MeMoMan2 - Methods for real-time accurate Model-based Measurement of HuMan Motion
Final Report Abstract
In MeMoMan we designed, realized and investigated a system that is capable of realtime detection and tracking of humans while performing everyday manipulation activies. To this end we developed methods for detecting and tracking of persons, combining tracking data with object detection to detect actions, and representations for tracking and action data. The data is logged at runtime of the system and made accessible for further investigation and analysis through our web-based knowledge representation and processing service OPEN EASE. We have successfully shown the application and strengths of the MeMoMan system during review meetings with external reviewers of the two EU projects RoboHow and SAPHARI. Within Robo-How we used the MeMoMan system to enable a robot to synchronize its actions with a human that is preparing a pizza. SAPHARI aims at safe physical human robot collaboration. Here we used MeMoMan to enable a robot to be aware of humans within its workspace. The semantically annotated data from the experiments can be accessed through the website at http:/http://www.open-ease.org/. With publishing our data on OPEN EASE we aim at establishing “Open Research” in the robotics community and enable other researchers without access to a mobile robotic platform to access data from real world experiments. Reviewers are invited to visit the website, create a free account and investigate all the experiments to get a better impression. In contrast to the original idea of using stereo and time-of-flight cameras we changed to using the kinect. The kinect combines an RGB and depth sensor and was developed for the mass market and is therefore cheaply acquirable. In addition with the position change of Michael Beetz we took the chance of making use of synergistic effects and put more emphasis on pushing towards “Open Research”.
Publications
- (2015). Perception for everyday human robot interaction. KI – Künstliche Intelligenz: 1-7
Jan-Hendrik Worch, Ferenc Balint-Benczedi and Michael Beetz
(See online at https://doi.org/10.1007/s13218-015-0400-1) - Model-free detection, encoding, retrieval, and visualization of human poses from kinect data. IEEE/ASME Transactions on Mechatronics 20 (2): 865-875
Martin Stommel, Michael Beetz and Weiliang Xu
(See online at https://doi.org/10.1109/TMECH.2014.2322376)
