摘要
处理点云配准中初始对齐参数的求解问题旨在快速估计两帧数据间的刚体变换矩阵,给出精确的初始对齐.通过分析刚体变换的欧氏不变特征,提出一种距离差分矩阵算法,在尽量保留正确匹配的同时快速剔除点云数据的显著错误匹配.随后采用最小二乘后向方法经少量迭代从剩余的匹配集中估计出准确的刚体变换参数.算法中的阈值参数根据点云分辨率自适应决定,消除了人为设定的不便和误差.针对多个数据集的测试表明,该算法可以快速有效地剔除误差匹配,求解出更优的初始变换,增强点云配准的精度和效率.
This paper deals with the initial aligning problem of pairwise point cloud registration. A novel Distance Disparity Matrix algorithm derived from Euclidean invariants of rigid motion is proposed to prune the obvious outlier matches while keeping the most inliers. The pruned matches are then sent into a Least- Squares Backward procedure to estimate optimized rigid transformation in fewer iterations. The employed thresholds are automatically determined with respect to the actual resolution of input point clouds. Experimental results show that the proposed strategy effectively eliminates the outliers to achieve better initial alignment at a high speed, and further enhances the performance of point cloud registration.
作者
罗楠
王泉
LUO Nan WANG Quan(School of Computer Science and Technology, Xidian Univ. , Xi＇an 710071, China)
出处
《西安电子科技大学学报》
EI
CAS
CSCD
北大核心
2017年第5期69-74,共6页
Journal of Xidian University
基金
国家自然科学基金资助项目(61572385)
陕西省科技计划资助项目(2015KTCXSF-01)
关键词
点云配准
刚体变换
距离差分矩阵
特征匹配
最小二乘后向方法
点云分辨率
point cloud registration
rigid transformation
distance disparity matrix
feature matches
least-sqaures backward search
point cloud resolution
