用于结构件模型重建的特征参数提取技术

Technique of Feature Extraction for Structural Part Model Reconstruction

提出了一种基于测量数据的特征参数提取新方法。首先将特征的边界曲面分解为相互邻接的若干基准面,运用极大似然估计法单独对每个分块数据集进行曲面拟合;然后利用曲面间与坐标无关的关系将拟合曲面与特征基准面匹配,并将拟合曲面用于特征参数的初步计算;最后引入马氏距离来量化重构特征与测量数据之间的误差,对特征进行仿射变换使误差最小,得到优化的特征参数。

Feature extraction is one of key techniques in structural part model reconstruction. A novel algorithm of feature parameter extraction was presented from measurement data. Firstly,boundary surfaces of feature were decomposed into several joint primitives, and each data patch was solely fitted by surface incorporating maximum likelihood estimation. Afterwards, fitted surface was matched with feature primitive utilizing coordinate--free relations among surfaces, thus preliminary feature parameters can be attained. Finally, the Mahalanobis distance was introduced to quantitate the errors from the reconstructed feature to measurement data, and feature parameters were refined by imposing appropriate similarity transformation minimizing the errors on the feature.