is used in HALCON to perform three-dimensional pose matching based on traits. And complete the estimation of the three-dimensional pose of the object. Finally, the development of the three-dimensional pose estimation system of the object is completed.

Keywords: Machine vision；Monocular vision；Three-dimensional pose；Feature extraction；Three-dimensional pose matching；Filter；smooth；Sharpen

目 录

摘 要 I

Abstract II

目 录 III

第一章 绪论 4

1.1 课题研究的背景及意义 4

1.2 研究的现状介绍 5

1.2.1机器视觉的发展 5

1.2.2单目视觉姿态估计 6

1.3论文的研究方向和内容 6

第二章 相机的选择 8

2.1 坐标系 8

2.2 相机的选择及参数 9

第三章 图像预处理 11

3.1 图像的获取 11

3.2 图像的初级预处理 12

3.3 图像的形态学处理 14

3.4 图像的blob分析 16

第四章 图像匹配 17

4.1 基于像素的图像匹配 17

4.2 基于特征的图像匹配 17

第五章 3D 匹配姿态估计 19

5.1 3D 匹配姿态估计的方法 19

5.2 基于形状的 3D 匹配的姿态估计 20

5.3 基于形状的3D匹配的姿态估计过程中可能存在的异常 21

参考文献 23

结语 25

致谢 26

第一章 绪论

请支付后下载全文，论文总字数：19860字