Understanding presumption system from an image sequence using HMM

In this paper, the understanding presumption system from the gesture recognition using Hidden Markov Model (HMM) is proposed. The features of this system are 1) not limiting the gesture recognized, and 2) automatically extracting the feature points by using HMM without a user's hand. In particular, the time-line pictures of subject's face are r st input into the system. Then, the motion of their face region, pupils, and eyebrows are extracted as a feature vector from each still picture. Next, to the feature vector sequence is changed into the symbol sequence, gesture has been recognized by estimating likelihood of HMM which learned gesture beforehand, using Viterbi algorithm. At the end, their degree-of-comprehension is presumed from the appearance probability of the recognized gesture according to their understanding. At the time, we take a video of their solving a problem during the evaluation experiment. And their degree-of-comprehension are presumed for their picture as an input of a system. Consequently, it is shown that understanding presumption by the proposed method is possible.