Special Seminar: Darius Burschka “Vision-Based Interaction in Dynamic Scenes”
While perception and modelling of static environments became a well-research problem, independent motions in the scene are still difficult to acquire and represent for robotic tasks. The challenges range from the required significantly higher sampling rate for a correct representation of the motion to appropriate representation of actions and behaviours in a knowledge database. For an observation of human actions, the typical sampling rate of a standard video-camera is not sufficient to catch the details of the transport action beyond the registration of the resulting position change of an object. A high-speed motion tracking system is necessary to analyse the intentions of the agent while performing an action. At the same time, the dynamic change in the scene is often used not only for task analysis but also for the implementation of reactive behaviours on systems. An interesting aspect in this context is to find a robust representation for the information exchange that is insensitive to calibration errors in the visual and the control part of the system. Experiments show that exchange in the three-dimensional Cartesian space is not optimal although it is easier to understand by the human operator.
In my talk, I will present the newest research results in my group that allow a fast labelling and estimation of physical properties of dynamic objects in manipulation scenarios and that allow also to implement low level reactive behaviours on mobile and flying robots without exact camera calibration. The developed hybrid stereo system allows motion acquisitions up to 120Hz providing a better sampling of the human behaviours. I will also present our work on motion representation for trajectory planning and collision avoidance on our robotic car platform RoboMobil.
Darius Burschka received his PhD degree in Electrical and Computer Engineering in 1998 from the Technical University Munich (TUM) in the field of vision-based navigation and map generation with binocular stereo systems. In 1999, he was a Postdoctoral Associate at Yale University, New Haven, Connecticut, where he worked on laser-based map generation and landmark selection from video images for vision-based navigation systems. From 1999 to 2003, he was an Associate Research Scientist at the Johns Hopkins University. Later 2003 to 2005, he was an Assistant Research Professor in Computer Science at JHU.
Currently, he is an Associate Professor in Computer Science at the TUM in Germany, where he heads the computer vision and perception group. He has a close collaboration with the German Aerospace Agency (DLR). He is a Co-Chair of the IEEE RAS Technical Committee for Computer and Robot Vision, Co-Chair of the Computer Vision and Perception Topic Group at euRobotics (EU Horizon2020), and a Senior Member of IEEE.