基于多视图约束的结构光扫描系统冗余点去除算法  

An Algorithm for Removing Redundancy Points in Structured Light Scanning System Based on Multi-view Constraints

作　　者：袁建英[1,2] 郭进[3] 李柏林[1] 王荔芳[2] 刘甲甲[1] 

机构地区：[1]西南交通大学机械工程学院,成都610031 [2]昆明学院自动控制与机械工程学院,昆明650214 [3]西南财经大学天府学院,绵阳621000

出　　处：《计算机辅助设计与图形学学报》2013年第12期1903-1909,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家科技支撑计划课题(2012BAF10B08);云南省教育厅科学研究基金(2013Y257);四川省科技支撑计划(2012GZ0102)

摘　　要：针对结构光多视扫描拼接后在扫描重叠区域内重复计算带来的数据冗余问题,提出一种重复区域内冗余点删除算法.首先根据多视图几何约束原理,将不同视下相机外参数绝对定向.然后计算当前视下三维点并将其反投影至上次扫描图像,若反投影像点在上次扫描中曾参与物点计算,且若2次物点距离大于给定阈值,则保留当前视下三维点,反之保留反投影残差较小者;否则,保留该点.理论分析和实验结果表明,对于重叠扫描区域,该算法能达到和单次扫描相同密度的点云,且去除了扫描过程中精度较低的点,同时保留了曲率较大处表面的细节信息.A novel algorithm has been proposed to remove the redundancy points in the overlapped regions when structured light scanning system works at different angles. All the external parameters in different views are oriented absolutely to a global coordinate system, and the 3D points in current view are reprojected to the last scanned image. If the reprojected image points are triangularized in last scan process, and the distance between the points between current scan and the last scan is larger than a given threshold value, the 3D points of current scan are preserved. Otherwise, the points with small reprojected residual are preserved. If the reprojected image points are not triangularized in last scan, 3D points in current view are preserved. Both theoretical analysis and experimental results demonstrate that the new algorithm has the benefit of removing the lower accuracy points and preserving more detailed information of the surface with large curvature, with the total data density remaining the same as single view scanning.

关 键 词：结构光扫描 多视拼接 冗余点删除 多视图几何 图像重复区域 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]