条纹投影三维测量的若干关键技术的研究

发布时间：2018-04-02 20:46

本文选题：条纹投影　切入点：系统标定　出处：《深圳大学》2017年硕士论文

【摘要】：三维测量技术是一种关键的机器视觉技术,它在逆向工程、图像识别、人工智能、虚拟现实以及生物医学等方面都有着重要的应用。其中传统的接触式三维测量存在测量耗时长、弹性材料测量受限等缺点,因此无法满足工业生产的需要。条纹投影的三维测量技术是一种非接触式的方法,这种测量技术具有效率高、测量准确的优点,因此成为了三维测量领域研究的热点。本文针对条纹投影中的一些关键技术展开研究,比如基于相位匹配的多传感器三维测量系统、大视场的三维测量系统以及3D点云配准等若干关键技术,并对这些技术进行了分析和总结,本文的主要内容如下:1.介绍了条纹投影三维测量技术的基本原理以及国内外研究现状,阐述了PMP(相位测量轮廓术)、格雷编码、摄像机成像模型,以及对获取到的3D点云数据处理的基本方法。2.由于单传感器测量范围的限制,多传感器测量系统逐渐成为获取物体完整3D面型的重要的方法。然而,传统的多传感器测量系统,它们相互之间的系统标定是分开的,整个测量系统的标定耗时长、操作复杂,不能满足现场标定的需求。针对传统方法的不足,本文提出了一种基于相位传递的多传感器测量系统,每套测量系统的标定相互关联,并且在完成整个系统标定后,每套测量系统获得的3D点云数据无需再次配准,自动转化到统一的坐标系下,实现3D点云的全局配准。3.为了能够满足大视场三维场景的测量需求,我们设计了两套三维数据采集系统,第一套采用了一种线结构光旋转扫描的方法来获取室内环境的3D点云数据,该系统能够实现360度场景的测量,但是测量速度慢、效率低。第二套则是利用鱼眼镜头具有宽视场角和大景深的优势,来实现双目立体视觉的大视场三维测量,但是由于镜头本身的特性,基于小孔成像模型的双目立体视觉3D测量方法不再适用于鱼眼镜头模型。因此,本文提出了一种基于球面模型的双目立体视觉测量系统,将测量物体的所在的世界坐标直接地转化到单位球面坐标系,由于采用球面模型代替了平面透视投影模型,因此测量系统的FOV(视场角)不再受限。同时,针对双目测量系统中的特征点匹配问题,本文提出了一种菱形匹配算法,极大的提高了查找的速度和测量的精度。4.在获取测量场景的三维信息后,需要对获取的多组三维点云数据进行点云配准处理。本文提出了一种基于2D图像的3D点云配准算法。首先,通过计算相邻两帧2D图像之间的仿射变换矩阵,完成相邻两组3D点云之间的粗配准;然后,通过2D坐标确定配准对应点的搜索区域,完成精配准;最后,再次通过3D坐标对应的2D坐标来进行全局配准。实验验证,这种方法极大的提高了配准的速度,完成了多个视角3D点云数据的拼接。
[Abstract]:3D measurement technology is a key machine vision technology, which has important applications in reverse engineering, image recognition, artificial intelligence, virtual reality and biomedicine.The traditional contact 3D measurement has the disadvantages of long measuring time and limited measurement of elastic materials, so it can not meet the needs of industrial production.The three-dimensional measurement technology of fringe projection is a kind of non-contact method. It has the advantages of high efficiency and accurate measurement, so it has become a hot spot in the field of three-dimensional measurement.In this paper, some key techniques in fringe projection, such as multi-sensor 3D measurement system based on phase matching, 3D measurement system with large field of view and 3D point cloud registration, are studied.These techniques are analyzed and summarized. The main contents of this paper are as follows: 1.This paper introduces the basic principle of fringe projection 3D measurement technology and the current research situation at home and abroad, and expounds the basic methods of PMP (phase measurement profilometry, Gray coding, camera imaging model) and 3D point cloud data processing.Due to the limitation of single sensor measurement range, multi-sensor measurement system has gradually become an important method to obtain the complete 3D surface shape of object.However, the calibration of the traditional multi-sensor measurement system is separate from each other. The calibration of the whole measurement system is time-consuming and complex, so it can not meet the requirements of field calibration.In view of the shortcomings of traditional methods, a multi-sensor measurement system based on phase transfer is proposed in this paper. The calibration of each set of measurement systems is related to each other, and after the calibration of the whole system is completed,The 3D point cloud data obtained by each measurement system need not be registered again, and it is automatically transformed to a unified coordinate system to realize the global registration of 3D point cloud.In order to meet the needs of 3D scene measurement in large field of view, we designed two sets of 3D data acquisition systems.The system can measure 360 degree scene, but the measuring speed is slow and the efficiency is low.The second set uses the advantages of wide field of view angle and large depth of field of field of field of view to realize the large field of view measurement of binocular stereo vision, but due to the characteristics of the lens itself,Binocular stereo vision 3D measurement method based on pinhole imaging model is no longer suitable for fish-eye lens model.Therefore, a binocular stereo vision measurement system based on spherical model is proposed in this paper, in which the world coordinates of measuring objects are directly transformed to unit spherical coordinate system, because the spherical model is used to replace the plane perspective projection model.Therefore, the FOV (Field of View Angle) of the measurement system is no longer limited.At the same time, aiming at the problem of feature point matching in binocular measurement system, a rhombus matching algorithm is proposed in this paper, which greatly improves the speed of searching and the precision of measurement.After obtaining the 3D information of the measurement scene, it is necessary to deal with the point cloud registration of multiple groups of 3D point cloud data.This paper presents a 3D point cloud registration algorithm based on 2D images.Firstly, by calculating the affine transformation matrix between two adjacent frames of 2D images, the rough registration between adjacent groups of 3D point clouds is completed. Then, the searching area of corresponding points is determined by 2D coordinates to complete the fine registration.Again, the global registration is carried out by 2D coordinates corresponding to 3D coordinates.Experimental results show that this method greatly improves the registration speed and completes the 3D point cloud data splicing of multiple views.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】