Optimisation in Quadric Primitive Representation and Classification from Range Data  

作　　者：ZHANGLei WANGGuoyu HOUKESZweitze 

机构地区：[1]DepartmentofComputerScience,NorthwestUniversity,Xi＇an710069,China [2].DepartmentofElectronicEngineering,OceanUniversityofChina,Qingdao266003,China [3]DepartmentofElectricalEngineering,UniversityofTwente,7500AEEnschede,TheNetherlands

出　　处：《Chinese Journal of Electronics》2004年第4期676-681,共6页电子学报（英文版）

摘　　要：This paper focuses the modelling of uncertainties occurring in feature extraction from range images for surface-based primitive classification. The feature descriptor of a surface primitive consists of the algebraic invariants extracted from a quadric patch, which are concerned in applications of man-made object recognition. The uncertainties of the estimates are cbaracterised with their covariance matrices. We propose an explicit probabilistic model to describe the statistical performance of the feature descriptor. Consequently, the classification of surface primitive and sensor-based modelling process can be implemented within an optimal framework. Experimental results of primitive classification with synthetic and real range data are presented.

关 键 词：不变量提取 二次配件 概率模型 最佳分类 图像处理 面貌特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]