Activities are complex action sequences often getting their meaning from interaction with objects or the overall context. Our current focus of research in the field of activity recognition is to enable a robot ways to acquire brief knowledge about what is going on in a room. This way, the robot can easily adapt to the room situation and react accordingly. For instance, after identifying that somebody is unloading the dishwasher, the robot could offer his help or even take over the task without any need of explicit commands.
To recognize activities automatically, we use the fact that human activities can be discriminated to a certain degree by their single motions, even under different environmental conditions. Also, the existence of some involved objects contains information about the activity itself. For instance, it is very likely that a person performing an activity with a knife is cutting something. However, instead of developing a new classifier for each object and each motion type, we use both kinds of information implicitly by employing low-level image-based features like motion and image gradients to classify an activity.
- L.Rybok, S. Friedberger, U. D. Hanebeck, and R. Stiefelhagen. The KIT Robo-Kitchen Data set for the Evaluation of View-based Activity Recognition Systems, IEEE-RAS International Conference on Humanoid Robots, October 2011, Bled, Slovenia (pdf, bib)
- D. Gehrig, P. Krauthausen, L. Rybok, H. Kuehne, U. D. Hanebeck, T. Schultz, and R. Stiefelhagen. Combined Intention, Activity, and Motion Recognition for a Humanoid Household Robot, IEEE/RSJ Int. Conference on Intelligent Robots and Systems - IROS11, September 2011, San Francisco, USA. (pdf, bib)
- C. Wojek, K. Nickel, and R. Stiefelhagen. Activity Recognition and Room Level Tracking in an Office Environment , IEEE Int. Conference on Multisensor Fusion and Integration for Intelligent Systems - MFI06, September 2006, Heidelberg, Germany. (pdf)
- See publications page for more!
Contact: Dipl. Inform. Lukas Rybok