基于二次曲面拟合的机械模型交互式分割方法  

Interactive Segmentation of Scanned Mechanical Models Based on Quadratic Surfaces Fitting

作　　者：吴仲琦 郭建伟[2] 肖俊 曾祥永 王颖 严冬明[2] Wu Zhongqi;Guo Jianwei;Xiao Jun;Zeng Xiangyong;Wang Ying;Yan Dongming(School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing 100049;National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences, Beijing 100190;Beijing Shengkai Interactive, Beijing 100191)

机构地区：[1]中国科学院大学人工智能技术学院,北京100049 [2]中国科学院自动化研究所模式识别国家重点实验室,北京100190 [3]北京盛开互动科技有限公司,北京100191

出　　处：《计算机辅助设计与图形学学报》2019年第7期1210-1220,共11页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61471338,61761003,61502490,61501464,61772523,61802406);北京市自然科学基金(4184102);北京市科技计划课题(Z181100003818019);中国科学院青年促进会(2015361);中国科学院前沿科学重点研究项目(QYZDY-SSW-SYS004);北京市科技新星计划(Z171100001117048)

摘　　要：针对三角网格全自动分割方法的自适应能力较差的问题,提出了一种基于二次曲面拟合的交互式半自动分割方法.首先采用变分方法对网格模型进行二次曲面拟合和分割,然后引入用户的交互信息来指导模型的分割,并配合迭代优化算法交替使用.具体地,算法包括以下3个步骤:基于random sample consensus(RANSAC)的初始化、迭代优化、交互式合并与切分.其中,初始化方法可以取得一个较好的划分结果,迭代优化可以提高算法的分割质量,而交互式合并与交互式切分步骤可以与迭代优化步骤交替使用,利用用户的主观先验知识对分割不够准确的地方进行局部操作,从而使得分割的效果更为准确、真实.通过比较不同算法对多个机械模型进行分割的效果,算法在收敛速度和拟合误差方面都有明显的优势,并且可以对具有不同特征的三维机械模型以及带有噪声的模型进行准确、合理的分割.Aiming at the problem of poor adaptive ability of the triangular mesh automatic segmentation method, an interactive semi-automatic segmentation method based on quadric surface fitting is proposed. Firstly, based on the framework of variational method, quadratic surface fitting and segmentation on the input mesh are processed. And then user interaction is introduced to guide the segmentation process, the interactive operations can be used interchangeably with iterative optimization algorithm. Specifically, the algorithm includes the following three steps: initialization based on random sample consensus (RANSAC), iterative optimization, interactive merging and segmentation. Among them, the initialization method can obtain a good segmentation result. Then an iterative fitting operation is adopted to optimize the segmentation result. Finally, the interactive operations can be used interchangeably with iterative optimization steps, meanwhile the user’s subjective prior knowledge is used to perform local operation on the part where the segmentation is not accurate enough, so that the segmentation effect is more accurate and authentic. The experimental examples show that compared with other methods, our approach has obvious advantages in convergence speed and fitting error. Further, we can accurately segment the mechanical models with complex structure and noise.

关 键 词：网格分割 用户交互 二次曲面拟合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]