跨尺度曲面微细结构的多视层析显微视觉三维检测方法

Following with China becoming a strong industrial manufacturing country, the demand for accurate measurement of trans-scale curved surface becomes more and more urgent. For example, the air bearing of a gyroscope inside a major strategic weapon of our country has a sophisticated trans-scale curved surface. The etching of micro cannelures on the curved surface requires very precise machining accuracy, because the machining accuracy directly influences the impact accuracy of the missile. However, feasible measurement method has not been proposed because of three problems. First, the micro cannelures are just several hundreds of micrometers wide and several micrometers depth, but extend spirally along the curved surface for nearly a centimeter. Second, the parameters of the cannelures need to be measured very precisely. What’s more, the slope of the curved surface varies much. The lack of feasible measurement is a hidden peril in the development of the major strategic weapon in our country. This study improves the optical imaging system and researches the image processing algorithms based on computational imaging theory, and proposes a multi-field and multi-view microscopic vision method for the microstructure on trans-scale curved surface . It contains four main contents. First, a multi-field and multi-view microscopic vision method for trans-scale, high-slope, and large-longitudinal-range surface is proposed based on the computational imaging theory. Second, a model of three-dimensional measurement based on multi-view microscopic vision is built, meanwhile the calibration and motion error alignment methods of the system are proposed. Thirdly, a method of feature extraction based on prominent feature indexing on blurred region is proposed to solve the problem of poor and repetitive texture on the microscopic image. Fourthly, an optimized three-dimensional surface reconstruction method based on Bayesian clustering of focus criterion function confidence is proposed. This technique provides technical support for developing major strategic weapon of our country. It also provides a conductive method and technical approach to extending the measurement ability of a microscopic vision system.

随着我国逐步走向工业制造强国，跨尺度复杂曲面工件的精确检测越来越重要，如我国重大战略武器中陀螺仪的气浮轴承，其曲面上微细沟槽的尺寸精度直接影响导弹的命中精度。但由于沟槽跨度大、精度要求高和表面倾角大等问题，一直没有合适的检测方法，是我国重大武器型号研制的隐患。本项目基于计算成像理论，通过对光学成像系统的改进和图像处理算法的研究，提出一种用于跨尺度曲面微细结构三维尺寸检测的多视显微视觉方法。主要内容包括：①基于计算成像技术理论提出一种能跨尺度、大倾角和大轴向范围检测的多视显微视觉方法；②建立多视显微视觉三维检测模型，研究系统标定方法以及运动误差矫正方法；③提出基于模糊区域突出特征索引的特征提取方法，解决显微图像无纹理或重复纹理的问题；④提出基于聚焦评价函数置信的贝叶斯聚类优化融合三维重建算法。为我国重要战略武器的研制提供技术保障，为扩展显微视觉系统的检测能力提供可供借鉴的方法和技术途径。

本项目针对我国重大战略武器中陀螺仪的气浮轴承、智能制造中的加工刀具等跨尺度复杂曲面工件的关键精确检测需求，基于计算成像理论，通过对光学成像系统的改进和图像处理算法的研究，提出了一种用于跨尺度曲面微细结构三维尺寸检测的多视显微视觉方法。主要内容包括：①基于计算成像技术理论提出了一种大视场、大倾角、大轴向范围和恒定放大率，用于跨尺度曲面微细结构尺寸三维检测的多视显微视觉方法；②为满足跨尺度曲面微细结构的多视显微视觉三维检测的轴向层析成像要求，既要轴向层析成像，又要保证在层析过程中放大率恒定，本项目利用4f 准直系统扩展显微物镜的孔径光阑面，将液体变焦透镜放置在孔径光阑面的共轭面上，并结合载物台平移扫描和旋转反射镜多视扫描的方式，设计了一种放大率恒定的多视场多视角层析显微视觉成像方法；③建立多视显微视觉三维检测模型，研究了多视层析显微视觉系统标定方法以及运动误差矫正方法；④针对多视显微视觉图像基线宽，纹理较少或者没有纹理，并且跨尺度曲面微细结构的关键特征可能会被遮挡或半遮挡等特点，本项目提出了基于聚焦评价函数置信的背景区域去除、基于面片的多视立体三维优化匹配方法。本项目突破了跨尺度、高精度和大倾角工件的测量问题，为我国重要战略武器的研制提供技术保障，为扩展显微视觉系统的检测能力提供可供借鉴的方法和技术途径。基于本项目提出的方法研制的原理样机的测量精度为：横向测量误差1.4μm（3σ），纵向测量误差1.2μm（3σ），多视变焦三维测量精度7.2μm（3σ）。本项目研究成果已与中国工程物理研究院合作开展核武器内壳关键尺寸的三维检测。本项目培养研究生6名，发表SCI检索论文15篇，申请国家发明专利1项，参加国际国内会议十数次，与国内外专家学者进行了广泛的交流。