基于SIFT特征的低区分度点云数据匹配  

Low Contrast Point Cloud Data Registration Based on SIFT Feature

作　　者：杨红娟[1] 陈继文[2] 张运楚[1] 张君捧[1] 郝丽丽[1] 

机构地区：[1]山东建筑大学信息与电气工程学院,济南250101 [2]山东建筑大学机电工程学院,济南250101

出　　处：《计算机辅助设计与图形学学报》2016年第3期498-504,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61303087);山东省高等学校科技计划项目(J12LN32;J14LB04);山东省绿色建筑协同创新中心团队建设科研基金(LSXT201513)

摘　　要：针对复杂产品测量视角变化较大、存在测量误差的低区分度点云数据匹配问题,从稳健的特征点识别方法入手,提出基于SIFT特征的低区分度点云数据匹配方法.该方法将反映产品外形及其空间关系的点云数据x表示为受空间位置变化的影响较小的低频部分,以及受到物体自身特性影响较大,随空间位置变化较大的高频部分.设计高斯同态滤波器,在频率域中降低低频分量,增强高频分量,提高点云数据的区分度;然后提取点云数据的SIFT特征向量,以欧氏距离作为相似性度量标准进行SIFT特征向量的匹配,获得点云数据的匹配点对;最后采用四元数法估计点云数据匹配参数,计算旋转矩阵和平移矩阵,实现低区分度点云数据匹配.通过某汽车用刹车壳体零件的点云数据匹配实例,验证了文中方法的有效性.Aiming at registration of low contrast point cloud data, which is measured with multi view that changes largely and is accompanied with noise from complex product, a data registration method based on SIFT feature is proposed, which focus on robust feature point identification. Point cloud datax, which reflects the product shape and relationship, is divided into the low frequency part and the high frequency part. Spatial location change has little effect to the low frequency part, while the high frequency part changes greatly with spatial location and product geometric shape characteristics. Gauss homomorphic filter is designed to reduce the low frequency part and increase the high frequency part, improving the contrast of point cloud data. SIFT feature vector of point cloud data is extracted. SIFT feature vector matching is realized with Euclidean distance as the similarity measure to achieve matching points of point cloud data. Quaternion is used to estimate the parameters of point cloud data registration and calculate rotation matrix and translation matrix, achieving low contrast point cloud data regi-stration. Registration example of point cloud data from a brake shell shows the validity of the proposed methods.

关 键 词：同态滤波 SIFT特征 四元数法 点云数据匹配 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]