A vision-based gesture recognition system for human-robot interaction

The paper describes a gesture recognition system which can effectively recognize static single-hand gestures and be applied in complex environments. The system involves a vocabulary of 20 gestures consisting of Chinese sign language for certain letters and digits. Segmentation based on color learning and normalization based on image moment invariants are used to extract candidate hand regions. Its real-time performance is due to a novel combination of the voting theory and an improved relief algorithm. The module of multiple-region recognition is also added to our system to decrease effects of noise region. The recognition system is now used in our human-robot interaction project.