Ke Zhu was born in Nanjing, China. He graduated from the University Duisburg-Essen with a Dipl.-Inf.(Univ.) degree in Computer Engineering (Angewandte Informatik). During his studies he specialized in Path Planning of mobile robots, Simultaneous Localization And Mapping (SLAM) and Computer Vision. His diploma thesis is "Data Association Methods in the Range of Mobile Robot".
Since February 2009 he works as a PhD-Student at Remote Sensing Technology of TUM. His field of work is about 3D reconstruction. His work is a part of the project 'Safe Earth' which was established in the International Graduate School of Science and Engineering (IGSSE) at TUM as result of the german excellence initiative in 2009.
Besies the investigation on computational performance improvement of various stereo methods using modern hardware, like Compute Unified Device Architecture(CUDA), he develops robust dense stereo matching methods on chanllenging real-world data from airborne imagery:
Dense stereo matching has been an active area of research in computer vision for over two decades. Many of the best stereo matching algorithms are framed as global energy minimizations, but these algorithms tend to have problems when applied to real-world data due to radiometric changes, large baseline and complicated scenarios. Our work introduces two methodological novelties aiming at general applicability for almost all global optimization methods:
- Based on the performance study of matching cost functions using both standard computer vision benchmark and remote sensing data, a merging strategy is introduced to design robust matching costs.
- A condence-based surface prior is developed within a probabilistic framework. Depending on the reliability of the prior (confidence) from a previous matching, the introduced surface prior is modeled as a Gaussian distribution, which can be probability fused with the current approach.
Zhu K, Neilson D, d'Angelo P (2013):
Confidence-based Surface Prior for Energy-Minimization Stereo Matching.
German Conference on Pattern Recognition (DAGM), accepted [PDF]
Tel.: ++49 / (0)89 / 289-23895
Fax: ++49 / (0)89 / 280-9573