基于最优几何匹配的时间连贯3D动画重建  

Time coherent 3D animation reconstruction based on optimal geometric matching

作　　者：王建华[1] 冉煜琨 WANG Jianhua;RAN Yukun(School of Engineering and Technology,Chengdu University of Technology,Leshan 614000,China)

机构地区：[1]成都理工大学工程技术学院,四川乐山614000

出　　处：《电子设计工程》2021年第20期36-42,共7页Electronic Design Engineering

基　　金：四川省重点实验室开放基金重点项目(scsxdz2019zd01)。

摘　　要：为了优化RGB-D数据的3D重建,提出一种时间连贯的3D重建方法。在数据采集后,从映射到深度数据的RGB数据中提取光学特征点,以得到两个帧之间的初始稀疏3D对应关系。利用最优几何匹配程序进行特征点细化,使用这些特征点对连续帧中的两个3D点云进行独立于帧分辨率的匹配;利用运动向量对齐方法重建时间连贯的3D动画。实验结果表明,所提方法可以重建时间连贯的3D动画。与其他方法相比,所提方法数据丢失较少,在平均误差方面具有一定优势,而且在计算方面也具有高效性。To optimize the 3D reconstruction of RGB-D data,a time coherent 3D reconstruction method is proposed.After data acquisition,optical feature points are extracted from RGB data mapped to depth data to obtain the initial sparse 3D correspondence between two frames.The optimal geometric matching program is used to refine feature points,and these feature points are used to match two 3D point clouds in consecutive frames independently of frame resolution.The motion vector alignment method is used to rebuild a 3D animation.The experimental results show that the proposed method can reconstruct 3D animation with temporal coherence.Compared with other methods,the data loss is less,the average error has some advantages,and the calculation is also efficient.

关 键 词：RGB-D数据 3D重建 特征点 最优几何匹配 运动向量对齐 平均误差 

分 类 号：TN219[电子电信—物理电子学] TP391[自动化与计算机技术—计算机应用技术]