电子学报 ›› 2016, Vol. 44 ›› Issue (12): 3044-3052.DOI: 10.3969/j.issn.0372-2112.2016.12.033

• 综述评论 • 上一篇    下一篇

单目图像序列光流三维重建技术研究综述

张聪炫1,2, 陈震1,2, 黎明1   

  1. 1. 南昌航空大学无损检测技术教育部重点实验室, 江西南昌 330063;
    2. 南昌航空大学测试与光电工程学院, 江西南昌 330063
  • 收稿日期:2015-11-24 修回日期:2016-01-19 出版日期:2016-12-25 发布日期:2016-12-25
  • 通讯作者: 陈震
  • 作者简介:张聪炫,男,1984年7月出生于河南省焦作市.分别于2007年和2014年在南昌航空大学和南京航空航天大学获得学士和博士学位.现为南昌航空大学讲师,硕士生导师.主要研究方向为图像检测与智能识别.E-mail:zcxdsg@163.com;黎明,男,1965年2月出生于江西省樟树市.分别于1985年、1990年和1997年在上海交通大学和南京航空航天大学获得学士、硕士和博士学位.现为南昌航空大学教授,博士生导师.主要研究方向为图像处理与模式识别、智能计算.E-mail:limingniat@hotmail.com
  • 基金资助:

    国家自然科学基金(No.61401190,No.U1233125,No.61462062);江西省主要学科学术带头人培养计划项目(No.201208421);江西省自然科学基金重点项目(No.20133ACB20004);江西省科技落地计划项目(No.201408083);航空科学基金(No.2015ZC56009);江西省优势学科团队计划项目(No.20152BCB24004);无损检测技术教育部重点实验室开放基金(No.ZD201529001)

Review of the 3D Reconstruction Technology Based on Optical Flow of Monocular Image Sequence

ZHANG Cong-xuan1,2, CHEN Zhen1,2, LI Ming1   

  1. 1. Key Laboratory of Nondestructive Testing, Ministry of Education, Nanchang Hangkong University, Nanchang, Jiangxi 330063, China;
    2. School of Measuring and Optical Engineering, Nanchang Hangkong University, Nanchang, Jiangxi 330063, China
  • Received:2015-11-24 Revised:2016-01-19 Online:2016-12-25 Published:2016-12-25

摘要:

由单目图像序列光流重建物体或场景的三维运动与结构是计算机视觉、图像处理与模式识别等领域的重要研究内容,在机器人视觉、无人机导航、车辆辅助驾驶以及医学影像分析等方面具有重要的应用.本文首先从精度与鲁棒性等方面对单目图像序列光流计算及三维重建技术近年来取得的进展进行综述与分析.然后采用Middlebury测试图像序列对HS、LDOF、CLG-TV、SOF、AOFSCNN和Classic+NL等典型光流算法以及Adiv、RMROF、Sekkati和DMDPOF等基于光流的间接与直接重建方法进行实验对比分析,指出各对比方法的优点与不足,归纳各类方法的性能特点与适用范围.最后对利用分数阶微分模型、非局部约束、立体视觉以及深度线索解决亮度突变、非刚性运动、运动遮挡与模糊情况下光流计算及重建模型的局限性与鲁棒性问题进行总结与展望.

关键词: 单目图像序列, 光流, 运动估计, 三维重建, 困难场景, 鲁棒性

Abstract:

The research on 3D motion estimation and structure reconstruction of the object or scene based on the monocular image sequence optical flow is important in computer vision,image processing and pattern recognition,and the research achievements are applied in many fields.such as robot vision,unmanned aerial vehicle(UAV) navigation,driver assistance system,medical image analysis,and so on.Firstly,the research progress of the technologies of the monocular image sequence optical flow estimation and 3D reconstruction is reviewed and analyzed from the aspects of accuracy and robustness.Secondly,the test image sequences of Middlebury database are employed to make comparison among the optical flow methods of HS,LDOF,CLG-TV,SOF,AOFSCNN and Classic+NL,and the direct and indirect reconstruction methods of Adiv,RMROF,Sekkati and DMDPOF based on optical flow are compared.Through the comparison results,the advantages and disadvantages of the methods are pointed out,and the characteristics and applications of the methods are generalized.Finally,the limitations and robustness of the models of optical flow estimation and 3D reconstruction in brightness changing,non-rigid motion,motion occlusion and blur are summarized,and the solutions by the fractional order differential model,non-local constraint,stereo vision and depth cue are presented.

Key words: monocular image sequence, optical flow, motion estimation, 3D reconstruction, hard scene, robustness

中图分类号: