Research Interests: Computer vision
Office Phone: 86-10-6275 5793
Zha, Hongbin is a professor in the Department of Machine Intelligence, School of EECS. He has served as a Vice Dean of the School of EECS since 2002, and the Director of the Key Lab of Machine Perception (MOE), PKU, since 2007. He obtained his B.Sc. from Hefei University of Technology in 1983, and Ph.D. from Kyushu University, Japan, in 1990, respectively. His research interests include 3D reconstruction and modeling, geometric data processing, motion tracking and analysis, all in computer vision.
Dr. Zha has published more than 300 papers, including more than 70 papers in top-tier journals, such as IEEE T-PAMI, IJCV, IEEE T-VCG, IEEE T-RA, IEEE T-SMC,ACM T-IST, JMLR, PR, and proceedings of conferences, such as ICCV, ECCV, CVPR, CHI, ICML, AAAI, and ICRA. He received the State Technological Invention Award (Second Class), China, in 2011, IEEE SMC Society Franklin V. Taylor Award in 1999, the Best paper award from EuroMed 2010, and the Best Paper Finalist from IEEE ICMA 2011. He served or has served as Vice Chair of Beijing Section, ACM, since 2014, Executive Committee Member and Technical Committee Chair of Beijing Section, IEEE, in 2006-2007, and Officer and Secretary of Asia Federation of Computer Vision (AFCV) since 2012. He also served as Associate Editor of the Computational Visual Media, General Chairs of ACCV’10, 3DIMPVT’11, CVM’12, ACCV’16, Program Chairs of VSMM’06, IEEE ROBIO’06, ACCV’07, ACCV’09, VSMM’10, IEEE ICMA’11, IEEE MFI’14, IEEE ICMA’17, Track Chair of ICPR’18, Area Chairs of ICCV’11, ACCV’15, and PC Editor of IEEE/RSJ IROS’13.
The main contributions of Dr. Zha to the computer vision research in recent years include:
1) 3D scene reconstruction and data fusion:
He proposed a number of methods for enhancing data quality for reconstructing 3D scenes from real sensing data. By using high-quality 3D data stored in an available data-base, the methods can automatically find corresponding surface parts in the data-base to help to remove noises in the sensing data, leading to improved surface descriptions of the scenes. At the same time, the scene quality can be enhanced further by an effective use of data fusion techniques by combining data from different sensor sources. The dynamic data fusion is achieved with an elegant framework based on statistical analysis approach.
2) 3D object recognition and structure analysis
A big problem in 3D object recognition is how to deal with the deformation and structural changes in the object shapes. He proposed a number of methods for the non-rigid object alignment and recognition by using a flexible manifold-based dimensional reduction approach. By transforming the object features into a low-dimensional manifold, the methods can find and recover the surface deformation and structural changes with great robustness, resulting in systematic procedures for the object recognition tasks required in a wide range of applications.