Head Gesture Recognition Using HMM

초록

In this paper, we present a technique for head gesture recognition. The method includes face motion tracking, face location, and head gesture recognition using HMMs(Hidden Markov Models). In the first step, motion tracking is performed based on the techniques of low resolution image difference and Kalman filter. The purpose of this step is to detect face movement. In the second step, we construct mosaic image called octet face from the extracted region of the first step. The octet face is checked by neural network whether it is an exact face region or not. The central coordinates of the face regions from sequent images are transformed into the directional vectors. The vectors are inputted to HMMs to decide whether it is positive, negative or neutral head motion.