Age-dependent degradation estimation focusing on dynamic features of head motion in daily conversation
Kana Natsumi, Yumi Wakita, Chitose Matsuda · 2020
We are developing a system that can estimate the degrees of decline for the elderly from daily conversation and inform them when the estimated result indicates a decline. In the paper, we discuss the effectiveness of the dynamic features of gestures during conversations to identify age-dependent degradation. By focusing on the head motion when a speaker responds to a partner while talking, we compare the optical flow values of images between independent elderly speakers and facility elderly speakers. The results of identification experiments using Support Vector Machine (SVM) indicate the average of correct rate is 83.3%. This result suggests that the dynamic feature of head motion has a tendency to be effective for age-dependent degradation estimation.