【作者】 罗小宾；

【导师】 殷国富；

【作者基本信息】 四川大学 ， 机械制造及其自动化， 2004， 博士

【摘要】 随着全球经济一体化，市场竞争日益激烈，迫使企业要快速响应市场的变化和客户需求。在机械产品开发过程中，采用先进的产品设计方法和手段快速响应市场需求是企业求生存与致胜的关键因素之一。反求设计是面向21世纪新产品开发的一种先进制造模式，已成为利用现有产品进行改型、创新及快速设计的一条有效途径，它涉及几何测量、图像处理、机器视觉和CAD／CAM等技术领域，对反求设计基本原理与关键技术进行研究具有重要应用价值。 如何将实物零件／几何模型经过图像数据获取和处理，实现三维模型重构、交互设计和创新设计，以便于快速成型投放市场是反求设计研究的一个关键技术问题。资料统计表明，视觉信息来源占人类获取信息的70％以上，在通过获取产品数字化信息而进行产品反求设计过程中，图像信息的获取和处理占十分重要的地位，对基于图像处理的反求设计方法研究具有重要的理论意义和应用价值。目前，基于视觉的机械产品三维重建是困扰广大研究者的一道难题，属于本学科领域研究的一个前沿问题。本论文作者在综述反求设计研究现状和发展趋势的基础上，以四川省应用基础研究课题“机械产品三维快速检测与反求设计方法的研究”为背景，重点进行基于图像处理的集成反求设计的关键技术与实施方法研究，建构基于图像处理的集成反求设计系统的体系结构。本论文在理论探讨和应用研究的主要内容和特色如下： (1) 通过快速响应工程与实现快速设计方法研究概况和关键技术的综述分析，阐述了反求设计的工作原理和应用前景。较深入地研究了基于图像处理的集成反求设计的数字化建模技术，构建了一种基于特征识别的机械产品动态约束的信息模型，提出了基于知识的反求设计体系结构的体系模型。采用UML建模语言把基于图像处理的集成反求设计的各个子系统的功能集成在一起。 (2) 针对应用计算机视觉原理实现三维实物数字化的需求，探讨了计算机视觉成像方法，分析论述了参考图像中对应关系的建立、图像深度信息处理及光照模型等问题。对于图像数据获取中存在噪声问题，提出了一种新的自适应中值滤波算法，把多个方向的子窗口同时作用于该象素，比较这些子窗口的灰度一致性，然后选择灰度一致性较高的子窗口的中值，作为处理后的该象素四川大学博士学位论文的灰度值。以一种标准中值滤波算法与本自适应算法进行测试图像和实际图像对比实验，本算法在去噪效果和细节保护特性方面均优于标准中值滤波方法。 (3)在基于图像处理集成反求设计模型的基础上，提出了一种基于ICT图像的局部曲面重建的实体建模方法，分析了基于光线追踪的体绘制算法的原理和影响显示效率的因素，提出了改善机械产品体绘制效率的方法。开发了集成反求设计的SCU一IRE实验系统，并进行应用验证。 (4)根据机械产品分类识别的需要，将模糊系统与神经网络算法相结合，提出了一种改进的模糊神经网络(FNN)新算法，采用了一种新颖的鲁棒最小二乘法RLS与BP算法相结合的RLSBP学习算法，使用模糊数学方法将一组机械零件已有的知识转变为规则，按本文所提出的算法对这些机械零件自动识别，有较好的识别效果。 (5)针对DavidM时的视觉计算理论的缺陷，较深入研究了主动视觉方法，采用基于注意机制算法，构建了一种以任务驱动，由上到下的模式，采用Mallat多分辨率边缘检测方法，进行显著特征影射的模型系统的初步探讨。模拟了生物视网膜对视觉信息非均匀采样及生物视觉具有选择注意能力的特点，采用改进的采样模式处理感兴趣的区域扭oI)。在应用中，结合小波等算法对图像数据处理，利用目标的多尺度特征，合理分配计算资源，达到实时高效对物体快速检测和识别的目的。构建了基于视觉的多DSP在线检测系统的框架结构，并探讨了其实施的关键技术。设计了使用该算法的图像采集卡，再使用注意机制算法在工厂实验中进行验证。关键词:机械产品，反求设计，数字化，重构，特征识别，图像处理，模糊神 经网络，注意机制更多还原

【Abstract】 With the global economic integration and fierce competition in market, the enterprise must rapidly respond to the change of market and requirements of customers. In the development of mechanical products, advanced product design method is one of key factors for survival and success of enterprise. Reverse design is an advanced manufacturing model in 21st century, and has become an effective approach for improvement, innovation and rapid design for products in existence. Its relevant domain involves geometry measurement, image processing, machine vision and CAD/CAM, etc., and has important application value for research on basic principle and key technologies of reverse design.How to realize reconstruction of 3D solid objects /geometry model, mutual design and creative design for rapid prototype by acquisition and processing of image data for solid objects and geometry model is one of key problem to reverse design, and research on reverse design based on image processing has important meaning. Now, 3-D model reconstruction of mechanical product based on machine vision belongs to a hot issue. In the dissertation, based on summarization of state-of-the-art and tendency of reverse engineering and on the background of project of Sichuan province, the key technology and implementing method of integrated reverse engineering based on image processing is studied in depth, and its architecture is established. The main characteristics and creative ideas of the dissertation are as follows:(1) Based on summarization and analysis of rapid responding engineering, rapid design and their key technologies, the working theory and applied foreground of reverse design is put forward. The digitization modeling technology of integrated reverse design based on image processing is studied. A Whole Dynamic Restriction information model of mechanical product is established based on feature recognition. The reverse design system based on knowledge is built. The function of each sub-system is integrated by applying UML modeling language.(2) To facing the need of three-dimension solid digitization by computer vision, computer imaging processing methods are discussed. The problems of image match, image depth and illumination model are discussed. To deal with image noise problem, a new adaptive median filter algorithm is presented. The most important feature is that the shape of the filter window may be changed according to the mean square error of the child window. A standard testing image and an experimental result are presented. The test results show that this algorithm is superior to standard median filter in preserving the details and removing the noise.(3) A solid modeling method of reconstructing partial surface based on ICT image is put forward. It is on the basis of the integrated reverse design model of handling pictures. Moreover, after analyzing the theory of IBMR based on tracing rays and the factors of affecting the displaying efficiency, the method of improving the efficiency of IBMR for mechanical product is brought forward. The experiment system of SCU-IRE software is developed, and it is validated in application.(4) According to the need of classified recognizing mechanical products, combining fuzzy system with ANN, an improved FNN algorithm is present forward. A novel robust least square algorithm with back-propagation algorithm is put forward. And the existing knowledge of a group of mechanical parts is changed to rules with fuzzy math. In test, the algorithm in this paper has auto-recognised mechanical parts in efficiency.(5) For dealing with the flaws of David Marr’s vision theory, the active vision method is researched. With the algorithm based on attention mechanism, a task-drived and "Up-Down" model is built. The Mallat wavelet multisolution method for edge detection is adopted. The model system of the saliency feature reflection is preliminarily discussed. Biology retina non-uniformly sampling and selecting to notify vision information is simulated. Range of interest (ROI) is solved by the improved sampling model. In pract更多还原