ISMAR sessions > Keynote speakers

Georg Klein - Tuesday Oct. 10 - 9:30-10:30

Georg Klein is a scientist at Microsoft Corporation.
He obtained an MEng and a PhD from the University of Cambridge in 2001 and 2006.
While his first attempts at computer vision involved the visual guidance of a mobile robot, his PhD work focused on applying model-based visual tracking to augmented reality.
During a post-doc at the University of Oxford, his research emphasis shifted to advancing SLAM techniques for AR, producing the well-known PTAM monocular SLAM system.
In 2009 he joined Microsoft, where he has helped develop the Microsoft Hololens, a commercially available self-tracked optical see-through HMD. 

Registration on Hololens

Microsoft Hololens is an optical see-through HMD that provides users the illusion of world-locked holograms: virtual content that appears stationary in the world even as the user moves around. In this talk I will describe some of the enabling technologies developed to achieve this, from sensors to displays to the Holographic Processing Unit.

Marie-Odile Berger - Wednesday Oct. 11 - 10:30-11:30 

Marie-Odile Berger is Research Director at Inria Nancy Grand Est (France). Her main research interest is the investigation of computer vision methods to support augmented reality (AR) tasks. Since the late 1990s, she has been engaged in many aspects of 3D reconstruction and pose computation, which are two main issues in AR. This research led to various theoretical and practical results in the areas of calibration, matching and 3D tracking, interactive reconstruction and visual perception. Her research also focuses on developing AR for deformable worlds with effective applications to interventional radiology. She is currently the head of the MAGRIT computer vision group at Inria.

Pose estimation for effective AR tasks.

The aim of my research in Augmented Reality is  to develop reliable and effective methods which allow significant progress in terms of ease of implementation, robustness to external conditions and capacity of handling complex environments. 
In this talk, I will present various works on pose estimation and scene modeling that have in common the goal to meet the expected robustness in different application contexts.
Though tracking strategies are now relatively mature, automatic pose initialization is still a barrier in many AR systems. Difficulties especially originate in repeated patterns, common in man-made environments, and in large variations of the viewpoint between the data stored in the model and the current view.
I will describe solutions for the estimation of the initial pose, both in the rigid and in the deformable context. These contributions are based on different methodologies depending on the application context. They range from A Contrario statistical models, to the use of view synthesis and to contextual learning-based methods specific to urban environments.
Finally, I will briefly address the role of interaction in the development of AR applications through in-situ modeling.