Visit ShanghaiTech University | 中文 | How to find us
HOME > News and Events > Events
Highly-Accurate Two-Frame Optical Flow Algorithm, and Motion Estimation for Autonomous Driving in the Rain
Date: 2016/12/2             Browse: 601
Seminar Topic: Highly-Accurate Two-Frame Optical Flow Algorithm, and Motion Estimation for Autonomous Driving in the Rain

Speaker: Prof. Hongdong Li
Time: Dec. 2, 10:30 a.m. - 11:30 a.m.
Venue:  Room 1A-200, SIST Building

This talk focuses on dealing with a challenging, frequently encountered, yet not properly investigated problem in two-frame optical flow estimation. That is, the input frames are compounds of two imaging layers – one desired background layer of the scene, and one distracting, possibly moving layer due to transparency or reflection. In this situation, the conventional brightness constancy constraint – the cornerstone of most existing optical flow methods – will no longer be valid. In this talk, the speaker will illustrate a robust solution to this problem in his recently published work on CVPR’16. The proposed method performs both optical flow estimation, and image layer separation. It exploits a generalized double-layer brightness consistency constraint connecting these two tasks, and utilizes the priors for both of them. Experiments on both synthetic data and real images have confirmed the efficacy of the proposed method. Remarkably, this is the first attempt towards handling generic optical flow fields of two-frame images containing transparency or reflection.

Hongdong Li is currently a Chief Investigator and a Program Leader for the Australia Centre of Excellence for Robotic Vision (ACRV).  He is a Reader with the College of Engineering and Computer Science, Australian National University (ANU).   His main research interests are 3D modelling, Camera Geometry and 3D reconstruction, Augmented Reality, and Robot navigation and autonomous driving, and mathematics optimisation.  Professor Li is a leading researcher in the field of 3D Computer Vision, and Structure from motion.  He served as an Area Chair at the IEEE CVPR, ICCV, ECCV, BMVC and 3DV etc major vision conferences in recent years.   He is program Chair for DICTA and ACRA in 2015,  and DICTA 2017, and a Program Chair for the future ACCV in 2018. He is Associate Editor for IVC (Image and Vision Computing, journal) and Associate Editor for IEEE Transaction on PAMI- the no.1 journal in comptuer science.  He was a winner of the prestigious CVPR Best Paper Award in 2012, the ICPR Best Student Paper Award in 2010, the ICIP Best Student Paper Award in 2014, and several other awards  (CiSRA Prize and DSTO Award, etc.).  He is the joint winner of VOT-TIR Challenge (first place award) in ECCV 2016.   His researches are funded by the Australia Research Council (ARC), ARC Centre of Excellence program and many major global ICT and automobile companies. Prior to his current appointment at the ACRV and ANU, he was a Research Fellow and Senior Fellow then a Fellow with the Research School of Information Science and Engineering (RSISE) of ANU during 2003-2008.  He was seconded to NICTA (National ICT Australia) as Senior Scientist in 2009-2010.   He was an Associate Investigator of the BVA (Bionic Vision Australia), working on the “Australia Bionic Eyes” project.  He has been teaching robotics and computer vision classes to undergraduate students at the ANU in the past ten years. 

Seminar 16079