Our paper proposes a vision-based hand gesture recognition system. It is implemented in a camera-projector system to achieve an augmented reality tool. In this configuration the main problem is that the hand surface reflects the projected background, thus we apply a robust hand segmentation method. Hand localizing is based on a background subtraction method, which adapts to the changes of the projected background. Hand poses are described by a method based on modified Fourier descriptors, which involves distance metric for the nearest neighbor classification. The proposed classification method is compared to other feature extraction methods. We also conducted tests on several users. Finally, the recognition efficiency is improved by the recognition probabilities of the consecutive detected gestures by maximum likelihood approach. copy; Springer-Verlag 2004.
|Number of pages||11|
|Journal||Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)|
|Publication status||Published - Dec 1 2004|
ASJC Scopus subject areas
- Theoretical Computer Science
- Computer Science(all)