Estimating energy expenditure of mobile device users using HMM

In this paper, we apply HMM to estimate amount of energy expenditure of mobile device users using a 3-axes acceleration sensor. HMM is used to recognize user's contexts from a time-series of acceleration data obtained by SunSPOT attached to the user. The energy expenditure can be calculated based on MET values corresponding to the contexts. In the preliminary experiments, we tried to distinguish between two contexts of “running” and “walking”. We observed acceleration data for 20 minutes with 100Hz for each context. As a result, HMM could recognize more than 90% cases correctly.