Abstract

The wide use of gestures in human computer interaction has lead to many different ways for recognizing gestures (e.g. via a camera, a smart phone or a pen device). Each way has different properties (like intrusiveness or handling) which influence its usability for different applications but also for specific situations. To optimally use gesture interaction, it is therefore required to utilize the different recognition ways for the same application. In this diploma thesis, a novel approach for a device independent hand gesture recognition is proposed. The proposed approach allows to recognize gestures that are performed with different gesture devices, with one classifier that is trained with one arbitrary gesture device. Performing a training with one arbitrary gesture device (e.g. a camera) facilitates the use of various other devices (e.g. game controllers, smart phones) for the interaction afterwards. To enable support for gesture input from vision systems, accelerometers or pen devices, the performed gesture trajectories are reconstructed from the gesture device sensor data. This allows a device independent representation of different gesture device sensor data. Based on this, a so called 'pre-shape' is generated in the Kendall shape space. This 'pre-shape' is free of scale and position information and is compared to templates in a database by a rotation invariant similarity measurement. The 'pre-shape' pair that provides the smallest similarity distance is the matching one. A comprehensive study shows the feasibility of the approach, but also the difficulties of reconstructing trajectories from accelerometers.

@inproceedings{...,
author = {...},
title = {...},
booktitle = {...},
year = {...},
isbn = {...},
pages = {...},
location = {...},
doi = {...},
publisher = {...},
address = {...},
}