一种新的散乱点云特征提取方法研究  

Study on a New Method of Feature Extraction from Scattered Point Clouds

作　　者：张靖[1] 周明全[1,2] 张雨禾[1] 耿国华[1] 李姗姗[1] 

机构地区：[1]西北大学信息与科学技术学院,西安710127 [2]北京师范大学信息科学与技术学院,北京100875

出　　处：《小型微型计算机系统》2017年第7期1601-1607,共7页Journal of Chinese Computer Systems

基　　金：国家自然科学基金面上项目(61373117)资助;国家自然基金项目(61305032)资助;高等学校博士学科点专项科研基金项目(20136101110019)资助

摘　　要：结合显著特征点与马尔科夫随机场的特性,提出一种新的散乱点云特征提取算法研究.首先构造点云的显著度函数,计算点云模型中散乱点的显著度;然后根据点的显著度及点间测地距构造Reeb图,由Reeb图及判断Reeb图显著点的两个条件提取显著特征点;接着根据点到显著特征点及中心点的距离计算马尔科夫随机场的联合密度函数,通过贝叶斯估计求最大后验概率分布函数,采用马尔科夫最大后验概率框架归约简化得到最优标号的目标函数;最后采用图割法求解目标函数,依赖全局能量变化调整标号集,得到点云的最优标号集,并根据最优标号集与点集间对应关系提取特征点.实验结果表明:该算法能够有效的检测出散乱点云的特征点,改善了算法的时间性能;其中,算法根据随机场的全局能量变化调整标号,得出最优标号集,避免了传统方法人工调参的问题.Based on the characteristics of the salient feature points and the Markov random field, an algorithm of basing on the improved Markov random field model of feature points extraction is proposed. Firstly, the prominence function of scattered points is constructed, and the prominence of the point clouds is calculated. Then, the Reeb map is constructed according to the prominence of points and the Geodesic distance between points, and extract salient feature points based on Reeb graph and two rules. Third, according to the distance of point to the salient points and center point calculate joint density function of the Markov Random Field, through Bayesian estimation to calculate maximum posterior probability distribution function, and the optimal label objective function by simplifying andreductting the maximum posterior probability framework. Finally, a graph cut method is usedto solve the objective function, and adjust label according to the change of global energy. The optimal label set of points is obtained, and the feature points are extracted according to the relationship between the optimal label set and the point set. The experimental results indicate that the method proposed in this paper can effectively detect feature points of the scattered point clouds and improve the efficiency. According to the change of airport＇s global energy to adjust the label and optimal label set is obtained finally. It avoids the traditional method of manual adjusting parameters.

关 键 词：散乱点云 显著度 特征点 马尔科夫随机场 标号集 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]