【作者】 陈济棠；

【导师】 徐杜；

【作者基本信息】 广东工业大学 ， 通信与信息系统， 2011， 硕士

【摘要】 计算机视觉是一门新兴的技术,而基于计算机视觉的测量技术作为一种非接触式的先进测量技术,具有精度高、效率高、成本低等诸多优点,有着广阔的应用前景。双目立体视觉是计算机视觉的一个重要分支,其原理是由不同位置的两台摄像机或者一台摄像机(CCD)经过移动或旋转拍摄同一幅场景,通过计算机空间点在两幅图像中的视差,获得该点的三维坐标值。由于双目测量具有一定的三维测量精度和测量的实时性,且花费的代价较其他方法要小的多,所以,双目视觉在测量方面应用广泛。除了对实际物体的测量之外,在其他领域的应用也取得快速发展。采用双目视觉测量,仅需从两幅对应图像中抽取必要的特征点的三维坐标,其信息量少,处理速度快,有利于提高检测精度,尤其适于动态情况。随着计算机与机器人技术的发展,双目的研究逐渐深入。从以前的射线三角定理发展到现在的平行校正定理,优化了搜索匹配点的方法,实现了对图像进行平行化后可以快速的匹配方案,简化了深度信息的还原公式。双目视觉现在已经普遍应用于三维定位装置,为CNC加工提供三维运动轨迹信息,同时可以为机器手的装配工作提供三维信息。目前热门研究的应用包括体感游戏、虚拟屏幕(可以在空间对屏幕进行操作)、机器人识路等。本文从角点的提取、摄像机标定、图像预处理,特征提取、立体匹配、平行校正以及目标空间定位几个方面对双目三维测量技术开展研究。本文主要研究工作有：提出一种在保持原有边缘信息的基础上减小噪声的图像预处理方法,提出并编程实现一种平整度的测量方法,编程实现了双目视觉系统的标定以及三维物体的测量。本文由六个章节构成。第一章为绪论部分,主要介绍了此课题的来源、研究的意义、研究内容和国内外发展的状况。第二章主要介绍了计算机视觉系统、图像预处理方法,以及双目立体视觉和三维重构。第三章从光学的物理特性推导出数学模型,介绍了内参与外参的概念,为标定建立一种从物理世界到数学的模型。介绍了几种经典的标定算法,并给出详细的算法过程,为三维还原建立了基础。第四章分析了几种图像预处理方法,从图像预处理出发,对拍摄得到的图像进行优化,使图像特征更明显,最后利用经典Harris算子提取出角点亚像素信息。除此之外,还介绍了几种经典的标定方法,并分析其优劣和详细的算法过程。从而为第五章的匹配的正确率提供了保障。第五章首先介绍图形特征的几种描述方法以及其匹配的方法,接着重点介绍了SIFT算法的应用,并给出匹配的结果。第六章重点讲解了平行校正原理,对第二章标定的结果进行平行校正,校正后的图像简化了深度的测量方法。紧接着给出了两种被测物体的三维测量实验数据的实例,实验结果证明标定结果可靠,三维还原信息准确。更多还原

【Abstract】 Computer vision is an emerging technology, as a non-contact measurement technology based on computer vision.It got the advanced of high precision, high efficiency, low cost and so on,It has broad application prospects. Stereo binocular vision is an important branch of computer vision. Taking photo form different locations with one camera or take photo using two camera.So as to obtained the parallax,and use the parallax to calculate the 3d coordinates.Because of binocular measuring both can satisfy the measurement of real-time, and cost price much smaller than other methods are. So, binocular vision application in the measurement is more widely. In addition to the actual object measurement, in other areas outside it also get some progress. To measure by binocular vision, we only need extraction of essential feature from two corresponding image to get the 3d coordinate. The Information is less and can calculate fast.It’s helpful for improving precision, especially suitable for dynamic situation.Along with the development of computer and robot.Binocular research is also gradually in-depth, from the former triangle theorem to the rays theorem.It optimize the search method of matching point. Achieved quickly match scheme and simplified formula of reduction depth.Binocular vision is now generally applied to the 3d positioning device, provide 3d trajectory information for CNC machining, and also provide 3d information for a machine assembly hand.. Currently popular research applications including body feeling games, virtual screen (it can control screen in the air), machine navigate,etc.This paper study the corner extraction, image pre-processing, cmera calibration, feature extraction, Stereo matching, parallel correction and target space positioning aspects. This paper mainly do the following:1. design research work that can keep original Edge information and also to remove noise.2. programe a sofeware to mease flatness;3. programe a sofeware to do the calibration and calculate depth.This paper consists of six chapters.Chapter 1 is the introduction part, mainly introduced the computer vision technology research at home and abroad. Then, starting from the image preprocessing, we optimize the photo, so as to make the image features more apparent, extracted with classical Harris operator to get corner subpixel information. And introduces several classic calibration method, analyzes the advantages and disadvantages, and give the detailed algorithm process.Chapter 2 mainly presents the computer vision system, image processing, binocular stereo vision and 3-D reconstruct technology.Chapter 3 translates optical physical properties to mathematical model, and introduces the concept of outside parameters and inside parameters participation. establish a relationship for physical world and Mathematical model, This paper introduces some typical calibration algorithm, and give detailed algorithm process. Establish foundation For the 3d reduction.Chapter 4 explains general image preprocessing method, the purpose is to make feature more prominent. Providing matching accuracy for the fourth chapter..Chapter 5 introduces several description method and its matching method, then Emphasis on sift algorithm, and give the Matching results.Chapter 6 emphasized parallel correction method, using the calibration results of the second chapter to do the parallel correction.After parallel correction,it can simplify the Depth measurement equation.In this chapter we give examples of 3d reconstruction, measure the length of the object. Experimental proof calibration results are reliable, and 3d it can restore information accurately.更多还原