曲面边界样点逆向均值漂移识别  

Reverse mean shift detection algorithm for boundary points of surface

作　　者：李延瑞[1] 孙殿柱[2] 张英杰[1] 白银来 

机构地区：[1]西安交通大学机械工程学院,陕西西安710049 [2]山东理工大学机械工程学院,山东淄博255049

出　　处：《计算机集成制造系统》2015年第7期1719-1724,共6页Computer Integrated Manufacturing Systems

基　　金：国家自然科学基金资助项目(51075247)~~

摘　　要：针对现有的曲面边界样点识别算法难以适应非均匀分布的实物表面采样数据的问题,将目标样点的k-近邻点集作为曲面局部样本,基于均值漂移算法使得曲面局部样本在一定程度上向目标样点邻近的采样数据稀疏区域扩展,实现对曲面局部样本的增益优化,并对增益优化后的曲面局部样本进行核密度估计,获取目标样点对应的模式点,并通过比较目标样点与其对应模式点的偏离程度进行边界样点判定。实验表明,该算法可快速准确地识别曲面裁剪边界、几何连续的相邻面片公共边界以及曲率变化较大的过渡曲面上的特征样点,并且对非均匀分布的采样数据具有良好的适应性。For solving the problem that current surface boundary points detection algorithms were difficult to adapt non-uniform distributed sampled data of physical surface, a boundary detection algorithm based on reverse mean shift was proposed. Based on mean shift algorithm, the surface local sample which used by k-nearest neighbors of objective point was extended to the sampled data sparse region of adjacent objective point, and the gain optimization for the surface local sample was realized. The kernel density estimation was applied for gain optimized sample to ob- tain the corresponding mode point of objective point. The boundary points were detected by comparing the deviation extent between the objective point and its mode point. The experimental results showed that the proposed algorithm could detect the characteristic points of surface trim boundary, public boundary of geometric continuous adjacent surfaces and transitional curved surface with great curvature change, and had good adaptability for the sample data of non-uniform distribution.

关 键 词：实物表面采样数据 曲面边界样点识别 均值漂移 核密度估计 动态空间索引 

分 类 号：TP391.72[自动化与计算机技术—计算机应用技术] TP301.6[自动化与计算机技术—计算机科学与技术]