光学三维成像数据的自动量化分析方法与应用的研究

发布时间：2018-01-04 02:28

  本文关键词：光学三维成像数据的自动量化分析方法与应用的研究　出处：《南京大学》2016年博士论文　论文类型：学位论文

  更多相关文章： 光学三维成像 自动量化分析 三维图像处理 立体视觉 光学相干层析成像 柔性激光加工 生物医学成像

【摘要】：基于光学原理的三维成像方法是人们对现实三维世界作立体感知的重要手段,近年来,伴随着计算机技术和光电子器件的快速发展,光学三维成像数据在图像的分辨率、信噪比、数据量上均比传统的成像手段有了大幅的提升。但是,在获得了这些高质量的成像数据后,如何高效地对其进行分析成为了一个新的难题：一方面,在如生物医学领域上的许多应用中,仍存在对人工分析和解读的大量依赖,这些人工方法已不能满足大数据条件下的分析要求,使用计算机来进行自动数据分析已成为大势所趋；另一方面,三维成像获得的数据总是冗余的,需要从背景中将真正有意义的目标数据提取出来,而这一过程往往还涉及了对一些关于三维目标的几何结构、动态特性等关键参数的量化测量。在这一背景下,本文针对光学三维成像数据的自动量化分析方法进行了研究：首先,在光学三维成像数据的分析理论上,回顾和总结了当前图像分析和处理的一般性理论框架,针对光学工程领域对量化测量的需求,提出了一种新的多层次三维数据自动量化分析体系,打通了从三维成像到图像分析再到图像伺服之间的通路,为光学三维成像的数据后处理方法提供了一个较为完善的框架。其次,在三维成像数据的空间关系校准方法上,研究了单视图和多视图的空间关系校正方法,提出了适用于多种立体视觉系统结构的传感器内外部参数校准方案,显著提升了三维成像测量的精度。此外,还对三维图像伺服中多变换共存的空间关系校准问题进行了研究,提出了一种伺服投影矩阵的量化校准方案,实现了三维图像数据空间与现实世界中的伺服机构坐标系的精确校准。再次,在三维图像数据的特征提取方法上,分别从三维图像的边缘特征检测和区域分割算法上重点开展了研究：首先,在分析了当前三维图像的特征提取方法的研究现状的基础上,提出了一种基于动态规划算法的边缘检测方法,实现了稳定性较高的连续的单线线条提取；其次,还提出了一种使用区域属性来对生长过程进行约束的区域生长和分割的算法思想,结合实际应用设计的算法在层析堆叠图像的区域分割中取得了比传统方法更为优秀的特征提取效果。再次,在三维目标的自动量化评价方法上,进行了三维目标的模型简化、目标结构参数的自动量化测量等方法的研究,提出了一种利用投影空间关系来统一二维-三维数据的混合量化分析方法,扩展了分析方法的适用范围。此外,还提出了一种通过对二维图像特征提取精度的统计分析来反推三维数据质量的方法,实现了在硬件结构确定的情况下仅通过图像分析来进行三维成像精度的量化判定。最后,在光学三维成像数据的自动量化分析方法的应用上,分别对立体视觉伺服的柔性激光加工、基于立体视觉的羽毛形貌质量量化评价、基于内窥OCT的呼吸道管壁多层结构数据的自动分析、以及基于多普勒OCT的大脑动脉血流动态特性的量化分析等若干三维数据分析应用开展了探索。通过根据这些应用的实际需求来设计分析方法,提出了一些具有针对性的自动量化分析手段和系统,取得了较好的分析效果。总之,本文针对光学三维成像的数据分析问题展开了较全面的研究,主要创新性体现在提出了一些具有独创性和普适性的三维图像数据自动量化分析的新方法,大幅提高了三维图像数据分析的效率,在一定程度上解决了当前对光学三维成像数据的自动量化分析方法的迫切需求。
[Abstract]:The 3D imaging method based on optical principle is an important means of people to make stereo perception of reality 3D world in recent years, with the rapid development of computer technology and optoelectronic devices, optical 3D imaging data in image resolution, signal-to-noise ratio, the amount of data than the traditional imaging methods have been greatly improved. But and with these high quality imaging data, how to efficiently analyze it becomes a new problem: on the one hand, many applications in biomedical fields such as the, still exist to rely heavily on manual analysis and interpretation, these artificial methods cannot satisfy the requirement of big data analysis the use of computer, to carry out automatic data analysis has become represent the general trend; on the other hand, 3D imaging data obtained are always redundant, need from the background will be truly meaningful target data extraction Take out, and this process is often involved on the 3D object geometry, quantitative measurement of key parameters of dynamic characteristics. In this background, this paper aimed at the automatic quantification of optical 3D imaging data analysis methods are studied: firstly, the theoretical analysis on the optical 3D imaging data, and review summarizes the general theoretical framework of the current image analysis and processing, according to the demand for quantitative measurement of optical engineering field, proposes a new three-dimensional multi-level data automatic quantitative analysis system, open from 3D imaging to image analysis and image between the servo channel, provides a better framework for optical the 3D imaging data postprocessing method. Secondly, the spatial relationship of calibration of 3D imaging data, study the correction method of single view and multi view spatial relations, is proposed The internal and external parameters of the sensor calibration scheme is applicable to a variety of stereo vision system, significantly enhance the 3D imaging accuracy. In addition, the spatial relationship of 3D image servo calibration problem of multi transform coexistence, is proposed to quantify the calibration scheme of a servo projection matrix, to achieve accurate calibration of servo coordinate system the 3D image data space and in the real world. Once again, in the feature extraction method of three-dimensional image data, respectively from the edge feature of 3D image detection and segmentation algorithm focuses on the research: firstly, based on the analysis of the present status of the feature extraction method of 3D images on a the detection method of dynamic programming algorithm based on edge to achieve higher stability of the continuous single line extraction; secondly, it proposes a region is used To carry out regional growth and constraints on the growth process of the segmentation algorithm, combined with the practical application of the design algorithm has better characteristics than the traditional extraction method results in segmentation of the image stack chromatography. Again, the automatic quantitative evaluation method in three-dimensional target, simplified 3D target model. Research on automatic quantitative measurement method of target structure parameters, is proposed by using the projection space relationship to unify two hybrid quantization 3D data analysis method, expanded the scope of analysis method. In addition, it is presented by using the method of statistical analysis to the accuracy of the extraction quality of 3D data on backstepping 2D image feature and realizes the quantitative judgment of 3D imaging accuracy through image analysis only identified in the hardware structure. Finally, the automatic optical 3D imaging data The quantitative analysis of the application of the method, flexible laser processing of stereo visual servoing respectively, evaluation of feather morphology quality quantization based on stereo vision, automatic analysis of endoscopic OCT airway wall multilayer structure based on the data flow, and Doppler OCT of the cerebral arterial blood based on the characteristics of quantitative analysis of 3D data analysis applications have been done. Through exploration. According to the actual needs of these applications to the analysis and design method, put forward some specific quantitative analysis method and the automatic system, have achieved good results. In conclusion, based on data analysis of optical 3D imaging were studied comprehensively, the main innovation is reflected in some unique a new method of 3D image analysis and quantitative data automatic adaptive P, improve the efficiency of 3D image data, to a certain degree The urgent need for the automatic quantitative analysis of the optical three-dimensional imaging data is solved.

【学位授予单位】：南京大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41
，

本文编号：1376650