Abstract

The information of a person’s activity can be used for a multitude of application. Thus, the activity recognition and the improvement of the activity recognition performance is a focus of research. Unfortunately, activities similar to each other are still likely to be mixed up during the activity recognition, resulting in a decrease in activity recognition performance. We propose a lightweight method to improve the recognition performance. By reevaluating recognized activities, we group similar activities to a new activity. We refer the created activity as grouped activity. We evaluate our method with an activity-set of seven activities and compare the results with two common approaches: the average recognition performance of the individual activities and the recognition performance using a generalized activity. We show an F-Measure improvement of up to 6.6 % by using the grouped activity.

Weitere Informationen finden Sie auf der Website der Conference.