航空叶片多光学传感器多尺度测量点云高效拼合方法

项目名称： 航空叶片多光学传感器多尺度测量点云高效拼合方法

项目编号： No.51505134

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 机械、仪表工业

项目作者： 何万涛

作者单位： 黑龙江科技大学

项目金额： 19万元

中文摘要： 航空叶片多光学传感器多尺度测量点云具有点集规模大、重叠区域小、不同分辨率、不同精度等级，给点云拼合效率、稳定性、精度带来一系列问题。针对上述问题，本项目探索基于变分隐函数、微分流形、非线性优化的大规模点集表示与操作方法，提出自动、高效、稳定的多传感器数据拼合方法。具体内容包括：建立适合GPU并行计算的大规模细粒度点集数据结构，研究快速最近点搜索方法；根据多传感器数据特点，定义新的变分隐函数表示的局部曲面特征描述算子，利用变分隐函数良好的内外插性质，研究小重叠区域特征快速匹配方法，提出匹配关系正确与否的评价方法，实现快速稳定的自动预拼合；应用对偶四元数表示刚体变换，建立刚体变换SE(3)流形上的测地距离最小二乘精拼合优化模型，推导同时迭代求解旋转和平移变换参数的线性方法。应用上述理论方法，开发点云拼合软件模块，为航空叶片的设计-加工-测量大闭环制造、全测全检与创新设计提供必要的技术保障。

中文关键词： 多光学传感器；多尺度；预拼合；全局优化；并行计算

英文摘要： Multi-scale point clouds which measured by multiple optical sensors have characteristics include large-scale, small overlapping regions, different resolutions, different levels of precision and so on. These characteristics bring series of problems such as calculation efficiency, feature points matching and global optimization for data registration. To solve the above problems, this project explores the large-scale point clouds representation and operation methods based on variational implicit function, differentiable manifold and non-linear optimization to present an automatic, efficient and stable multi-sensor data registration method. Point set structure of massive and fine-grained is established for parallel computing on GPU to search the nearest point . A new description operator of local surface feature is defined by variational implicit function. To achieve a stable automatic pre-registration, the good nature of variational implicit function on interpolation and extrapolation is used to study the feature matching methods and evaluation mechanisms . Rigid transformation is represented by dual quaternion to establish geodesic distance least squares optimization model on SE(3) manifold, while iterative linear method is present to solve rotation and translation transformation parameters. Finally, a registration software module will be developed using the above theories to provide the necessary technical for the aero-engine blades closed loop manufacture, full size measurement and innovative design.

英文关键词： multiple optical sensors ;multi-scale;pre-registration;global optimization;parallel computing