Extention of hidden Markov models to deal with multiple candidates of observations and its application to mobile-robot-oriented gesture recognition

摘要

Aiming at improving the performance of gesture recognition in the condition of moving camera, we have proposed a modified HMM which can deal with multiple feature candidates. The conventional HMM is formulated to deal with only one feature candidate per frame However, for mobile robot, the background and the lighting condition change every moment. Therefore, the feature detection problem becomes very difficult. If we select only one candidate in the feature detection phase, many errors must be involved, and it cause serious degradation of total performance. In this paper, we propose a new gesture recognition framework in which multiple candidates of feature vectors are generated with confidence measures and HMM is extended to deal with these multiple feature vectors. Experimental results show the effectiveness of the proposed method.