吴毅红 English |
|||||||||||||
|
研究员,博士生导师 Email:
yhwu at
nlpr.ia.ac.cn Tel: +86-10-82544697 Fax: +86-10-82544594 |
||||||||||||
**中国科学院 自动化研究所 模式识别国家重点实验室**
**机器人视觉课题组**
主要研究方向: ---------------------------------------------------------------------------------------------------------- 3DV 2020 Tutorial: Visual Localization in the Age of Deep Learning
ACCV 2018 Tutorial: Developments of 3 Dimensional Computer Vision Since 2017 PDF _____________________________________________________________________________________________
研 究领域 1. 多视几何理论研究,几何不变量与括号代数在视
觉中的应用 3. SLAM应 用:机器人定位与导航,AR,VR
代 表 性论文
视觉定位研究展示 1. 基于视觉的移动增强现实 2. 动
态目标SLAM 3.
场景SLAM 4. SLAM重
定位 5.
结合IMU的SLAM
6. SLAM devices
研制多款软硬协同实时在线三维重建设备,室内外皆可
个
人简历 1995年毕业于山西雁北师院获本科学位;1998年毕业于陕西师范大学获硕士学位,导师李会师教授;2001年6月
毕业于中国科学院系统科学研究所获博士学位,导师李洪波研究员;之后到模式识别国家重点实验室做博 士后,导师胡占义研究员。2003年被评为副研究员;2008年
被评为研究员。2005年,2010年,被法国IRIT实
验室邀请合作研究。2006年至 2008年,被香港城市大学多次邀请合作研究。获1项高等学校科学研究自然科学奖二等奖,排名第三。获三星电子校企合作卓越
贡献奖。首批阿里菜鸟驼峰计划特约专家。申请多项
国内外发明专利,多项被授权。 教学
2017-今, 春季 机器人定位导航 中国科学院大学 学
术活动 Area Chair of CVPR 2021 Associate
Editor of Pattern Recognition Area Chair of ICPR 2018 Area Chair of PRCV 2021 SPC
of IJCAI 2019/2020/2021 Editorial
Board Member of ACTA AUTOMATICA SINICA Editorial
Board Member of Journal of CAD & CG Editorial
Board Member of Journal of Frontiers of Computer Science and Technology Editorial
Board Member of the Open Artificial Intelligence /Computer Science
Journal Editorial
Board Member of CADDM PC
Member of VISUAL 2007 PC
Member of ICCV 2007 PC
Member of PCM 2007 PC
Member of WCICA 2008 PC
Member of CVPR 2008 Session
Chair of ACCV 2007 Session
Chair of RIUPEEEC 2005 PC
Member of KES 2009 Reviewer of ICCV 2009/ACCV 2009 CSIG 三维计算机视觉专委会副主任 CSIG 机器视觉专委会常委 特 邀报告 1. 吴 毅红,国际会议,图像科学和信息处理中的数学与计算大会,“PnP Problem Revisited”,新加坡,2003年10月. 2. 吴毅红,国内会议,第三届全国几何设计与计算 学术会议,“计算机视觉的任务及目前研究的现状”,兰州,2007年7月. 3. 吴毅红,国内会议,全国数学与信息科学研究生 学术研讨会,“计算机视觉的历史和现状”,北京,2008年12月. 4. 吴毅红,国内会议,中国自动化大会暨钱学森诞 辰一百周年,“视觉定位和三维建模”,北京,2011年11月. 5. 吴毅红,国际会议,三星研究院计算机视觉研讨 会,“Efficient Pose Tracking and Mapping”,2014年7月. 6. 吴毅红,国际会议,WCR,“Camera Localization and 3D Mapping in Large Scale Environments”, 2015年9月. 7. 吴毅红,国内会议,系统科学青年学者论坛,“大数据的视觉定位与三维重建& rdquo;, 2015年12月. 8. 吴毅红,国内会议,RACV,“基于图像的定位:发展与应用& amp; rdquo;,2016年9月. 9. 吴毅红,国内会议,Vision China,“视觉建模与定位 ”,2016年10月. 10. 吴毅红,国内会议,华为北研工程技术论坛,& ldquo;视觉建模与定位研究”,2016年12月. 11. 吴毅红,国内会议,全球人工智能大会,“三 维视觉研究及应用”,2017年5月. 12. 吴毅红,国内会议,VALSE,“2017年以来的2Dto3D& rdquo;,2018年4月22日. 主 持课题 Single view based metrology (863) A
study on the theory and algorithm of camera
self-calibration Geometric
invariant computation and camera pose
determination from n perspective points Image
based 3-dimensional reconstruction Camera
parameter computation from video sequence
(Key NSFC) Ominidirectional
camera calibration Study on image invariants and applications under multiple camera models (NSFC) Image-based modeling for complex and large scale environment (NLPR) Visual SLAM (Nokia RC in Finland) Non-planar
object tracking (Samsung) Camera
pose tracking (Samsung) VR
pose tracking (Huawei)
举
办国际会议 The First Workshop on Community Based 3D Content and Its Applications in Mobile Internet Environments,In conjunction with ACCV 2009 The Second Workshop on
Community Based 3D Content and Its Applications,In
conjunction with ICME
2012 Developments
of 3 Dimensional Computer
Vision Since 2017 Virtual
reality (VR), augmented reality (AR), robotics, and
autonomous
driving have recently
attracted much attention from the academic as well as the industrial
community.
3 dimensional (3D) computer vision plays important roles in these
fields.
Autonomous localization and navigation is necessary for a moving robot,
where
using cameras is the most flexible and low cost approach for building
map and
localization. To augment reality in images, camera pose determination
or
localization is needed. To view virtual environments, the corresponding
viewing
angle is necessary to be computed. Furthermore, cameras are ubiquitous
and
people carry mobile phones that have cameras every day. There have been
some AR
applications in mobile phones. Therefore, 3D computer vision has great
and
widespread applications. This
tutorial will provide the developments of 3D computer vision of the
past two
years. The important works since 2017 will be introduced in image
matching,
camera localization including camera pose determination and
simultaneous
localization and mapping (SLAM), 3D reconstruction. The
contents have five parts: 1.
Preface Some
fundamental knowledge of 3D vision is introduced as well as some events
related
to 3D vision in this past two years. 2.
Developments in image matching Some
important works of image feature detectors and descriptors since 2017
are
introduced. Also, some important woks of image matching and two dataset
since
2017 are introduced. Among these works, deep learning based methods are
growing. 3.
Developments in camera localization A
complete classification of image based camera localization mapped as a
tree
structure is given. The important directions are pointed out. The
developments
of camera localization in both known environments and unknown
environments are
introduced since 2017. The ones in known environments are PnP works.
The ones
in unknown environments are SLAM works. SLAM works include the general
geometric SLAM, learning SLAM, semantic SLAM, and marker SLAM. Except
SLAM
under the traditional cameras, there are some SLAM works under the
event
cameras and RGBD cameras. 4.
Developments in 3D reconstruction This
part will introduce structure from motion (SFM) based 3D
reconstruction,
learning 3D reconstruction, RGBD 3D reconstruction or RGBD SLAM since
2017. 5.
Trends of 3D vision I
will share my views for 3D vision trends. Visual
Localization in the Age of Deep Learning A
short description: Deep learning is a subset of machine learning
and one of the foundations of Artificial Intelligence.
Recent years see its remarkable development and wide applications. It
has been
used in almost every task of computer vision field and shows its
powerful ability.
3D computer vision, developed under the Marr framework, re-blooms by
deep
learning again. Visual localization, a key task in 3D computer vision,
has many
applications in AR, VR, robot navigation, and driverless car. Deep
learning has
been used in visual localization by various ways. There are end-to-end
and
non-end-to-end deep learning visual localizations. The
main topics in
visual localization include image feature extraction and description,
image
matching, image retrieval, 2D-3D matching, RANSAC, PnP, feature
tracking, loop
closure detection, SLAM, bundle adjustment. All these have combined
deep
learning. The
tutorial consists of five parts. In the first
part, this
tutorial will give an overview of visual localization, share with some
personal
viewpoints to use deep learning in 3D vision. In the second part, the
tutorial
will introduce some important literatures in image feature extraction
and
description, and matching with deep learning. Then, work of camera pose
determinations from known 3D structure knowledge with deep learning is
reported
in the third part, including the work of image retrieval, 2D-3D
matching,
RANSAC, and PnP. In the fourth part, work of SLAM with deep learning is
reported, including the work of feature tracking, loop closure
detection,
bundle adjustment etc. As we see, every aspect combines deep learning
in visual
localization. It shows free exploration to use deep learning. Finally,
i.e. in
the last part, some personal views are shared for some existing
problems when
using deep learning, for example, there are no clear training dataset
and test
dataset given in some public dataset. Besides, future trends for visual
localization are also shared. Syllabus: The
contents have five parts: 1.
An overview of visual localization Some
fundamental knowledge of 3D vision and an overview of image-based
localization
are introduced. A complete classification of image-based localization
mapped as
a tree structure is given. The possible best way to use deep learning
in 3D
vision is also analyzed. 2.
Image feature representation and matching with deep learning Some
important literatures of image feature detectors and descriptors with
deep
learning are introduced. Also, some important literatures of image
matching with
deep learning and dataset are introduced. 3.
Camera pose determination from known 3D structure knowledge with deep
learning Deserved
studies with deep learning are pointed out. The developments of camera
localization in known environments are introduced. PnP and RANSAC using
deep
learning are also presented. 4.
SLAM with deep learning This
part will introduce some learning-based SLAM, recent new dataset, and
challenges. It also includes feature tracking from videos, loop closure
detection, bundle adjustment etc. 5.
Problems and trends when view geometry using deep learning Some
personal views are shared for some problems and trends when view
geometry using
deep learning.
|
Last Updated by Yihong Wu, September 2020