基于自适应移动最小二乘曲面多视多传感器测量数据对齐方法

To resolve the accurate registration issue of multiview or multisensor measured data for rapid quality inspection or reverse engineering for complex shape products(e.g. aircraft turbine blade), an efficient registration approach based on adaptive moving least-square surface is presented , in which 1) an integral invariant based principal curvatures estimation algorithm is proposed by selecting adaptive scales reflecting the shape variation of sensing data points, 2) a smooth and accurate Moving Least-Squares (MLS) surface is reconstructed from measured points by defining a Gaussian kernel with a radius being consistent with scale and heterogeneous being consistent principle curvatures; 3) a simulation based optimization algorithm is presented to align a point-cloud to the MLS surface in a global optimal sense; Finally, registered point-clouds are used to estimate the MLS surface to extend its bound and improve the representation accuracy. Then the MLS surface can be employed to iteratively register other point-clouds or registered point-clouds to obtain a more accurate registration for multi-view or multi-sensor data points, which is very beneficial to reduce the total registration errors, or avoid the propagation of registration errors during the following processing. Such approach not only can be directly applied in quality inspection of complex shape parts, and is very useful to solve the data registration problems in the complex scene reconstruction, or computer vision, which is of great significance in theory research and engineering applications.

针对复杂工件（如飞机涡轮叶片等）外形快速测量评价过程中的多视多传感器测量数据精确对齐问题，提出基于自适应移动最小二乘曲面测量数据对齐方法，通过对测量数据基于积分不变性的多尺度分析，结合单元分割及加权逼近方法自动选择合适尺度，以获取主曲率特征稳定准确估计；利用主曲率特征及相应尺度定义具有各向异性的高斯核函数及核半径，以此构建自适应移动最小二乘曲面，使其既能反映工件整体丰富多变外形，且能准确重建局部各向异性形状变化特征；同时，将测量数据到移动最小二乘曲面对齐转化为仿真优化问题，通过细分矩形优化算法自动给出合理初值，克服现有对齐方法受初值影响的局限性，实现全局优化对齐；以此为基础，通过迭代实现多视多传感器测量数据精确对齐，从而提高工件外形逼近及评价准确性。该方法不仅可用于复杂外形零件加工质量检测、维修，且有助于解决反求工程、计算机视觉等应用中的测量数据对齐问题，具有重要理论研究及工程应用价值。

针对复杂工件（如飞机涡轮叶片等）外形快速测量评价过程中的多视多传感器测量数据精确对齐问题，项目组提出基于自适应移动最小二乘曲面测量数据对齐方法，通过对测量数据基于积分不变性的迭代多尺度分析，结合单元分割及加权逼近方法自动选择合适尺度，获取了复杂曲面轮廓主曲率特征稳定准确估计；利用稳定估计的主曲率特征及与之相适应尺度，定义具有各向异性高斯核函数及核半径，以此构建自适应移动最小二乘曲面，使其既能反映工件整体丰富多变外形轮廓，且能准确表征局部各向异性形状变化特征；依据自适应移动最小二乘曲面各向异性特征定义，推导给出了任意点到该曲面的最近点计算公式及其算法，以计算测量数据到该曲面的精确ICP对齐矩阵；同时，将测量数据到移动最小二乘曲面对齐转化为仿真优化问题，通过超拉丁立方采样，稀疏响应面方法，以及GPU并行快速响应面寻优算法，自动给出合理初值，克服现有对齐方法受初值影响的局限性，实现测量数据到自适应移动最小二乘曲面的全局优化对齐；以此为基础，通过迭代实现多视多传感器测量数据精确对齐，从而提高工件外形逼近及评价准确性。该方法不仅可用于复杂外形零件加工质量检测、维修，且有助于解决反求工程、计算机视觉等应用中的测量数据对齐问题，具有重要理论研究及工程应用价值。