Unobstrusive human activity recognition using smartphones and Hidden Markov Models

Accelerometer data is sufficient to compute human activity recognition, even with only a single accelerometer in use. Such data can be used for many pervasive computing applications, user activity being interpreted as real-time contextual information. This paper investigates activity recognition on smartphones, as they are a suitable platform for the implementation of context-aware pervasive systems. Many machine learning algorithms are suitable for this purpose, but Hidden Markov Models (HMMs) are particularly appropriate for their ability to exploit the sequential and temporal nature of data. This paper evaluates HMMs in unobstrusive activity recognition with the added restrictions resulting from the use of the smartphone platform.