3축 가속도계를 이용한 은닉 마르코프 모델 기반의 행동인식

Abstract

We consider the activity recognition using one triaxial-accelerometer for daily-life log. We assume that the sensor is included in the mobile phone and develop a device for activity recognition. Eight activities are modeled using three-state hidden Markov models. In the experiments, HMMs were trained by maximum likelihood (ML), maximum mutual information (MMI) and maximum margin framework (MM). The MMI and MM improved the recognition accuracy over the ML which is widely used in training the HMMs.