Skip to Main Content
This paper presents a gesture input device, magic wand, with which a user can input gestures in 3-D space, inertial sensors embedded in it generate acceleration and angular velocity signals according to a user's hand movement. A trajectory estimation algorithm is employed to convert them into a trajectory on 2-D plane. The recognition algorithm based on Bayesian networks finds the gesture model with the maximum likelihood from it. The recognition performance of the proposed system is quite promising; the writer-independent recognition rate was 99.2% on average for the database of 15 writers and 13 gesture classes.