@inproceedings {pub3744,
	title = {Personalized Online Learning of Whole-Body Motion Classes Using Multiple Inertial Measurement Units},
	author = {Viktor Losing AND Martina Hasenj{\"a}ger AND Barbara Hammer AND Heiko Wersing},
	year = {2019},
	month = {May},
	abstract = {Online action classification is a field with a variety of application scenarios. It is particularly important for body assisting devices which try to support the motions of its user. This paper investigates the benefits of personalized online learning in this domain. 
We let different subjects perform various motions which we categorized in eighteen classes. The data was recorded using the XSens bodysuit with 17 integrated inertial sensors, resulting in an challenging task with over 2000 actions in total. 
First, we utilize a greedy feature selection approach to show that only a few sensors are necessary to achieve a high classification performance. 
On this basis, we perform a thorough study, concluding that online personalized models require very few data to clearly outperform average user systems.
The performance gain increases further with additional personal training data.},
	publisher = {IEEE},
	booktitle = {International Conference on Robotics and Automation (ICRA)}
}