电子学报 ›› 2021, Vol. 49 ›› Issue (5): 936-943.DOI: 10.12263/DZXB.20170915

• 学术论文 • 上一篇    下一篇

基于稠密变形场能量优化的非刚性在线体积重建方法

林金花1, 马林2   

  1. 1. 长春工业大学应用技术学院, 吉林长春 130012;
    2. 长春一汽铸造有限公司, 吉林长春 130000
  • 收稿日期:2017-08-08 修回日期:2020-10-10 出版日期:2021-05-25
    • 作者简介:
    • 林金花 女,1980年9月出生,吉林长春人.长春工业大学讲师.主要研究方向为深度学习算法和金相图像处理. E-mail:ljh3832@163.com;马林 男,1979年8月出生,吉林长春人.一汽铸造有限公司高级工程师.主要研究方向为金相图像处理. E-mail:927168801@qq.com
    • 基金资助:
    • 国家自然科学基金 (No.62006027); 吉林省教育厅十三五科学技术研究 (No.JJKH20200681KJ)

An Non-Rigid Online Volume Reconstruction Method Based on Energy Optimization of Dense Deformation Field

LIN Jin-hua1, MA Lin2   

  1. 1. School of Application Technology, Changchun University of Technology, Changchun, Jilin 130012, China;
    2. FAW Foundry Co., Ltd, Changchun, Jilin 130000, China
  • Received:2017-08-08 Revised:2020-10-10 Online:2021-05-25 Published:2021-05-25
    • Supported by:
    • National Natural Science Foundation of China (No.62006027); Science and Technology Research Project of Department of Education of Jilin Province during the 13th Five-year Plan (No.JJKH20200681KJ)

摘要: 针对动态非刚性在线重建存在的重建漂移问题,提出了一种基于单个RGB-D传感器的稠密变形场在线融合策略,实现了对非刚性几何形状的动态重建.通过局部平滑和输入约束处理,将空间的最优变形转换为非线性正则变分优化问题,并使用数据并行闪存优化策略,以摄像机捕获速率,实现对非刚性场景的在线跟踪重建.通过实验表明,本文方法实现了对非刚性场景的鲁棒跟踪,减少了在线重建过程中的漂移现象,适用于快速运动和缺乏几何特征的场景重建.

关键词: RGB-D传感器, 动态重建, 鲁棒跟踪

Abstract: In order to relieve the drift phenomenon in dynamic non-rigid online reconstruction, an on-line fusion strategy for dense deformation field based on a single RGB-D sensor is proposed, realizing the dynamic reconstruction for non-rigid geometries. By local smoothing and input constraint strategy, the optimal deformation problem is transformed into nonlinear regular variational optimization problem. We use the data parallel flash optimization strategy to achieve on-line tracking of non-rigid scene in camera tracking rate. Experiments show that the proposed method achieves robust tracking of non-rigid scenes, which reduces drift in the process of online reconstruction, this algorithm is suitable for fast moving scene as well as the object lack of geometric features.

Key words: RGB-D sensor, dynamic reconstruction, robust tracking

中图分类号: